This is one of the reasons I believe Google will ultimately win the AI race once the VC funding tap runs dry. For all its breakthroughs, OpenAI lacks any fiscal discipline. Sam Altman seems convinced that as long as they’re delivering value and have a clear mission, the money will somehow take care of itself. That’s fine for now, but the music will stop playing one day.
When it comes to software engineering at scale, nothing beats Google. They already operate the world’s largest and (profitable) information index, they design their own chips, and employ the best engineers who know how to deliver massive systems cost-effectively. They can sustain large investment far longer than any other company simply because of how profitable all their existing businesses are. I just wish they would get their act together when it comes to product management.
I hold the exact same sentiment. The safest investment you can make in this AI race is Alphabet. The smaller players being wrapped up in this (CRWV, NOK, etc...) are going to get obliterated when the returns are not as good as previously thought. The bigger players will take a hit, but not as much as the small ones.
NVIDIA also seems like a risky bet given all the money they're passing around in circles to themselves via multi-billion dollar "investments"
> The safest investment you can make in this AI race is Alphabet.
From the technical/talent/funding side, I agree with you. With Alphabet, the real risk is that they lose interest/focus and kill (or pivot) products early. I don't really see that happening in the short term (considering the hype/investments/cash around AI) but maybe in the medium to long term if things start to get stale.
The chances of Alphabet losing interest in AI are vanishingly small. Google (pre-Alphabet remaming) was the AI company before AI companies were called that.
> Google (pre-Alphabet renaming) was the AI company before AI companies were called that.
I disagree. They're an advertising/attention company.
The jury's still out on whether their AI offerings will compliment their existing business or will cannibalize it. And in the case that it cannibalizes the advertising business, will they undermine/kneecap it to protect that cash cow?
It's been a while since they've been the "don't be evil" and "organizing the world's information"-mission Google.
They didn’t lose interest in messaging when they cycled through half a dozen different apps but the problem was they were building products they wanted to exist rather than what users wanted. I’d worry about AI getting similar churn if PMs make their name by launching a new product rather than improving an existing one.
That makes sense framing it as a skepticism about Google Gemini (a single product) rather than their interest in a category (AI).
I'm less worried about them genuinely losing interest then Google outlasting their AI competition and then kneecapping their AI product to protect their cash cow.
The root comment was investor-centric, i.e. if you want to divert some investment towards AI, Alphabet is a smart bet.
Investors got great returns as Google churned messaging platforms, and this will likely be the case ad Google churns through AI products, or even entire architectures.
We did get good returns but that was because other units did very well. It’s not a given but still too likely that they’d botch AI products but still be profitable due to ad sales continuing to grow. I do think AI has better than average chances, though, because it seems to have scared their senior management enough to do their jobs: from outside, everything appears to radiate “this can’t fail” in a way they haven’t had since the 2000s.
If you look back at the dot com era, hardware companies did survive but weren't the financial "winners". Cisco is an example and a good comparison is Amazon - their bread and butter now is ecommerce, they dominate it and that was borne out of that bubble.
Alphabet also has massive market share with YouTube, solid cloud offerings, Waymo which already has driverless/unsupervised robot taxis and their old standby search which is morphing more into more AI answers when you throw questions in your browser.
Alphabet's P/E ratio is lower than the S&P 500's. Some may argue the S&P is mostly being held up by Tech and AI related companies so stock in Alphabet could be seen as undervalued and NVDIA as overvalued.
Hardware almost always ends up getting commoditized so investing in hardware companies for the long haul does carry maybe more risk than mainly software companies.
I remember when SCO had an all hands meeting that they evaluated that linux wasn't a threat because SCO was the safe choice, the big player in UNIX on commodity hardware.
It’s everyone’s wet dream that Apple buys Anthropic, but I’m guessing their valuation now is already too high even for Apple. And why would Amazon let that happen?
It’s a solid match but not happening unless the AI market craters and Apple swoops in to buy them for vastly less.
All reports are that Apple isn’t really focusing on building their own server side models and is mostly just going to be a platform for other companies to fight over just like they said they don’t pay OpenAI anything for the current integration and Google pays them $20B+ a year to be the default search engine.
It makes no sense for Apple to waste money on having their own LLM when they are becoming commoditized - just chose the one that will give them the best deal.
I also think apple has put a lot of energy into providing on device hardware for this kind of stuff, like they have gone through a ton of iterations at this point on their own hardware that is also ai focused. so now they offer a lot of value in the mid to near future because if you can run ai on device why wouldn't you since you don't need to pay cloud costs anymore since users pay that for you by buying iphones.
So they’ll just die when Google has a seamless AI-first mobile experience where consumers just ask their Pixel to do X and it happens. Disruption comes for everyone, hardware isnt a moat
The issue is that Android won’t have an AI first experience where you depend on local inference because most Android phones being sold are cheap crappy devices and Android users don’t monetize.
But there is much more to a phone than “just doing stuff agentically”. Either way, you give way too much credit for Google to ever make a good end user product and be able to sell it. Google’s sells of pixels for a year are about the same number Apple sells in three weeks.
Apple’s hardware/ecosystem has been a most for 50 years.
Also Google is non existent in the country that has 1/5 of the worlds population.
Google broke my ability to set timers via voice (a core thing my phone used to do forever) for like a year. I no longer use that functionality on my phone (I can't trust Google not to break it, if it can't routinely work, it can't be part of my routine, and I personally don't enjoy cooking dinner twice after a long day).
Google replaced 'Google play the news' from playing news clips from sources I chose and trust, to an AI generated news feed that I have zero faith isn't hallucinated and from sources I have no idea about. Another multiple times daily function I used my phone for, broken by 'Google AI' because 'I should be happy with whatever they give me, it's a free product'.
Google's Youtube/Music algorithm punishes me if I flag AI slop in my feeds by removing those genres/topics, trying to force AI slop into my feed. This one I actually (well until it runs out the end of the month because I canceled) pay for.
Google has trained their users not to use Pixel's voice/AI capabilities, and to resent them (in the typical Google way, be damned with the old it's been replaced with our new half baked product that will itself be replaced shortly after the bugs are gone and it starts to work kind of OK (still really bad, but Google justifies bad is OK because it's 'FREE').
No one is integrating anything new from Google into their routine. No one is integrating Google's broken AI stuff into their daily routine, because it's friggen BROKEN. Google just one day removed me from setting voice timers with my phone, a core use case, for a year, because they could. My routine is not going back to using Google for things I count on just working because Google will arbitrarily break it and I will get the impact (maybe a burnt dinner, maybe faulty news reporting that causes me a heart attack, maybe who knows, maybe a missed important appointment tomorrow?).
Google doesn't care that I'm hit because 'it's good enough for a free product' to them. Everyone I know with a Pixel hates the ecosystem right now. Removing sideloading might fix that and improve good will I guess though? God Google doesn't know and hates their users/advocates/supporters. Google has moved themselves from 'a good solution' to 'good enough for a cheap device and free tools', and made damn sure I know it (via burnt dinners, having to move to my podcast app for (less current) news, by degrading my music/youtube feeds).
Google is a zombie company, they are dead, they just don't know it yet. Coming soon... Alphabet, a Bending Spoons company.
apple is quite smart with this, there was a time when people thought they were dumb for not buying up tesla and ultimately shutting down their self driving division, but now it seems like a good bet.
NVIDIA also seems like a risky bet because Google has their own chips R&D. It's not just another data center buying NVIDIA's GPUs. BTW Apple is following suit regarding chips, don't know if at some point Apple will offer any cloud service or continue to work on their ecosystem only.
This makes sense to me. Where I work our ai team set up a couple h100 cards and are hosting a newer model that uses up around 80GB vram. You can see the gpu utilization on graphana go to like 80% for seconds as it processes a single request. That was very surprising to me. This is $30k worth of hardware that can support only a couple users and maybe only 1 if you have an agent going. Now, maybe we're doing something wrong, but it's hard to imagine anyone is going to make money on hosting billions of dollars of these cards when you're making $20 a month per card. I guess it depends on how active your users are. Hard to imagine anthropic is right side up here.
I've been a GCP consultant for close to a decade now. Google messed up big time lagging behind in AI - they just kept increasing prices in other products for about 5 years straight (CDN, for example) and did nothing to productize AI. I'm of the belief that ChatGPT should've been their pilot project. What they have now (AI mode) in Google should've been there well before ChatGPT.
But, all that aside, you know what they're really good at? Google Cloud. I'm a user of all the major Cloud providers and nothing beats GCP's interface.
Azure? Complicated, buggy and unreliable. There's some exploit every quarter.
AWS? Overly complex security policies even to deploy a basic app, very enterprise focused and startup-unfriendly.
GCP - you can be up and running on a serverless/VM instance in your lunch break. Simple, reliable, scales effortlessly. We serve 10M+ visitors on there and we've had zero issues in the last half a decade with them.
They suck at a lot of other things and they have a lot of other problems. But boy, are they good at Cloud. No wonder even Apple is their customer. It's one of the few products from Google where you can say "it just works".
I think App Engine was really ahead of its time in showing how simple cloud deployments can be. It had a similar ease of use as setting up a YouTube account. For that reason, a lot of people thought of it as a toy, which was kind of unfair because companies like Niantic were able to build global products on it. So a lot of Google Cloud afterward ended up being designed to be more "normal" like how Amazon is. Now people are seeing what normal gets them, so maybe it's going to be time for the Google way of doing things to finally shine. (Disclaimer: I'm a Google employee)
App Engine caused a huge innovator's dilemma for Google when it came to cloud.
It only addressed a set of use cases around web application development- Folks would ask "why shoudl we build a low-margin cloud?" and I'd tell people: "if I can't import numpy, App Engine is useless to me". Eventually, the useful bits of AE were extracted to other services with APIs (datastore is one example), but it wasn't until google built a full GCP that they started to see real cloud growth (revenue).
I totally second this. AppEngine was way ahead of its time. I used it for all my projects back then. The only reason I stopped using it for clients because even internal Google employees had no idea when the plug might be pulled out on AppEngine. AppEngine supported Elixir far earlier than anyone else (Elixir is my main go-to stack)
Ah, that's a shame they went that way. What I tell people when they're first getting into GCP is that it's gonna be a passion to set up what you want. But, the offerings are great, and once it's working, it'll just work
I am also a cloud consultant focused on AWS and a former employee of AWS ProServe (I have no love lost for the company).
But if you are a consultant, why pray tell are you spending that much time in the console except for occasional monitoring? Everything is either CLI commands or infrastructure as code.
And the issue with Google in particular is their customer support and business enterprise go to market sucks. The sales guys at ProServe use to run circles around GCP and never really took them seriously or had talking points. For big contracts we mostly had to compete with Azure because big boring enterprise was already using Microsoft.
When we did have to compete against GCP - and their sales team didn’t blow up the deal themselves - we just had to say “do you really want to trust Google with your workloads? Look at their history of abandoning products and raising prices”
Apple is a customer of all of the big cloud providers and was on stage at last year’s reinvent.
AWS throws money and people at startups. I’ve been on three sides - working at a 60 person startup who hosted on AWS, working at ProServe and now a third party consulting company where many deals come about because of AWS funding.
Google has the resources to win AI, but I still don't see a way to resolve the tension between good AI and ad revenue. A useful AI tells me what I need to know, but Google's cash cow is selling users to advertisers. Those are fundamentally oppositional and I don't see a clear way to resolve it.
Google can kick ass in AI all day, but soon Gemini is going to be saying shit like, "it sounds like your air conditioner is broken, maybe you should call Clearwater HVAC - they're the best!"
Once Google has to start making money from AI, talking to Gemini is going to feel like talking to that friend that invited you to dinner only to try to pitch you on an MLM.
Wouldn't one possible solution be just adding a sidebar with ads for products/services relevant to the current topic? They don't need to change the model output to have ads. (Not advocating for ads, just pointing out there's lots of ways to deliver ads)
I ask, because earlier this year I had the chance to use a Waymo for the first time. I moved a good chunk of my portfolio to GOOG after that.
To me, Waymo felt like either Magic or stepping 15+ years into the future; Demolition Man stuff (im a child of the 80s, Xenial).
Stuff like Waymo, AlphaFold and similar are just so far away from anybody else, that Alphabet could split in several companies and all of them would be hugely dominant in the future. And none of them base their revenue in advertising.
> Google can kick ass in AI all day, but soon Gemini is going to be saying shit like, "it sounds like your air conditioner is broken, maybe you should call Clearwater HVAC - they're the best!"
That is the nightmare scenario. The puzzle for me is why you apparently think it just a trap for Google. All of the pure AI companies are burning cash. That is going to stop. Either they go broke, or they do something with ads.
There is a third option I guess, and that is they create something that is useful enough to be worth paying for. It won't be just an LLM, but as far as I can tell they are all just fine tuning LLM's. We already know what the limits of an LLM are, as we are already there.
All except Google that is, who are building AI's for all sorts of things, such as protein folding, weather prediction and driving cars. LLM's were their invention, created by them for doing translation and OCR. And unlike everyone else, they aren't dependent on NVidia as they use their own internally developed hardware. Surely there is no argument about Google being the world's power house of AI?? Most of these efforts will be experiments Google is infamous for abandoning. But it's seems likely one or two will become the youtubes of AI.
I can't see anything remotely comparable from the other AI companies. It's difficult to see how this doesn't end with Google completely dominating the space, and all these LLM companies going broke.
Google is literally, today, making money from AI. Not only that but the presence of competing AI services has resulted only in record revenue and straight quarters of ad revenue growth since ChatGPT was launched.
I presume they will do what they have done successfully for the last decade or two and separate the recommendations from the ads. So you'll have gemini saying unbiasedly what it thinks and then a separate "sponsored" bit say "Buy Clearwater!"
Wtf does “put ads in” mean? Google ads is one of the most sophisticated and profitable machines on the planet. People seem to think it’s as simple as <your ad here>.
There is no way for anyone to resolve it, you'll either be paying premium, running last gen stuff locally, or getting ads, regardless of whether you use Google, OpenAI or anyone else.
"They already operate the world's largest and (profitable) information index, they design their own chips, and employ the best engineers who know how to deliver massive systems cost-effectively."
Google was incorporated in California in September 1998 and re-incorporated in the State of Delaware in August 2003. In 2015, we implemented a holding company reorganization, and as a result, Alphabet Inc. ("Alphabet") became the successor issuer to Google.
We generate revenues by delivering relevant, cost-effective online advertising; cloud-based solutions that provide enterprise customers of all sizes with infrastructure, platform services, and applications; sales of other products and services, such as fees received for subscription-based products, apps and in-app purchases, and devices."
Would "(profitable) information index" be included in (a) "online advertising" (b) "cloud-based solutions" or (c) "sales of other products and services"
HN comment claims Alphabet operates (profitable) information index
But Alphabet SEC filing discloses its revenue-generating operations are (a) online advertising delivery, (b) cloud-based solutions and (c) sales of other products and services
US law requires the statements in Alphabet's disclosure to be accurate, not misleading
Nothing requires statements in HN comments to be accurate, not misleading
What is "(profitable) information index"
Is it part of (a) online advertising delivery,
b) cloud-based solutions or (c) sales of other products and services
I think it'll be Google and effectively Microsoft -- OpenAI is already so partnered with Microsoft, and if OpenAI messes up financially, Microsoft will end up bailing it out in exchange for majority ownership. So yes the music may stop playing but that doesn't mean OpenAI disappears, but maybe Sam does.
Unless there are antitrust concerns, but that's hard to see because Google is competition and Anthropic will probably remain the third more niche player.
Google also own the ad space in a way that's unsurpassed, and that's the most direct route from where they are with their AI product and profitability.
Tangentially, it's sad to see AI become yet another attention factory. Sora 2 is OAI trying to get the jump on this, but I think it's quite likely that they'll get clobbered by Google and Meta when those companies start juicing their users attention in the same way - they have the ad buyers and the infrastructure for that business already, it's literally how they've made money this whole time.
> This is one of the reasons I believe Google will ultimately win the AI race once the VC funding tap runs dry.
They've already successfully monetised their chatbot (Gemini)[1] while all the others are still wondering how to get there.
-------------------
[1] I've got about 15 - 20 additions to the settings to prefix all chats with "don't show me video links", but apparently the "show youtube link" is at the system layer and cannot be overridden at all.
Wouldn't this same logic apply to IBM back in its heyday?
The biggest risk for Google is that it is already a behemoth and it has lost a lot of what made it so special. I also think they have a huge conflict of interest when it comes to AI, because it is a threat to their core business.
To be fair, Alphabet id REQUIRED to be strong in AI. Google as a search engine is losing clicks to its own AI summaries. If they didn't have that, they would risk the ad business in the future
Money alone cannot win the race; it will require a strong vision, courage, and luck. OpenAI and XAI and Anthropic (though underfunded) seem to lead in those factors so far.
idk google is so flaky, one must not forget that they had the llm concept first and couldnt really materialize it into anything commercial until chat gpt came around. Google often could dominate a market, yet they kill the product/tech instead of keeping a small team to maintain it.
I think the winner will be between xai or meta, im leaning more with xai since elon seems to always have 100% conviction even with seemingly bad ideas and he is way more technical than sam & zuck
>I believe Google will ultimately win the AI race once the VC funding tap runs dry.
Your mention of "VC funding" may be a placeholder for conversational purposes but just to clarify, OpenAI is way beyond the small pocketbooks of typical Venture Capitalists. They have investments from huge sovereign wealth funds (e.g. billions from Middle East oil funds) and trillion-dollar corporations like NVIDIA, Microsoft, AMD, etc.
In other words, OpenAI has embedded itself into the money and tech ecosystem much more deeply than the dotcom failed startup Webvan burning through the limited cash of Sequoia VC.
There are lots of heavily funded vested interests that don't want to see Google be the ultimate winner (or only winner).
Google's new AI voice assistant finally seems to reliably set timers. The problem is they released it what, a year ago? And during that time it didn't and I now no longer trust the AI on my phone to do tasks. Google themselves trained me not to trust their AI tools.
Youtube (Google) currently degrades my feed because when I block AI videos, they assume I dislike the related topic. They have made it either I love AI videos/music, or they punish me.
You are right, they have stakeholders causing inertia to protect their fiefdoms across a large range of products. You are right, they have (profitable) current products which will prove to be barriers to nimble, future, nearterm non-profitable products. I'm not sure that's a strength seeing how they have responded so far.
They are too big and too slow to adapt. They have too many people with interests in protecting their (legacy) niche, too much inertia to overcome. The classic signs of a zombie company. Going purely off their ability to adapt in the last year, they are dead and don't even know it yet.
"When it comes to software engineering at scale, nothing beats Google"
I agree with many of your statements, but this one simply isn't true. They've really struggled to integrate AI into products in a useful way. Do you remember glue on pizza? Founding Fathers reimagined for DEI? The Brain/DeepMind merger was largely an acknowledgement of their many misses from a product perspective.
It's not really impressive to me in absolute terms that GCP grew by 34% because it's still in last place behind Azure. I worked with all three major CSPs at a past company and Google, by far, had the worst support, the worst network, the most issues, and at the same time they were the least willing to build custom features for us while Azure and AWS would constantly fight for our business.
IMHO I think Microsoft is a lot closer in terms of creating a real business around AI because they understand Enterprise customers. In fact they were first to offer private OpenAI deployments with AuthN/AuthZ that you could drop in your own VPC and the ability to handle sensitive data while everyone else was handing out API keys off a public endpoint like a startup. Don't get me wrong Google has had some okay technology (i.e. Gemini 2.5 is middle of the pack) but they seem like a very consumer-oriented company and don't seem to know how to market them.
I build tools for our security teams, I guarantee you the only reason we use Microsoft Azure is corporate deals. All the critical infrastructure is on AWS (including our Azure monitoring tools), and if we could move from Azure to GCP all our security experts would be happy (except maybe the red team members whose job would be harder)
I have used the major hyperscalers for years. I agree with you that Google is the worst business partner, but saying they have the worst network and more issues does not align with my experience at all. Azure is miles behind on technical prowess.
We use GCP at work, it works well for what we use it (VMs, container storage, cloud file storage), I wish they would stop deprecating stuff though, it just causes developer busy work with no additional value.
> it just causes developer busy work with no additional value.
Anyone remotely familiar with Google as a third party developer will notice the pattern: this will ramp up until it is almost your entire job simply dealing with their changes, most of which will be not-quite-actual-fixes to their previous round of changes.
This is not unique to Google, but it is a strategy employed to slow down development at other companies, and so help preserve the moat.
Old-timers who date back to when Joel Spolsky's early musings on the business of software development were fixtures on the HN front page will remember him using the phrase "fire and motion" for Microsoft's old strategy of constantly making changes so that everyone trying to keep up was—like the Red Queen—running as fast as they could but not getting anywhere.
> Watch out when your competition fires at you. Do they just want to force you to keep busy reacting to their volleys, so you can’t move forward?
> Think of the history of data access strategies to come out of Microsoft. ODBC, RDO, DAO, ADO, OLEDB, now ADO.NET – All New! Are these technological imperatives? The result of an incompetent design group that needs to reinvent data access every goddamn year? (That’s probably it, actually.) But the end result is just cover fire. The competition has no choice but to spend all their time porting and keeping up, time that they can’t spend writing new features. Look closely at the software landscape. The companies that do well are the ones who rely least on big companies and don’t have to spend all their cycles catching up and reimplementing and fixing bugs that crop up only on Windows XP.
I used to think that made sense (as sibling mentions, Spolsky's "fire and motion" thesis)... until I worked at large-ish tech company whose internal platforms also kept doing this. Heck, the platform I owned also underwent a couple cycles of this. And so a large part of our work was just doing the Red Queen's race of deprecations and migrations.
So it was definitely not "fire and motion," as there was no competition. I think platforms genuinely need to evolve as new use-cases onboard and technology progresses, and so the assumptions underpinning the platform's design and architecture no longer hold.
However, I do think a small part of the problem was also PDD: "Promotion Driven Development."
> it is a strategy employed to slow down development at other companies, and so help preserve the moat.
Worked at Google for 7 years, and your post reminds me it is time to share a secret: it is Koyaanisqatsi* and people's base instincts unbridled, no more. There is no quicker route to irrelevancy than being the person who cares about something from last years OKRs.
* to be clear, s/Koyaanisqatsi/too big to exist healthily and yet it does exist -- edited this in, after I realized it's unclear even if you watched the movie and know the translation of the word
If they actually incentivized a group to support stability and continuity among enterprise customers, they would probably be able to diversify their revenue away from ads. Microsoft understands this…
The real sick thing is it doesn't matter, right? Like we're commenting on an article about how they won the day yesterday and Cloud revenue continues to skyrocket.
To be clear, I agree with you, and am puzzled by the lack of consequences from the real world for the stuff I saw. But that was always the mystery of Google to me, in a nutshell: How can we keep getting away with this?
A large part of that is the Google-are-super-geniuses PR effort. Anyone pointing out that Google's products don't reflect this to their boss faces having their own credibility reduced instead.
If it's so obvious, and Google supposedly know this internally, and can obviously tell by others that some are avoiding Google because they're fast at sunsetting services, why are they not doing anything about it?
Imaginary conversation between an honest VP and earnest year 0 me, here's what the VP says:
"We definitely care about deprecations: now tell me how to accomplish that with Sundar's Focus™* Agenda over the last 2 years"
* no net headcount increases, random firings, and any new headcount should be overseas. i.e. we have the same # of people we did in 2021 with 50% more to do.
Unfortunately the software deployed on top of them will.
So you either:
1) postpone all your updates for years until a bad CVE hits and you need to update or some application goes end of life and you’re screwed because updating becomes a massive exercise
2) do regular updates and patches to the entire stack, including Linux, in which case, you’re in the same position you were before with running on the stack rot treadmill
So you might’ve moved the rot to a different place, but I don’t know if you’ve reduced any of it. I’ve owned stuff deployed off of vanilla VMs and I actually found it harder to maintain because everything was a one-off.
My rationale for staying up to date aggressively is that it minimizes integration work. Basically integration work multiplies, it doesn't just accumulate. So the further you fall behind the more that can break when you finally do upgrade. And you create needlessly more work related to testing and fixing all that. Upgrading a system that was fully up to date until a few days/weeks ago is generally easy. There's only so much that changes. Doing the same to something that was last touched five years ago can be a bit painful. Apis that no longer exist. Components that are no longer supported. Code that no longer compiles. Etc.
I see a lot of teams being overly conservative with keeping their stuff up to date running with years out of date stuff with lots of known & fixed bugs of all varieties, performance issues that have long since been addressed, etc. All in the name of stability.
I treat anything that isn't up to date as technical debt. If an update breaks stuff, I need to know so I can either deal with it or document a work around or a (usually temporary) version rollback. While that happens, it doesn't happen a lot. And I prefer knowing about these things because I've tried it over being ignorant of the breakage because I haven't updated anything in years. It just adds to the hidden pile of technical debt you don't even know you have. Ignorance is not an excuse for not dealing with your technical debt. Or worse compounding it by building on top of it and creating more technical debt in the process.
Dealing with small changes over time is a lot less work than with dealing with a large delta all at once. It's something I've been doing for years. If I work on any of my projects, the first thing I do is update dependencies. Make sure stuff still works (tests). Make sure deprecated APIs are dealt with.
If you’re willing to put in the maintenance work, you’ll probably be in good shape whether you’re on plain VMs or a snazzy cloud provider managed service.
If business understands that you need time to work on these things :’)
RedShift1 complains that GCP is "deprecating stuff". I wouldn't put doing regular updates in the same problem category as having to deal with part of your stack disappearing.
To me "I wish they would stop deprecating stuff" sounds like any part of the stack has something like a 1% or even 10% chance in any given year to be shut off.
I would expect that by carefully choosing your stack from open source software in the Debian repos, you can bring the probability of any given part being gone with no successor to less than 0.1% per year. As an example - could you imagine Python becoming unavailable in 2026? Or SQLite? Docker?
I have a bunch of scripts that do the work for me so I can just run that in the background and do something else, and for one off tasks grumble and mumble a bit.
Oh my god I thought I was the only one. Everyone says that you should be using the command line anyway, but I'd rather use a GUI but it's disgustingly slow.
It surprises me that it's built in Python (as is AWS'), which doesn't seem like a very appropriate language for exactly this reason. Go would seem much more apposite.
It's not negligible versus a single request. Even `gcloud --help` takes over a second on this machine - actually getting it to do a simple list request takes almost no time longer over any reasonable connection.
Plus Python is notoriously not easy to deploy for - a Go (or Rust or whatever) binary would have almost no dependencies to worry about.
The execution speed of the language the CLI is written in doesn't matter. Any difference is dwarfed by the slowness of network calls, specifically their API response times.
It's not just busy work, it's like being chained to a sandpaper treadmill especially in some specific "hot" domains like AI. It feels like you can build something with it and 3 months later you've got to update your code because half the dependencies are deprecated.
In AI, whatever you wrote is going to be deprecated in theee months, and in six months SOTA LLMs will one-shot it directly. That's what it means for a domain to be "hot". AI is currently the largest and most intense global R&D project in the history of humanity. So for this one field, your complaint makes no sense.
> Fuck yooooouuuuuuuu. Fuck you, fuck you, Fuck You. Drop whatever you are doing because it’s not important. What is important is OUR time. It’s costing us time and money to support our shit, and we’re tired of it, so we’re not going to support it anymore. So drop your fucking plans and go start digging through our shitty documentation, begging for scraps on forums, and oh by the way, our new shit is COMPLETELY different from the old shit, because well, we fucked that design up pretty bad, heh, but hey, that’s YOUR problem, not our problem.
> We remain committed as always to ensuring everything you write will be unusable within 1 year.
Impressive results. Microsoft also reported strong cloud growth yesterday.
Amazon reports today. AWS has been slipping and at current growth rates would slip to the number 2 cloud provider in a few years time… let’s see if that trend holds true today.
IMO, if Microsoft and Google are gaining this much, AWS is losing customers and mindshare. I'm not saying they probably won't report growth, but I doubt it's going to be the same as MSFT and GOOG.
Alternatively, with so much money being poured into "AI", the size of the entire cloud pie is growing. Everyone and their uncle are deploying AI, but no one does on-prem anymore - all that money has to go somewhere.
Amazon and Apple are two companies taking it conservatively in this bubble. Possibly because they actually have gone through the last bubble with some pain. Microsoft to some degree but they have the luxury of OpenAI.
We are at a point in this bubble where planned capex is approaching theorectical limits of actual energy capacity. Let’s see when they start pumping coal companies because they restart those plants…not even joking.
AWS is lagging in AI, but not by choice. Their leadership is in an all out panic at the moment trying to catch up.
A lot of folks agree though that the smarter play would have been just just sit more on the sidelines intentionally and wait for the bubble to burst, then buy up stuff cheap in the resulting fire sales. AWS should have just focused on the core compute and storage services, which they do well.
AWS simply doesn’t have the right talent to do well in AI at the moment and the folks they did have mostly fled elsewhere.
It can be both, they're taking it easy meaning they aren't on a hiring spree of "top talent" (like Meta paying 8-9 figures for such) which in turn becomes ineptitude since they aren't betting on it.
Even the labour market for the required skills is absurdly overvalued, if they don't see it as a major priority they won't spend the necessary rates for it.
By some napkin calc, Google has some billions of customers globally and about low hundred billions annual net income, so makes about 15-20$ per user.
With Google I enjoy free full web service coverage (search, mail and the other popular products), a huge video site, a phone ecosystem and now AI, which otherwise would cost me hundreds of dollars in subscriptions. Let advertisers and paying developers subsidise this free web service world offer (and by ad blocking you avoid "being the product").
Google will win AI as far as the most profitable for the same reason that Apple keeps winning the cell phone market - again profits not market share - vertical integration.
Google has the best infrastructure, its own chips TPUs and a huge customer base. Ben Thompson (Stratechery) and Horace Deidi (Asymco) have referenced Clay’s “Innovator’s Dilemma” about the difference between a disruptive technology and a sustaining technology.
AI is going to help BigTech that exists today get bigger.
The cloud providers - AWS, Google and Microsoft - are just exposing models as just another API to their existing customers. Google added AI overviews and an AI tab. Even Apple is going to come out ahead because developers who want local (free) inference can depend on iPhones to have decent hardware and local models.
Facebook is using AI for better ad targeting.
Where does that leave OpenAI? It’s been said plenty of times that no one is going to care about their models that are a commodity and they could put any decent model behind ChatGPT and it would be just as popular.
There was a blog post on here the other day about moving back to bare metal hosting with practical explanations and problem solving.
Cloud is great but its just borrowing someone else's machine for a fee. It's like a hyper-scalable and granular mainframe but like all mainframes, the client is powerless without it.
We need to get our acts together and not allow ourselves to be smothered in silicon nimbus (clouds) and lose track of the open sky that is the internet.
I am not suggesting going back to the 2000s with a small tower PC under someone's desk with a blinking light running a business critical cron job.
Just that we need more appetite to take responsibility of our servers and systems. Yes it takes time to manage and get up to speed, but that's empowering right....?
> Yes it takes time to manage and get up to speed, but that's empowering right....?
I rather not, but instead these compute clouds should be treated similarly to utilities - you don't really want to be running your own generators for electricity do you? So why is it that we can setup utilities properly, where everybody can trust that it stays on and usable for a price low enough?
My theory is that vendor lock in is the cause. Cloud should be a commodity, but it isn't because nobody in the cloud business wants to be a commodity. Hence, they have every incentive to prevent it from happening.
I've seen how the cloud changed over 20 years from s3/ec2 in 2006 to what we have today. I've also seen how it's built at AWS. It's ironic they call it utility computing.
What I always feared as a user was that they'd invent a new billable metric, which happened a few times. Have you ever seen a utility add them at this pace? The length of your monthly usage report shows all those items at $0 that could eventually be charged. Let that sink in.
Another interesting element is that all higher level services are built on core services such as s3/ec2. So the vendor lock in comes from all propaganda that cloud advocates have conditioned young developers with.
Notice how core utilities in many countries are state monopolies. If you want it to be a true utility, perhaps that's the solution to get them started. The state doesn't need a huge profit, but it needs sovereignty and keep foreign agents out of its DCs. Is it inefficient? Of course. But if all you really need is s3/ec2 and some networking / queuing constructs, perhaps private companies can own higher tier / lock-in services while guaranteeing it runs on such a utility. This would provide their users reduced egress fees from a true utility which doesn't need (and is not allowed to have) a 50x profit on that line item.
The cloud providers themselves run their own generators. Computers aside, every place I worked at that did critical manufacturing ran their own generators as well. That said, I agree with your general statement, that this should be a commodity, but we’ve accepted that vendor lock in is better than having a department / the cost of humans lock-in.
I think Google has at least 1 datacenter that uses batteries instead of generators.
>In Belgium, we’ll soon install the first ever battery-based system for replacing generators at a hyperscale data center. In the event of a power disruption, the system will help keep our users’ searches, e-mails, and videos on the move—without the pollution associated with burning diesel.
They only run the generators if the power company fails. To bring that analogy to servers, that would be running your loads in the cloud, but in the case of an outage, failing over to bare metal servers you run yourself. That sounds like the worst of both worlds to me.
It's not unheard of near me in a rural area to have a diesel generator in your garage as the electricity's unreliable just enough of the time, and especially given working from home is common nowadays losing electricity for 6h now and again isn't acceptable. That's not running your own generator for electricity as in your analogy, that's having a reasonable backup for short term redundancy. I think it's reasonable for companies to act similarly, rely on cloud but have a backup that provides at least degraded functionality as opposed to none.
The most pernicious effect of Big Cloud is the general erosion of infra know-how in the general public. People increasingly need to go out of their way to learn about computer hardware, networking, databases, disaster recovery, etc, when the default business decision is to outsource those tasks to a cloud service provider. For Cloud to become a commodity, computer professionals need to understand both infra and software.
Sort of an aside but utilities are done differently in many states and the expansion they are doing specifically to accommodate data centers is going to massively increase consumer rates
Seems kinda obvious why they wouldn’t want to be a commodity.
It’d lead to direct comparisons between clouds (as someone who had estimated moving legacy work loads it’s a shit load of work to get a reasonable “drive away” price for one of them and the work is duplicated to do it for the other).
If they are truly commodities then than goes away as does their margins to large degree - none of them want that.
If I lose power I can have a home generator running. These are incredibly common where I am due to hurricane outages - you’ll hear sometimes multiple on a single block powering whole homes.
S3 goes down, your entire data infrastructure (and systems attached to it) is out of reach. Local backups that can be deployed and run like S3 as if your services are still up, unless I’m mistaken, aren’t very common and would mean doing the thing you got S3 to handle in the first place. A generator is much easier to set up and use to solve the problem in a 1:1 way comparatively with zero dependence on local utilities and click on almost quite literally the second you lose power. Yeah they can’t run forever and aren’t the most cost efficient, but people use them weeks and have 100% function of all electronics in their home again.
I am not responsible for building and deploying these systems, I just depend on some of them at my job and interact with S3/media convert a ton. My understanding has always been that backups that can be restored very quickly are the aim, not trying to keep all the things attached to S3 running as if it’s still going. But if I am wrong please let me know and I would love to hear more from folks about this! I actually find this whole dance very interesting
I believe this will only be possible once the provisioning in colo is as easy as or as fast as cloud provisioning.Think Hertzner but faster/more customizable , and for that the physical commodification would need to be more nimble/modular than it is now.
Hyperscalars can drive down their costs by heavy customization of hardware/software while most colo and other builders are basically doing OEM machines or commercial rack units.
The day one can be able to rent a space, spec out the machine, its network bandwith in lessthan a day and scale it in similar timeline - the power balance will shift a bit back from the hyperscalarers.
Imagine what they could do if they focused a little bit more on making TPUs easier to use in real world applications. They are fantastic value but you feel like you're doing so much busy work to use them.
Depends on the service but for the most part googles infra has its own stack that GCP is built on top of. There is always initiatives going on to get more internal stuff running on GCP as opposed to the internal infra directly, but none of them have really stuck that I saw.
Not for the purposes of these numbers. They totally use GCP (and even "bill" themselves) but it is all isolated to one "Organization" that often gets special handling.
> engineers who know how to deliver massive systems cost-effectively
Is that true? Google just has an infinite money printing machine and a city's worth of engineers. They don't strike me as a company that fits the description you wrote.
I admit I'm biased at present since they killed off support for the Nests in my house.
Google is definitely not doing nothing. A lot of smart people are there who work very hard.
But, as a Google employee for about a decade at this point, I really do think that the company is trash when it comes to planning and coordination and is getting worse at it over time. I think that there are a lot of things it never really needed to learn how to do because the ads money is just so outrageous.
Like the resource curse for companies that stumble on a cash spigot. Still, maybe that's fine, because they wouldn't exist otherwise. Just don't expect too much from them outside that?
I wish they'd just spin out more companies so they could rise or fall on their own.
They have completely monopolized access. I was thinking am a weird exception for not ranking for the exact name/domain of my product until I have met a couple other entrepreneurs who have names/domains and still don't rank first but instead some other random website will show up.
So just the simple fact of "existing" on the Internet now costs money. Not only that, with their ads, you can target competitors search names. So you need to outbid your competitors so that your users don't get directed to your competitors by "accident". When I lost uBlock last month, I was surprised how many sites buy ads for their own names.
> So just the simple fact of "existing" on the Internet now costs money. Not only that, with their ads, you can target competitors search names. So you need to outbid your competitors so that your users don't get directed to your competitors by "accident".
To be clear, they have been doing this for so long I was first told it by a friend in the travel sector almost 20 years ago, along with the explanation of why this ends up absorbing a frightening proportion of the spare money in the economy.
The whole reason Google had such a nightmare over Facebook was FB is the only thing that broke their monopoly on this, which is why FB also prints money.
Makes sense they are concerned about losing the "browser" because it's no longer about the search/service quality and more about holding the platform that most people use for access.
It's interesting that not many investors sees this and there is little investment in an alternative browser. It is literally a software worth hundreds of billions of $$$. I hate Sam Altman as an individual but I have to admit that he could see it which is probably why he tried to grab "chrome" when an opportunity presented itself.
This is one of the reasons I believe Google will ultimately win the AI race once the VC funding tap runs dry. For all its breakthroughs, OpenAI lacks any fiscal discipline. Sam Altman seems convinced that as long as they’re delivering value and have a clear mission, the money will somehow take care of itself. That’s fine for now, but the music will stop playing one day.
When it comes to software engineering at scale, nothing beats Google. They already operate the world’s largest and (profitable) information index, they design their own chips, and employ the best engineers who know how to deliver massive systems cost-effectively. They can sustain large investment far longer than any other company simply because of how profitable all their existing businesses are. I just wish they would get their act together when it comes to product management.
I hold the exact same sentiment. The safest investment you can make in this AI race is Alphabet. The smaller players being wrapped up in this (CRWV, NOK, etc...) are going to get obliterated when the returns are not as good as previously thought. The bigger players will take a hit, but not as much as the small ones.
NVIDIA also seems like a risky bet given all the money they're passing around in circles to themselves via multi-billion dollar "investments"
> The safest investment you can make in this AI race is Alphabet.
From the technical/talent/funding side, I agree with you. With Alphabet, the real risk is that they lose interest/focus and kill (or pivot) products early. I don't really see that happening in the short term (considering the hype/investments/cash around AI) but maybe in the medium to long term if things start to get stale.
Even with that risk though, I agree with you.
The chances of Alphabet losing interest in AI are vanishingly small. Google (pre-Alphabet remaming) was the AI company before AI companies were called that.
> Google (pre-Alphabet renaming) was the AI company before AI companies were called that.
I disagree. They're an advertising/attention company.
The jury's still out on whether their AI offerings will compliment their existing business or will cannibalize it. And in the case that it cannibalizes the advertising business, will they undermine/kneecap it to protect that cash cow?
It's been a while since they've been the "don't be evil" and "organizing the world's information"-mission Google.
A company can be multiple things. Just like Apple is a Phone, Tablet, & General Computing Company
Agreed, but there's no denying that selling advertising is Google/Alphabet's primary business. And it's not even close.
Its the majority of their revenue, but the share is shrinking
Revenue attributable to advertising:
Q1: 74.1%
Q2: 74.0%
Q3: 72.5%
They didn’t lose interest in messaging when they cycled through half a dozen different apps but the problem was they were building products they wanted to exist rather than what users wanted. I’d worry about AI getting similar churn if PMs make their name by launching a new product rather than improving an existing one.
That makes sense framing it as a skepticism about Google Gemini (a single product) rather than their interest in a category (AI).
I'm less worried about them genuinely losing interest then Google outlasting their AI competition and then kneecapping their AI product to protect their cash cow.
The root comment was investor-centric, i.e. if you want to divert some investment towards AI, Alphabet is a smart bet.
Investors got great returns as Google churned messaging platforms, and this will likely be the case ad Google churns through AI products, or even entire architectures.
We did get good returns but that was because other units did very well. It’s not a given but still too likely that they’d botch AI products but still be profitable due to ad sales continuing to grow. I do think AI has better than average chances, though, because it seems to have scared their senior management enough to do their jobs: from outside, everything appears to radiate “this can’t fail” in a way they haven’t had since the 2000s.
If you look back at the dot com era, hardware companies did survive but weren't the financial "winners". Cisco is an example and a good comparison is Amazon - their bread and butter now is ecommerce, they dominate it and that was borne out of that bubble.
Alphabet also has massive market share with YouTube, solid cloud offerings, Waymo which already has driverless/unsupervised robot taxis and their old standby search which is morphing more into more AI answers when you throw questions in your browser.
Alphabet's P/E ratio is lower than the S&P 500's. Some may argue the S&P is mostly being held up by Tech and AI related companies so stock in Alphabet could be seen as undervalued and NVDIA as overvalued.
Hardware almost always ends up getting commoditized so investing in hardware companies for the long haul does carry maybe more risk than mainly software companies.
Man I remember when IBM was the safe bet.
I remember when SCO had an all hands meeting that they evaluated that linux wasn't a threat because SCO was the safe choice, the big player in UNIX on commodity hardware.
I don't think tech works that way.
It's not that easy. Tim is probably rubbing his hands until he can bring anthropic under the apple umbrella.
Might have missed the boat on that.
It’s everyone’s wet dream that Apple buys Anthropic, but I’m guessing their valuation now is already too high even for Apple. And why would Amazon let that happen?
It’s a solid match but not happening unless the AI market craters and Apple swoops in to buy them for vastly less.
> why would Amazon let that happen?
Google too. Google owns 14%.
I wouldn’t be surprised if Apple eventually decides that AI isn’t even a market worth existing in beyond some basic on-device tools.
They might say, have fun making single digit or negative profit margins on AI, we’ll sell subscriptions and iPhones and extended warranties.
All reports are that Apple isn’t really focusing on building their own server side models and is mostly just going to be a platform for other companies to fight over just like they said they don’t pay OpenAI anything for the current integration and Google pays them $20B+ a year to be the default search engine.
It makes no sense for Apple to waste money on having their own LLM when they are becoming commoditized - just chose the one that will give them the best deal.
I also think apple has put a lot of energy into providing on device hardware for this kind of stuff, like they have gone through a ton of iterations at this point on their own hardware that is also ai focused. so now they offer a lot of value in the mid to near future because if you can run ai on device why wouldn't you since you don't need to pay cloud costs anymore since users pay that for you by buying iphones.
So they’ll just die when Google has a seamless AI-first mobile experience where consumers just ask their Pixel to do X and it happens. Disruption comes for everyone, hardware isnt a moat
The issue is that Android won’t have an AI first experience where you depend on local inference because most Android phones being sold are cheap crappy devices and Android users don’t monetize.
But there is much more to a phone than “just doing stuff agentically”. Either way, you give way too much credit for Google to ever make a good end user product and be able to sell it. Google’s sells of pixels for a year are about the same number Apple sells in three weeks.
Apple’s hardware/ecosystem has been a most for 50 years.
Also Google is non existent in the country that has 1/5 of the worlds population.
Google broke my ability to set timers via voice (a core thing my phone used to do forever) for like a year. I no longer use that functionality on my phone (I can't trust Google not to break it, if it can't routinely work, it can't be part of my routine, and I personally don't enjoy cooking dinner twice after a long day).
Google replaced 'Google play the news' from playing news clips from sources I chose and trust, to an AI generated news feed that I have zero faith isn't hallucinated and from sources I have no idea about. Another multiple times daily function I used my phone for, broken by 'Google AI' because 'I should be happy with whatever they give me, it's a free product'.
Google's Youtube/Music algorithm punishes me if I flag AI slop in my feeds by removing those genres/topics, trying to force AI slop into my feed. This one I actually (well until it runs out the end of the month because I canceled) pay for.
Google has trained their users not to use Pixel's voice/AI capabilities, and to resent them (in the typical Google way, be damned with the old it's been replaced with our new half baked product that will itself be replaced shortly after the bugs are gone and it starts to work kind of OK (still really bad, but Google justifies bad is OK because it's 'FREE').
No one is integrating anything new from Google into their routine. No one is integrating Google's broken AI stuff into their daily routine, because it's friggen BROKEN. Google just one day removed me from setting voice timers with my phone, a core use case, for a year, because they could. My routine is not going back to using Google for things I count on just working because Google will arbitrarily break it and I will get the impact (maybe a burnt dinner, maybe faulty news reporting that causes me a heart attack, maybe who knows, maybe a missed important appointment tomorrow?).
Google doesn't care that I'm hit because 'it's good enough for a free product' to them. Everyone I know with a Pixel hates the ecosystem right now. Removing sideloading might fix that and improve good will I guess though? God Google doesn't know and hates their users/advocates/supporters. Google has moved themselves from 'a good solution' to 'good enough for a cheap device and free tools', and made damn sure I know it (via burnt dinners, having to move to my podcast app for (less current) news, by degrading my music/youtube feeds).
Google is a zombie company, they are dead, they just don't know it yet. Coming soon... Alphabet, a Bending Spoons company.
> Google is a zombie company, they are dead, they just don't know it yet. Coming soon... Alphabet, a Bending Spoons company
Google isn’t the next IBM. It’s the next Microsoft - boring, extremely profitable because of inertia. But crappy end user products.
apple is quite smart with this, there was a time when people thought they were dumb for not buying up tesla and ultimately shutting down their self driving division, but now it seems like a good bet.
There is literally 0 value in buying Anthropic when they can just sign a sweet deal to license Gemini (or some other model).
NVIDIA also seems like a risky bet because Google has their own chips R&D. It's not just another data center buying NVIDIA's GPUs. BTW Apple is following suit regarding chips, don't know if at some point Apple will offer any cloud service or continue to work on their ecosystem only.
Beyond that there are a lot of new chip companies attacking the market: https://news.ycombinator.com/item?id=45686790
This makes sense to me. Where I work our ai team set up a couple h100 cards and are hosting a newer model that uses up around 80GB vram. You can see the gpu utilization on graphana go to like 80% for seconds as it processes a single request. That was very surprising to me. This is $30k worth of hardware that can support only a couple users and maybe only 1 if you have an agent going. Now, maybe we're doing something wrong, but it's hard to imagine anyone is going to make money on hosting billions of dollars of these cards when you're making $20 a month per card. I guess it depends on how active your users are. Hard to imagine anthropic is right side up here.
But was that with batching? It makes a big difference. You can run many requests in parallel on the same card if you're doing LLM inferencing.
While Google has their own chips, they don't really have the market power there to buy up bleeding-edge manufacturing capacity.
Apple on the other hand... (though they're behind in other regards)
But they were just an example. They are many players arising. You can look at this: https://news.ycombinator.com/item?id=45746246 even <https://www.nextsilicon.com/> is not on that list.
I don't see a natural monopoly anymore.
I've been a GCP consultant for close to a decade now. Google messed up big time lagging behind in AI - they just kept increasing prices in other products for about 5 years straight (CDN, for example) and did nothing to productize AI. I'm of the belief that ChatGPT should've been their pilot project. What they have now (AI mode) in Google should've been there well before ChatGPT.
But, all that aside, you know what they're really good at? Google Cloud. I'm a user of all the major Cloud providers and nothing beats GCP's interface. Azure? Complicated, buggy and unreliable. There's some exploit every quarter. AWS? Overly complex security policies even to deploy a basic app, very enterprise focused and startup-unfriendly. GCP - you can be up and running on a serverless/VM instance in your lunch break. Simple, reliable, scales effortlessly. We serve 10M+ visitors on there and we've had zero issues in the last half a decade with them.
They suck at a lot of other things and they have a lot of other problems. But boy, are they good at Cloud. No wonder even Apple is their customer. It's one of the few products from Google where you can say "it just works".
I think App Engine was really ahead of its time in showing how simple cloud deployments can be. It had a similar ease of use as setting up a YouTube account. For that reason, a lot of people thought of it as a toy, which was kind of unfair because companies like Niantic were able to build global products on it. So a lot of Google Cloud afterward ended up being designed to be more "normal" like how Amazon is. Now people are seeing what normal gets them, so maybe it's going to be time for the Google way of doing things to finally shine. (Disclaimer: I'm a Google employee)
App Engine caused a huge innovator's dilemma for Google when it came to cloud. It only addressed a set of use cases around web application development- Folks would ask "why shoudl we build a low-margin cloud?" and I'd tell people: "if I can't import numpy, App Engine is useless to me". Eventually, the useful bits of AE were extracted to other services with APIs (datastore is one example), but it wasn't until google built a full GCP that they started to see real cloud growth (revenue).
"Normal" is what most of us want.
App Engine was the OG serverless before "serverless" was even a term:
https://trends.google.com/trends/explore?date=all&geo=US&q=s...
I totally second this. AppEngine was way ahead of its time. I used it for all my projects back then. The only reason I stopped using it for clients because even internal Google employees had no idea when the plug might be pulled out on AppEngine. AppEngine supported Elixir far earlier than anyone else (Elixir is my main go-to stack)
Ah, that's a shame they went that way. What I tell people when they're first getting into GCP is that it's gonna be a passion to set up what you want. But, the offerings are great, and once it's working, it'll just work
I am also a cloud consultant focused on AWS and a former employee of AWS ProServe (I have no love lost for the company).
But if you are a consultant, why pray tell are you spending that much time in the console except for occasional monitoring? Everything is either CLI commands or infrastructure as code.
And the issue with Google in particular is their customer support and business enterprise go to market sucks. The sales guys at ProServe use to run circles around GCP and never really took them seriously or had talking points. For big contracts we mostly had to compete with Azure because big boring enterprise was already using Microsoft.
When we did have to compete against GCP - and their sales team didn’t blow up the deal themselves - we just had to say “do you really want to trust Google with your workloads? Look at their history of abandoning products and raising prices”
Apple is a customer of all of the big cloud providers and was on stage at last year’s reinvent.
AWS throws money and people at startups. I’ve been on three sides - working at a 60 person startup who hosted on AWS, working at ProServe and now a third party consulting company where many deals come about because of AWS funding.
Google has the resources to win AI, but I still don't see a way to resolve the tension between good AI and ad revenue. A useful AI tells me what I need to know, but Google's cash cow is selling users to advertisers. Those are fundamentally oppositional and I don't see a clear way to resolve it.
Google can kick ass in AI all day, but soon Gemini is going to be saying shit like, "it sounds like your air conditioner is broken, maybe you should call Clearwater HVAC - they're the best!"
Once Google has to start making money from AI, talking to Gemini is going to feel like talking to that friend that invited you to dinner only to try to pitch you on an MLM.
>maybe you should call Clearwater HVAC
Wouldn't one possible solution be just adding a sidebar with ads for products/services relevant to the current topic? They don't need to change the model output to have ads. (Not advocating for ads, just pointing out there's lots of ways to deliver ads)
Do you define "AI" as LLMs?
I ask, because earlier this year I had the chance to use a Waymo for the first time. I moved a good chunk of my portfolio to GOOG after that.
To me, Waymo felt like either Magic or stepping 15+ years into the future; Demolition Man stuff (im a child of the 80s, Xenial).
Stuff like Waymo, AlphaFold and similar are just so far away from anybody else, that Alphabet could split in several companies and all of them would be hugely dominant in the future. And none of them base their revenue in advertising.
> Google can kick ass in AI all day, but soon Gemini is going to be saying shit like, "it sounds like your air conditioner is broken, maybe you should call Clearwater HVAC - they're the best!"
That is the nightmare scenario. The puzzle for me is why you apparently think it just a trap for Google. All of the pure AI companies are burning cash. That is going to stop. Either they go broke, or they do something with ads.
There is a third option I guess, and that is they create something that is useful enough to be worth paying for. It won't be just an LLM, but as far as I can tell they are all just fine tuning LLM's. We already know what the limits of an LLM are, as we are already there.
All except Google that is, who are building AI's for all sorts of things, such as protein folding, weather prediction and driving cars. LLM's were their invention, created by them for doing translation and OCR. And unlike everyone else, they aren't dependent on NVidia as they use their own internally developed hardware. Surely there is no argument about Google being the world's power house of AI?? Most of these efforts will be experiments Google is infamous for abandoning. But it's seems likely one or two will become the youtubes of AI.
I can't see anything remotely comparable from the other AI companies. It's difficult to see how this doesn't end with Google completely dominating the space, and all these LLM companies going broke.
Google is literally, today, making money from AI. Not only that but the presence of competing AI services has resulted only in record revenue and straight quarters of ad revenue growth since ChatGPT was launched.
I presume they will do what they have done successfully for the last decade or two and separate the recommendations from the ads. So you'll have gemini saying unbiasedly what it thinks and then a separate "sponsored" bit say "Buy Clearwater!"
They have an entire cloud division and they just announced it grew by 33%
and you think the competitors won't put ads in?
Wtf does “put ads in” mean? Google ads is one of the most sophisticated and profitable machines on the planet. People seem to think it’s as simple as <your ad here>.
The competitors will either figure it out or run out of money and collapse (or, both)
There is no way for anyone to resolve it, you'll either be paying premium, running last gen stuff locally, or getting ads, regardless of whether you use Google, OpenAI or anyone else.
"They already operate the world's largest and (profitable) information index, they design their own chips, and employ the best engineers who know how to deliver massive systems cost-effectively."
https://www.sec.gov/Archives/edgar/data/1652044/000165204425...
"Nature of Operations
Google was incorporated in California in September 1998 and re-incorporated in the State of Delaware in August 2003. In 2015, we implemented a holding company reorganization, and as a result, Alphabet Inc. ("Alphabet") became the successor issuer to Google.
We generate revenues by delivering relevant, cost-effective online advertising; cloud-based solutions that provide enterprise customers of all sizes with infrastructure, platform services, and applications; sales of other products and services, such as fees received for subscription-based products, apps and in-app purchases, and devices."
Would "(profitable) information index" be included in (a) "online advertising" (b) "cloud-based solutions" or (c) "sales of other products and services"
HN comment claims Alphabet operates (profitable) information index
But Alphabet SEC filing discloses its revenue-generating operations are (a) online advertising delivery, (b) cloud-based solutions and (c) sales of other products and services
US law requires the statements in Alphabet's disclosure to be accurate, not misleading
Nothing requires statements in HN comments to be accurate, not misleading
What is "(profitable) information index"
Is it part of (a) online advertising delivery, b) cloud-based solutions or (c) sales of other products and services
It's hard to understand what point you're trying to make with your weird quoting. Can you elaborate in a straightforward way?
I think it'll be Google and effectively Microsoft -- OpenAI is already so partnered with Microsoft, and if OpenAI messes up financially, Microsoft will end up bailing it out in exchange for majority ownership. So yes the music may stop playing but that doesn't mean OpenAI disappears, but maybe Sam does.
Unless there are antitrust concerns, but that's hard to see because Google is competition and Anthropic will probably remain the third more niche player.
Google is on pace to have >1B Gemini MAUs by Q2 2026.
This definitely doesn't look like it's going to be a winner takes all market.
OpenAI doesn't release MAUs, but it's possible Gemini could surpass OpenAI in MAUs (but really unlikely DAUs) by the end of 2026.
Google also own the ad space in a way that's unsurpassed, and that's the most direct route from where they are with their AI product and profitability.
Tangentially, it's sad to see AI become yet another attention factory. Sora 2 is OAI trying to get the jump on this, but I think it's quite likely that they'll get clobbered by Google and Meta when those companies start juicing their users attention in the same way - they have the ad buyers and the infrastructure for that business already, it's literally how they've made money this whole time.
> This is one of the reasons I believe Google will ultimately win the AI race once the VC funding tap runs dry.
They've already successfully monetised their chatbot (Gemini)[1] while all the others are still wondering how to get there.
-------------------
[1] I've got about 15 - 20 additions to the settings to prefix all chats with "don't show me video links", but apparently the "show youtube link" is at the system layer and cannot be overridden at all.
Wouldn't this same logic apply to IBM back in its heyday?
The biggest risk for Google is that it is already a behemoth and it has lost a lot of what made it so special. I also think they have a huge conflict of interest when it comes to AI, because it is a threat to their core business.
To be fair, Alphabet id REQUIRED to be strong in AI. Google as a search engine is losing clicks to its own AI summaries. If they didn't have that, they would risk the ad business in the future
Money alone cannot win the race; it will require a strong vision, courage, and luck. OpenAI and XAI and Anthropic (though underfunded) seem to lead in those factors so far.
If ability to sustain investment was the key apple wouldn’t be fumbling the ball so hard, wouldn’t they?
I can’t understand how they’re bleeding talent.
Isn't Microsoft maybe waiting for that and grab OpenAI at a huge discount?
Re AI their continued investment in TPUs might one day prove existentially competitive to Nvidia’a GPUs?
idk google is so flaky, one must not forget that they had the llm concept first and couldnt really materialize it into anything commercial until chat gpt came around. Google often could dominate a market, yet they kill the product/tech instead of keeping a small team to maintain it.
I think the winner will be between xai or meta, im leaning more with xai since elon seems to always have 100% conviction even with seemingly bad ideas and he is way more technical than sam & zuck
> they had the llm concept first and couldnt really materialize it into anything commercial until chat gpt came around.
Wasn't the "fuzzy search" that ignored keywords run by some sort of LLM? I.e. "lady gaga age" would return documents with "lady gaga old" etc.
> Wasn't the "fuzzy search" that ignored keywords run by some sort of LLM? I.e. "lady gaga age" would return documents with "lady gaga old" etc.
That was embeddings AFAIK, so yeah, similar technology to LLMs.
You honestly think talent is going to xAI? Why do neither of these companies have a SoTA model?
>I believe Google will ultimately win the AI race once the VC funding tap runs dry.
Your mention of "VC funding" may be a placeholder for conversational purposes but just to clarify, OpenAI is way beyond the small pocketbooks of typical Venture Capitalists. They have investments from huge sovereign wealth funds (e.g. billions from Middle East oil funds) and trillion-dollar corporations like NVIDIA, Microsoft, AMD, etc.
In other words, OpenAI has embedded itself into the money and tech ecosystem much more deeply than the dotcom failed startup Webvan burning through the limited cash of Sequoia VC.
There are lots of heavily funded vested interests that don't want to see Google be the ultimate winner (or only winner).
So did wework. Didn't stop it from going under. Some financial truths are like shouting at gravity.
Wework's CEO was not hobnobbing with the leaders of sovereign nations.
Google's new AI voice assistant finally seems to reliably set timers. The problem is they released it what, a year ago? And during that time it didn't and I now no longer trust the AI on my phone to do tasks. Google themselves trained me not to trust their AI tools.
Youtube (Google) currently degrades my feed because when I block AI videos, they assume I dislike the related topic. They have made it either I love AI videos/music, or they punish me.
You are right, they have stakeholders causing inertia to protect their fiefdoms across a large range of products. You are right, they have (profitable) current products which will prove to be barriers to nimble, future, nearterm non-profitable products. I'm not sure that's a strength seeing how they have responded so far.
They are too big and too slow to adapt. They have too many people with interests in protecting their (legacy) niche, too much inertia to overcome. The classic signs of a zombie company. Going purely off their ability to adapt in the last year, they are dead and don't even know it yet.
I have been saying the exact same thing for a while
And after people get used to using AI from Google, they discontinue it, just like they’ve done with so many other services.
Sarcasm, of course, for those who didn’t catch it - it’s hard to make jokes online.
Google has so much cash from other services…
"When it comes to software engineering at scale, nothing beats Google"
I agree with many of your statements, but this one simply isn't true. They've really struggled to integrate AI into products in a useful way. Do you remember glue on pizza? Founding Fathers reimagined for DEI? The Brain/DeepMind merger was largely an acknowledgement of their many misses from a product perspective.
The pizza glue and DEI stuff were temporary glitches on new products. Their basic engineering on search, gmail etc is pretty solid.
It's not really impressive to me in absolute terms that GCP grew by 34% because it's still in last place behind Azure. I worked with all three major CSPs at a past company and Google, by far, had the worst support, the worst network, the most issues, and at the same time they were the least willing to build custom features for us while Azure and AWS would constantly fight for our business.
IMHO I think Microsoft is a lot closer in terms of creating a real business around AI because they understand Enterprise customers. In fact they were first to offer private OpenAI deployments with AuthN/AuthZ that you could drop in your own VPC and the ability to handle sensitive data while everyone else was handing out API keys off a public endpoint like a startup. Don't get me wrong Google has had some okay technology (i.e. Gemini 2.5 is middle of the pack) but they seem like a very consumer-oriented company and don't seem to know how to market them.
Maybe I was not at your scale but I’ve had the exact opposite experience with GCP. I would prefer to use it but now I work at an all AWS shop.
I build tools for our security teams, I guarantee you the only reason we use Microsoft Azure is corporate deals. All the critical infrastructure is on AWS (including our Azure monitoring tools), and if we could move from Azure to GCP all our security experts would be happy (except maybe the red team members whose job would be harder)
You’re being way too dismissive of the importance of sales, customer support, inertia and reputation are when it comes to technology choices.
Google loses on all of those fronts.
I have used the major hyperscalers for years. I agree with you that Google is the worst business partner, but saying they have the worst network and more issues does not align with my experience at all. Azure is miles behind on technical prowess.
We use GCP at work, it works well for what we use it (VMs, container storage, cloud file storage), I wish they would stop deprecating stuff though, it just causes developer busy work with no additional value.
> it just causes developer busy work with no additional value.
Anyone remotely familiar with Google as a third party developer will notice the pattern: this will ramp up until it is almost your entire job simply dealing with their changes, most of which will be not-quite-actual-fixes to their previous round of changes.
This is not unique to Google, but it is a strategy employed to slow down development at other companies, and so help preserve the moat.
Old-timers who date back to when Joel Spolsky's early musings on the business of software development were fixtures on the HN front page will remember him using the phrase "fire and motion" for Microsoft's old strategy of constantly making changes so that everyone trying to keep up was—like the Red Queen—running as fast as they could but not getting anywhere.
> Watch out when your competition fires at you. Do they just want to force you to keep busy reacting to their volleys, so you can’t move forward?
> Think of the history of data access strategies to come out of Microsoft. ODBC, RDO, DAO, ADO, OLEDB, now ADO.NET – All New! Are these technological imperatives? The result of an incompetent design group that needs to reinvent data access every goddamn year? (That’s probably it, actually.) But the end result is just cover fire. The competition has no choice but to spend all their time porting and keeping up, time that they can’t spend writing new features. Look closely at the software landscape. The companies that do well are the ones who rely least on big companies and don’t have to spend all their cycles catching up and reimplementing and fixing bugs that crop up only on Windows XP.
—Joel Spolsky, "Fire and Motion," 2002
https://www.joelonsoftware.com/2002/01/06/fire-and-motion/
There’s a solution to the Fermi paradox somewhere in this string of comments.
> ODBC, RDO, DAO, ADO, OLEDB, now ADO.NET
Notably all of these still work, even if not getting new updates.
I used to think that made sense (as sibling mentions, Spolsky's "fire and motion" thesis)... until I worked at large-ish tech company whose internal platforms also kept doing this. Heck, the platform I owned also underwent a couple cycles of this. And so a large part of our work was just doing the Red Queen's race of deprecations and migrations.
So it was definitely not "fire and motion," as there was no competition. I think platforms genuinely need to evolve as new use-cases onboard and technology progresses, and so the assumptions underpinning the platform's design and architecture no longer hold.
However, I do think a small part of the problem was also PDD: "Promotion Driven Development."
> it is a strategy employed to slow down development at other companies, and so help preserve the moat.
Worked at Google for 7 years, and your post reminds me it is time to share a secret: it is Koyaanisqatsi* and people's base instincts unbridled, no more. There is no quicker route to irrelevancy than being the person who cares about something from last years OKRs.
* to be clear, s/Koyaanisqatsi/too big to exist healthily and yet it does exist -- edited this in, after I realized it's unclear even if you watched the movie and know the translation of the word
If they actually incentivized a group to support stability and continuity among enterprise customers, they would probably be able to diversify their revenue away from ads. Microsoft understands this…
The real sick thing is it doesn't matter, right? Like we're commenting on an article about how they won the day yesterday and Cloud revenue continues to skyrocket.
To be clear, I agree with you, and am puzzled by the lack of consequences from the real world for the stuff I saw. But that was always the mystery of Google to me, in a nutshell: How can we keep getting away with this?
> How can we keep getting away with this?
A large part of that is the Google-are-super-geniuses PR effort. Anyone pointing out that Google's products don't reflect this to their boss faces having their own credibility reduced instead.
If it's so obvious, and Google supposedly know this internally, and can obviously tell by others that some are avoiding Google because they're fast at sunsetting services, why are they not doing anything about it?
Because it’s not meaningfully hurting their bottom line.
Imaginary conversation between an honest VP and earnest year 0 me, here's what the VP says: "We definitely care about deprecations: now tell me how to accomplish that with Sundar's Focus™* Agenda over the last 2 years"
* no net headcount increases, random firings, and any new headcount should be overseas. i.e. we have the same # of people we did in 2021 with 50% more to do.
One approach to prevent being hit by "stack rot" is to build everything on top of plain Linux VMs.
Those rarely change.
Unfortunately the software deployed on top of them will.
So you either:
1) postpone all your updates for years until a bad CVE hits and you need to update or some application goes end of life and you’re screwed because updating becomes a massive exercise
2) do regular updates and patches to the entire stack, including Linux, in which case, you’re in the same position you were before with running on the stack rot treadmill
So you might’ve moved the rot to a different place, but I don’t know if you’ve reduced any of it. I’ve owned stuff deployed off of vanilla VMs and I actually found it harder to maintain because everything was a one-off.
My rationale for staying up to date aggressively is that it minimizes integration work. Basically integration work multiplies, it doesn't just accumulate. So the further you fall behind the more that can break when you finally do upgrade. And you create needlessly more work related to testing and fixing all that. Upgrading a system that was fully up to date until a few days/weeks ago is generally easy. There's only so much that changes. Doing the same to something that was last touched five years ago can be a bit painful. Apis that no longer exist. Components that are no longer supported. Code that no longer compiles. Etc.
I see a lot of teams being overly conservative with keeping their stuff up to date running with years out of date stuff with lots of known & fixed bugs of all varieties, performance issues that have long since been addressed, etc. All in the name of stability.
I treat anything that isn't up to date as technical debt. If an update breaks stuff, I need to know so I can either deal with it or document a work around or a (usually temporary) version rollback. While that happens, it doesn't happen a lot. And I prefer knowing about these things because I've tried it over being ignorant of the breakage because I haven't updated anything in years. It just adds to the hidden pile of technical debt you don't even know you have. Ignorance is not an excuse for not dealing with your technical debt. Or worse compounding it by building on top of it and creating more technical debt in the process.
Dealing with small changes over time is a lot less work than with dealing with a large delta all at once. It's something I've been doing for years. If I work on any of my projects, the first thing I do is update dependencies. Make sure stuff still works (tests). Make sure deprecated APIs are dealt with.
If you’re willing to put in the maintenance work, you’ll probably be in good shape whether you’re on plain VMs or a snazzy cloud provider managed service.
If business understands that you need time to work on these things :’)
RedShift1 complains that GCP is "deprecating stuff". I wouldn't put doing regular updates in the same problem category as having to deal with part of your stack disappearing.
To me "I wish they would stop deprecating stuff" sounds like any part of the stack has something like a 1% or even 10% chance in any given year to be shut off.
I would expect that by carefully choosing your stack from open source software in the Debian repos, you can bring the probability of any given part being gone with no successor to less than 0.1% per year. As an example - could you imagine Python becoming unavailable in 2026? Or SQLite? Docker?
Fair - I’m not very experienced with GCP, but I’ve seen AWS keep the deprecation treadmill moving as well.
In general, if I’m going to be maintaining stuff, I guess I’d rather be maintaining cloud than like… old Solaris or something.
But then you have to maintain it yourself, which overall usually wilk be more work than just migrating from time to time
Your cost just shifts elsewhere, then. Rolling your own stack from Linux up is a big endeavor too.
> it works well for what we use it
How do you handle the slow console?
I have a bunch of scripts that do the work for me so I can just run that in the background and do something else, and for one off tasks grumble and mumble a bit.
Microsoft, especially the Partner site for Microsoft Store and so on, is also exceptionally slow. Regular Azure not much better.
I guess the big providers learn from each other.
Do everything in terraform, never touch the console?
Oh my god I thought I was the only one. Everyone says that you should be using the command line anyway, but I'd rather use a GUI but it's disgustingly slow.
The CLI also takes like 5 seconds to do anything
It surprises me that it's built in Python (as is AWS'), which doesn't seem like a very appropriate language for exactly this reason. Go would seem much more apposite.
These CLIs are just thin request clients a la curl. The code execution time is peanuts in comparison to the request latency
It's not negligible versus a single request. Even `gcloud --help` takes over a second on this machine - actually getting it to do a simple list request takes almost no time longer over any reasonable connection.
Plus Python is notoriously not easy to deploy for - a Go (or Rust or whatever) binary would have almost no dependencies to worry about.
The execution speed of the language the CLI is written in doesn't matter. Any difference is dwarfed by the slowness of network calls, specifically their API response times.
It helps companies develop a vendor agnostic culture, its a great thing.
It's not just busy work, it's like being chained to a sandpaper treadmill especially in some specific "hot" domains like AI. It feels like you can build something with it and 3 months later you've got to update your code because half the dependencies are deprecated.
In AI, whatever you wrote is going to be deprecated in theee months, and in six months SOTA LLMs will one-shot it directly. That's what it means for a domain to be "hot". AI is currently the largest and most intense global R&D project in the history of humanity. So for this one field, your complaint makes no sense.
Obligatory Steve Yegge rant about the Google deprecation treadmill:
https://steve-yegge.medium.com/dear-google-cloud-your-deprec...
> Dear RECIPIENT,
> Fuck yooooouuuuuuuu. Fuck you, fuck you, Fuck You. Drop whatever you are doing because it’s not important. What is important is OUR time. It’s costing us time and money to support our shit, and we’re tired of it, so we’re not going to support it anymore. So drop your fucking plans and go start digging through our shitty documentation, begging for scraps on forums, and oh by the way, our new shit is COMPLETELY different from the old shit, because well, we fucked that design up pretty bad, heh, but hey, that’s YOUR problem, not our problem.
> We remain committed as always to ensuring everything you write will be unusable within 1 year.
> Please go fuck yourself,
> Google Cloud Platform
Great read, thanks for sharing
Impressive results. Microsoft also reported strong cloud growth yesterday.
Amazon reports today. AWS has been slipping and at current growth rates would slip to the number 2 cloud provider in a few years time… let’s see if that trend holds true today.
> Microsoft also reported strong cloud growth yesterday.
Is that why Azure was down, maybe it was out celebrating? Hahah
IMO, if Microsoft and Google are gaining this much, AWS is losing customers and mindshare. I'm not saying they probably won't report growth, but I doubt it's going to be the same as MSFT and GOOG.
Alternatively, with so much money being poured into "AI", the size of the entire cloud pie is growing. Everyone and their uncle are deploying AI, but no one does on-prem anymore - all that money has to go somewhere.
Amazon and Apple are two companies taking it conservatively in this bubble. Possibly because they actually have gone through the last bubble with some pain. Microsoft to some degree but they have the luxury of OpenAI.
We are at a point in this bubble where planned capex is approaching theorectical limits of actual energy capacity. Let’s see when they start pumping coal companies because they restart those plants…not even joking.
It's already happening. Coal use in the US was up 15% in H1/25 compared to the previous year. Partially due to a shift from higher prices gas.
Coal retirement in the US is almost entirely driven by economics. If someone needs the power, coal plants will stay open and capacity factor go up.
https://www.eia.gov/outlooks/steo/report/elec_coal_renew.php
AWS is lagging in AI, but not by choice. Their leadership is in an all out panic at the moment trying to catch up.
A lot of folks agree though that the smarter play would have been just just sit more on the sidelines intentionally and wait for the bubble to burst, then buy up stuff cheap in the resulting fire sales. AWS should have just focused on the core compute and storage services, which they do well.
AWS simply doesn’t have the right talent to do well in AI at the moment and the folks they did have mostly fled elsewhere.
Apple is "taking it easy" because of embarrassing ineptitude, not because it's their strategy
It can be both, they're taking it easy meaning they aren't on a hiring spree of "top talent" (like Meta paying 8-9 figures for such) which in turn becomes ineptitude since they aren't betting on it.
Even the labour market for the required skills is absurdly overvalued, if they don't see it as a major priority they won't spend the necessary rates for it.
They don't need the hiring spree if they don't train their own models.
They rolled out search without building a search engine and could've done the same with AI.
By some napkin calc, Google has some billions of customers globally and about low hundred billions annual net income, so makes about 15-20$ per user.
With Google I enjoy free full web service coverage (search, mail and the other popular products), a huge video site, a phone ecosystem and now AI, which otherwise would cost me hundreds of dollars in subscriptions. Let advertisers and paying developers subsidise this free web service world offer (and by ad blocking you avoid "being the product").
Google will win AI as far as the most profitable for the same reason that Apple keeps winning the cell phone market - again profits not market share - vertical integration.
Google has the best infrastructure, its own chips TPUs and a huge customer base. Ben Thompson (Stratechery) and Horace Deidi (Asymco) have referenced Clay’s “Innovator’s Dilemma” about the difference between a disruptive technology and a sustaining technology.
AI is going to help BigTech that exists today get bigger.
The cloud providers - AWS, Google and Microsoft - are just exposing models as just another API to their existing customers. Google added AI overviews and an AI tab. Even Apple is going to come out ahead because developers who want local (free) inference can depend on iPhones to have decent hardware and local models.
Facebook is using AI for better ad targeting.
Where does that leave OpenAI? It’s been said plenty of times that no one is going to care about their models that are a commodity and they could put any decent model behind ChatGPT and it would be just as popular.
There was a blog post on here the other day about moving back to bare metal hosting with practical explanations and problem solving.
Cloud is great but its just borrowing someone else's machine for a fee. It's like a hyper-scalable and granular mainframe but like all mainframes, the client is powerless without it.
We need to get our acts together and not allow ourselves to be smothered in silicon nimbus (clouds) and lose track of the open sky that is the internet.
I am not suggesting going back to the 2000s with a small tower PC under someone's desk with a blinking light running a business critical cron job.
Just that we need more appetite to take responsibility of our servers and systems. Yes it takes time to manage and get up to speed, but that's empowering right....?
> Yes it takes time to manage and get up to speed, but that's empowering right....?
I rather not, but instead these compute clouds should be treated similarly to utilities - you don't really want to be running your own generators for electricity do you? So why is it that we can setup utilities properly, where everybody can trust that it stays on and usable for a price low enough?
My theory is that vendor lock in is the cause. Cloud should be a commodity, but it isn't because nobody in the cloud business wants to be a commodity. Hence, they have every incentive to prevent it from happening.
I've seen how the cloud changed over 20 years from s3/ec2 in 2006 to what we have today. I've also seen how it's built at AWS. It's ironic they call it utility computing.
What I always feared as a user was that they'd invent a new billable metric, which happened a few times. Have you ever seen a utility add them at this pace? The length of your monthly usage report shows all those items at $0 that could eventually be charged. Let that sink in.
Another interesting element is that all higher level services are built on core services such as s3/ec2. So the vendor lock in comes from all propaganda that cloud advocates have conditioned young developers with.
Notice how core utilities in many countries are state monopolies. If you want it to be a true utility, perhaps that's the solution to get them started. The state doesn't need a huge profit, but it needs sovereignty and keep foreign agents out of its DCs. Is it inefficient? Of course. But if all you really need is s3/ec2 and some networking / queuing constructs, perhaps private companies can own higher tier / lock-in services while guaranteeing it runs on such a utility. This would provide their users reduced egress fees from a true utility which doesn't need (and is not allowed to have) a 50x profit on that line item.
The cloud providers themselves run their own generators. Computers aside, every place I worked at that did critical manufacturing ran their own generators as well. That said, I agree with your general statement, that this should be a commodity, but we’ve accepted that vendor lock in is better than having a department / the cost of humans lock-in.
I think Google has at least 1 datacenter that uses batteries instead of generators.
>In Belgium, we’ll soon install the first ever battery-based system for replacing generators at a hyperscale data center. In the event of a power disruption, the system will help keep our users’ searches, e-mails, and videos on the move—without the pollution associated with burning diesel.
https://blog.google/inside-google/infrastructure/cleaner-dat...
Disclosure: I work at Google, but no on anything related to this.
They only run the generators if the power company fails. To bring that analogy to servers, that would be running your loads in the cloud, but in the case of an outage, failing over to bare metal servers you run yourself. That sounds like the worst of both worlds to me.
It's not unheard of near me in a rural area to have a diesel generator in your garage as the electricity's unreliable just enough of the time, and especially given working from home is common nowadays losing electricity for 6h now and again isn't acceptable. That's not running your own generator for electricity as in your analogy, that's having a reasonable backup for short term redundancy. I think it's reasonable for companies to act similarly, rely on cloud but have a backup that provides at least degraded functionality as opposed to none.
The most pernicious effect of Big Cloud is the general erosion of infra know-how in the general public. People increasingly need to go out of their way to learn about computer hardware, networking, databases, disaster recovery, etc, when the default business decision is to outsource those tasks to a cloud service provider. For Cloud to become a commodity, computer professionals need to understand both infra and software.
Because utilities are commodities and natural monopolies. CSPs are neither.
Sort of an aside but utilities are done differently in many states and the expansion they are doing specifically to accommodate data centers is going to massively increase consumer rates
I feel like current VPSs feel like a commodity, most work the same, and even have the same ui
Utilities prices are also pretty damn high right now, not sure if that’s the best way to choose going forward.
Seems kinda obvious why they wouldn’t want to be a commodity.
It’d lead to direct comparisons between clouds (as someone who had estimated moving legacy work loads it’s a shit load of work to get a reasonable “drive away” price for one of them and the work is duplicated to do it for the other).
If they are truly commodities then than goes away as does their margins to large degree - none of them want that.
If I lose power I can have a home generator running. These are incredibly common where I am due to hurricane outages - you’ll hear sometimes multiple on a single block powering whole homes.
S3 goes down, your entire data infrastructure (and systems attached to it) is out of reach. Local backups that can be deployed and run like S3 as if your services are still up, unless I’m mistaken, aren’t very common and would mean doing the thing you got S3 to handle in the first place. A generator is much easier to set up and use to solve the problem in a 1:1 way comparatively with zero dependence on local utilities and click on almost quite literally the second you lose power. Yeah they can’t run forever and aren’t the most cost efficient, but people use them weeks and have 100% function of all electronics in their home again.
I am not responsible for building and deploying these systems, I just depend on some of them at my job and interact with S3/media convert a ton. My understanding has always been that backups that can be restored very quickly are the aim, not trying to keep all the things attached to S3 running as if it’s still going. But if I am wrong please let me know and I would love to hear more from folks about this! I actually find this whole dance very interesting
I believe this will only be possible once the provisioning in colo is as easy as or as fast as cloud provisioning.Think Hertzner but faster/more customizable , and for that the physical commodification would need to be more nimble/modular than it is now. Hyperscalars can drive down their costs by heavy customization of hardware/software while most colo and other builders are basically doing OEM machines or commercial rack units. The day one can be able to rent a space, spec out the machine, its network bandwith in lessthan a day and scale it in similar timeline - the power balance will shift a bit back from the hyperscalarers.
Sounds like you should apply to Oxide Computer.
Imagine what they could do if they focused a little bit more on making TPUs easier to use in real world applications. They are fantastic value but you feel like you're doing so much busy work to use them.
> Google’s search business generated $56.56 billion in revenue — up 15% from the prior year.
So, "search revenue" means ads, right? So with youtube ads, 66B out of 100B revenue is from advertising?
In other words Alphabet is still an advertising company first, and foremost?
the percentage of revenue due to ads is definitely lower than say last decade. by 2050 I expect the super majority of revenue to be non-ads.
Perhaps. I kind of expect that targeted LLM generated content will take over for ads, product placements etc.
What % of that revenue comes from this(FortNine video): https://www.youtube.com/watch?v=BfNIRyPi5QA
TLDR: Google shows the user an ad for a product they were already planning to buy at the very last minute.
How has Thomas Kurian performed as CEO? Seems like he hasn’t done too bad of a job.
GCP is growing 34% with a current annual run-rate of $60B in revenue. So I'd also say not too bad.
That is a good % of all money going to tech. There is where all the money that pay for wages is going.
One question is google it's own customer for GCP?
Basically no.
Well kind of. They use the same data center infrastructure, with internal versions of the external GCP services.
Depends on the service but for the most part googles infra has its own stack that GCP is built on top of. There is always initiatives going on to get more internal stuff running on GCP as opposed to the internal infra directly, but none of them have really stuck that I saw.
Not for the purposes of these numbers. They totally use GCP (and even "bill" themselves) but it is all isolated to one "Organization" that often gets special handling.
> engineers who know how to deliver massive systems cost-effectively
Is that true? Google just has an infinite money printing machine and a city's worth of engineers. They don't strike me as a company that fits the description you wrote.
I admit I'm biased at present since they killed off support for the Nests in my house.
It would be dandy if they were to specify how much they profit from genocide specifically.
“Over 70% of existing Google Cloud customers use its AI products, Pichai said.”
That’s not by choice, of course. It’s difficult-to-impossible to turn off.
AKA... 70% of existing google cloud users have filled out a support ticket, which starts with a chat bot.
Such a blerg nothing company with almost no innovation sitting on a cash volcano that never stops erupting.
I often think about what an incredible bargain YouTube was at $1 billion.
> I often think about what an incredible bargain YouTube was at $1 billion.
It's hard to tell if it's actually turning over a profit when they only report revenue for it.
Simple math proves that wrong. You don't make billions doing nothing.
Google is definitely not doing nothing. A lot of smart people are there who work very hard.
But, as a Google employee for about a decade at this point, I really do think that the company is trash when it comes to planning and coordination and is getting worse at it over time. I think that there are a lot of things it never really needed to learn how to do because the ads money is just so outrageous.
Like the resource curse for companies that stumble on a cash spigot. Still, maybe that's fine, because they wouldn't exist otherwise. Just don't expect too much from them outside that?
I wish they'd just spin out more companies so they could rise or fall on their own.
They have completely monopolized access. I was thinking am a weird exception for not ranking for the exact name/domain of my product until I have met a couple other entrepreneurs who have names/domains and still don't rank first but instead some other random website will show up.
So just the simple fact of "existing" on the Internet now costs money. Not only that, with their ads, you can target competitors search names. So you need to outbid your competitors so that your users don't get directed to your competitors by "accident". When I lost uBlock last month, I was surprised how many sites buy ads for their own names.
> So just the simple fact of "existing" on the Internet now costs money. Not only that, with their ads, you can target competitors search names. So you need to outbid your competitors so that your users don't get directed to your competitors by "accident".
To be clear, they have been doing this for so long I was first told it by a friend in the travel sector almost 20 years ago, along with the explanation of why this ends up absorbing a frightening proportion of the spare money in the economy.
The whole reason Google had such a nightmare over Facebook was FB is the only thing that broke their monopoly on this, which is why FB also prints money.
Makes sense they are concerned about losing the "browser" because it's no longer about the search/service quality and more about holding the platform that most people use for access.
It's interesting that not many investors sees this and there is little investment in an alternative browser. It is literally a software worth hundreds of billions of $$$. I hate Sam Altman as an individual but I have to admit that he could see it which is probably why he tried to grab "chrome" when an opportunity presented itself.
almost no innovation? from transformers to alphago to the quantum stuff.
A bit like Xerox, Bell, Kodak or IBM back in the day then?