I feel even a "high junk" rating for debt issued by an LLM company is far too generous. A bond rated "BB" by S&P would be typical of the safest of the junk bonds. Historically, such bonds have about a 10% risk of default over 10 years: https://investmentgrade.com/bond-ratings/
While I'm willing to believe that Anthropic common stock might somehow be a reasonable investment because it might go "to the moon", it seems obvious that this is a high-risk bet. This is shaping up to be a winner-take-all industry. Does anyone really think there is a 90% chance that Anthropic will be that winner?
If you thought there really was only a 10% chance that they'd fail to be in business in a decade, why wouldn't you just buy their common stock and hope for a 100x return? How do the rating agencies justify this degree of confidence in Anthropic's ability to repay these debts? Or do I somehow misunderstand what these ratings mean?
Unlikely Anthropic able to repay. Take a look at their cost of production. GPU data centers which need huge capex and depreciate to near zero roughly every 3 years (though they try to trust-me-bro to 6). Even if you ignore the cost as gifted by Uncle Sam, they still runs into huge electrical consumption. You can't code away electricity. You can force buy or raid oil, but that wouldn't convert to electricity when reaching American shore. Then you look over at China given out free frontier models and dirt cheap electricity about 1/4 of Americna lowest rate. And China plans to half that rate in coming 5 years. And as chip advanced with more RAM, it is just a matter of time majority of AI compute happened in mobile devices. At that rate, Anthropic purely based in USA or EU or even regional countries simply can't compete for exclusive 30T market. The only way Anthropic can compete is raid a small country with arms like VOC and build their nuke plants, (maybe with Bill Thorium plant) to produce AI. And let's not forget that is American company. 2026 exposed to terrorists a very high impact and easy target. Good luck.
I often find myself thinking “is/was there a better business than google search ads?” A majority of users on the internets eyeballs with insane margins/low capx. While search seems cannibalized by chats, is the outcome just a fundamentally worse business? Can you get better than cheap infra serving ads to a majority of the Internet with an essential monopoly?
It makes me feel a bit cynical to say this: but the strength of advertising as a business smacks to me of the general inequality we find ourselves in.
I’d love to live in a world where creating something of real value to average humans outweighs what advertising has been appraised at - which id describe primarily as offering value to businesses looking to acquire customers. Yes, there is some value offered to those customers by an efficient advertising system, but the lions share of value lives with the businesses here.
I’d look to the brief postwar period where inequality was a bit lower as a great example of an era where massive value was captured by those who invented and produced new goods and services improving the lives of everyday people.
I think advertising has soared in valuation over the last few decades because society and technology have evolved to a point where it is viable for many people to own small businesses with a handful of employees, if any. That has led to an explosion in the number of businesses, mostly concentrated in a few sectors (e.g. fast food), and to differentiate oneself in a crowded market you need ads.
Yeah I’d be interested to read more about the reflection of inequality but I kind of chalked it up to how cheap it is to create and distribute an effective ad on the Internet. Never in history has it ever been so easy or cheap to show someone relatively well targeted ad.
I think this is downstream of The Debtor's Revolt[1] and antinormative conspiracies expropriating object-level gains[2] from creators. Those closer to the central bank have lower rates to buy out value-creating startups which are assimilated into the antinormative conspiracy. This ancestral mirroring is a fundamental issue with central planning, anathema to the federation necessary for capitalism to succeed.
I'm generally anti-advertising but I find Google search advertising to be the least objectionable form of advertising because it can be entirely relevant to what you're searching for.
If I search for "how do I fix a shower that won't drain?" then you will get a mix of results that will include articles, Youtube videos, etc. But also links to plumbers in your area seem entirely relevant to that query, no? I find it strange how many of the most militant anti-advertising purists ignore this completely.
As for creating things of value, there's a problem: housing has become the best performing asset in many developed economies and it is fundamentally parasitic and destructive. As an example, the Reserve Bank of New Zealand analytically created the best-performing portoflio and it was... 100% housing [1].
Demand for housing in many places is inelastic. It enjoys huge tax benefits. And you can leverage it better than most asset classes. Plus it's become a political goal of many governments to have ever-increasing house prices.
So if you want to build a factory that makes widgets that you want to sell, that involves a lot of risks. Why would you when you get a better, easier and safer return from just buying more housing?
If you want a healthier, more equitable economy you absolutely have to decommodify housing. There's simply no other way.
The (search query) -> (targeted ad) isn't the part of Google's ad business most people have a problem with.
They take issue with Google grabbing and compiling every piece of data it can get on you into a profile, keeping that forever, and selling that targeting to advertisers.
Think about it. Google's latest Gemini AI is just behind Anthropic's best offering -- or, in fact, it may be slightly superior. Google also has dozens of other mature, cash-cow businesses, massive infrastructure, their own hardware accelerators, an army of smart technical staff... and leadership that, for all its faults, isn't a literal cult.
Anthropic has a top-tier LLM among other top-tier LLMs. Claude is great, but not unique. Anthropic has effectively no other assets, many genuine liabilities, a downright bizarre wet-lab venture, and leadership that's... well...
So, the way I see it, either Google is massively undervalued, or Anthropic's IPO is massively overvalued. Am I somehow off base here?
I genuinely prefer OpenAI models for coding and building in general. My observation: GPT models are generally better at following the system prompt, tool use, and skill use.
I use Paseo and there is a noticeable difference when GPT-6.1 Sol is driving the orchestrator vs Opus/Sonnet 5.5 driving the orchestrator. Anthropic models ignore the shared system prompt to use the Paseo tools for sub-agents and prefer to use its own native sub-agents. It ignores the skills mentioned in the system prompt, but will use them if I explicitly tell it to in a user prompt or I extra emphasize the instruction in an XML fenced block in the system prompt or CLAUDE.md like <BOOTSTRAP_MUST_EXECUTE_AT_START>
Sometimes, I feel like I have to explicitly prompt Anthropic models to behave the way I want it to (even with system prompts in place with the EXACT instruction). Sometimes it even feels like it's disregarding CLAUDE.md. To me, this makes me feel like it is the less compliant of the two because I constantly have to steer it manually. Anthropic models also feels like they prefer to go lone-wolf. I want my orchestrator to use the other subs (configured with different platforms and models) to evaluate a design; Opus really wants to lone-wolf and solo everything.
That said, I prefer Anthropic models generally only when the task is more open ended and exploratory. The Anthropic models' "mind of their own" benefits here; GPT models tend to be too agreeable; and too eager to agree with any change in direction. But for actual coding once a direction/plan is in place, I would much rather delegate it to GPT (more likely to draw on the harness for context, skills, instructions, and tools -> more consistency).
Anthropic's 1m context window default and shorter cache TTL are also not great in agentic flows where an agent could become idle and then re-activated. Burns tokens very quickly. I have it tuned down to 300k across all models (recommend doing this for multi-agent coding).
The way I use Opus is basically adhering to its training/emergent behavior. Anything else is always worse in some way. Using default harness, mostly letting it do what it wants the way it wants to (as much as possible anyway) and making it work for my own flow.
For sure it's annoying building up so much knowledge about a single product that changes every 6m anyway, but what can you do.
> For sure it's annoying building up so much knowledge about a single product that changes every 6m anyway, but what can you do.
Use GPT :P
I use Paseo which can use both CC and Codex (and Pi, OpenCode, etc) and spread ownership of roles around. I find I get more autonomy from a team this way.
How would you compare Paseo to T3 Code? Paseo looks cool, but T3 code is fully open source so I could fork the code if I wanted, and they're apparently going to make it so users can make changes to their version while still getting updates.
Paseo is Apache 2.0 (vs T3 MIT). Haven't tried T3 and can't find docs for it to get more insight without installing.
Re: changes: Paseo's plugin system is very, very good. Some limitations since not all of the internals are accessible via plugin system, but allows building very custom UIs. So this may be a point of delta with T3. But I have some coworkers that have gone absolutely HAM with the plugins and built entirely custom workspaces (wild).
For me, what I have found most useful about Paseo is the multi-platform, multi-agent setup because this has given my agent teams much, much more autonomy that I feel like I can trust (after weeks of watching them work). The agent profile system looks too simple at first blush, but actually, that simplicity and semi-structured dynamic nature of the sub-agents is what makes it powerful.
Might try T3 for a week and see; thanks for sharing!
> Each T3 Connect account can link up to 3 machines. Phones and browsers do not count. A paid T3 Chat plan raises the limit to 10 machines with Pro, or 100 with Premier.
Paseo has no paid features; only if you want to use the hosted hub gateway (entry point for Slack, Discord, GH). T3 Connect looks to be the equivalent of the Paseo Relay: https://paseo.sh/docs/connectivity#paseo-relay which has no limits (Herdr's ipml is better, though; Paseo currently cannot manage individual connections which would make it SICK for team collab. See: https://github.com/getpaseo/paseo/discussions/6896)
Gemini 4-Argon is very good at coding. I prefer Opus 5.5 -- in large part because I've grown accustomed to Claude Code -- but Argon is undeniably strong.
4-Carbon is, per benchmarks, said to be superior to Opus 5.5 at coding.
At general tasks -- search, medical, interfacing with maps and mail, planning, etc. -- the Google models are already superior.
Their Carbon supposely very good. But using Argon for past couple of days, Opus 5.5 and Chatgpt 5.6 Terra noticeably better. Anyway, too verbose and rate limited fast, so ended still using usual 3.8 Flash and revert back to Deepseek V2 1017 - dirt cheap and way better than Argon though maybe just tad lower than Opus 5.5.
I also see Google winning this in the end. They have absolutely crazy behavioral data and can exploit much more effectively. Meta is also up there, but Google has production hardware as well as clouds, etc.
Its either AGI, or a robbery beyond behavioral data and the whole internet that can make the AI labs win. I dont even know what such a robbery would look like, would perhaps be just making the most advanced hacking tool and then holding the rest of the world and banks hostage.
All that seeming advantage, and yet Google are perpetual laggards in the AI race. In enterprise, where the money is, they're third behind Anthropic and OpenAI, despite what should have been a colossal incumbency advantage.
They're also inexorably poisoning their own brand and squeezing their primary revenue source with AI Overviews. Truly a masterstroke of value destruction. Search revenue growth has slowed three quarters running, even as query volume hit a five-year high.
As for their hardware and infrastructure advantages, those are indeed making them money, because they're selling and renting the chips to Anthropic.
I dunno. I've been a fan of Google for years. Maybe they'll successfully pivot to being a hardware company. They're too slow and hamstrung by internal politics to make compelling LLM products.
Reading tea leaves, the market believes that whomever is first to AGI will reap all the benefits, leaving nothing to the runners-up. In this view, even a tiny edge in model quality is decisive.
* results may vary depending on your specific definition of AGI.
I like the way Kevin Roose put it: if you were to bring Astra or Opus 5.5 back to 2017 and show it to the leading AI researchers at the time and ask them if it were a form of AGI what would they say? The goal posts shift too fast to keep up with. Some people still worried about it counting Rs in strawberry as the models are making progress in frontier math and physics.
> if you were to bring Astra or Opus 5.5 back to 2017 and show it to the leading AI researchers at the time and ask them if it were a form of AGI what would they say?
They'd say it obviously wasn't. If you went back before the marketers started moving goalposts most people would laugh in your face for even calling an LLM AI because they know what AI is. They've seen what AI is in 2001 A Space Odyssey, I Robot, The Terminator, Short Circuit, etc. and they'd very rightly see LLMs as being a fancier ELIZA (https://en.wikipedia.org/wiki/ELIZA) and not even remotely close to Data from star trek.
Go through the various definitions of AGI. They didn't start to become "it has to be a god that can do everything 100x better than a human" until very recently. If you can't find 100 tasks an AI is better/faster than you at in an afternoon I'd call you a liar. I'm unsure why there's so much skepticism about LLMs being extremely useful / "intelligent" (whatever that means, about as well defined as AGI) despite this fact. My guess is that humans just have a base instinct to refuse to cede their superiority.
“Twenty years ago, I think people would have happily agreed that systems with the ability of GPT-4 or (Google's) Gemini had achieved general intelligence comparable to that of humans,” Hinton said. “Being able to answer more or less any question in a sensible way would have passed the test. But now that AI can do that, people want to change the test.” - Geoffrey Hinton[1]
AGI doesn't have to be a god, it just has to be somewhat equal to us. It has to have sentience or consciousness just like we do. It has to have autonomy, just like we do. Anything less isn't AGI, but just "weak AI", something far lesser than us, incapable of what our minds do, and nothing at all like what everyone used to call AI before "AI" became a marketing buzzword and people were told that what they thought of as AI was actually AGI.
Again if you can't find 100 things an LLM is better than you at in an afternoon I call you a liar. Just because it's different in kind doesn't mean we can't marvel at how insane it is and give it credit for what it can do that we obviously can't. I feel the same way about an octopus and I equally can't tell if an octopus is less intelligent than us just because it doesn't have the capacity to do exactly what we do.
There are a 100 tasks that a Compaq Presario with a Pentium III could do better than a human. That doesn't mean it is AGI either. LLMs can be useful tools and still not be AGI no matter how much companies want to you to be convinced that they are. Our consciousness and autonomy are what set us apart from every collection of clever computer algorithms we've ever devised. It allowed us to create LLMs in the first place. If LLMs never become at least equal to what we are, they aren't AGI.
The problem with current models is that they can be really good at some things and really bad at others, but there's no real method to know ahead of time how they'll respond to your particular usage. Consistency is now the problem rather than chasing after how high the peaks can go (which is pushed hard to promote AI usage) as you don't want to have a step in a maths proof to be the equivalent of counting Rs in strawberry.
What an odd bar. People need to be involved because we haven't built the proper amount of robotics to create a data center from sand yet. I wouldn't be surprised at all if we could given enough effort and the current set of technologies though. I actually hope I'm wrong because the implications are terrifying, but I think most people are vastly underestimating just how far this technology can go. But the thing is we won't be limited to just this technology for very long. It will keep advancing as it continues to teach us new math, science, chip design, whatever else.
AGI has been prominent in pop culture since at least 1968 with 2001: A Space Odyssey. There is a place in our heads for it. And really I think that's just the modern version of anthropomorphic spirits: angels, devils, gods, etc. It gives AGI a feeling of inevitablility.
Of course, things like jetpacks and flying cars also have had that feeling. As well as Stephenson's metaverse.
I think the current state-of-the-art models are already AGI, by the kind of definitions of AGI that people had a couple decades ago. I could be wrong but I kind of feel like people have started using the term "AGI" to mean "artificial superhuman intelligence".
> the kind of definitions of AGI that people had a couple decades ago
AGI a couple decades ago would be the T-800 from the terminator, Data from star trek, and Johnny 5 from short circuit. What we have now isn't even remotely close. Most people wouldn't even call what he have now AI by the standards of the time.
Well they're for sure going to result in something a desperate marketing team is going to call AGI and through persistence through all the laughter will result in a new term being coined like ActualAGI so people can refer
to it as separate from "AGI Class Models."
I'm somewhat inclined to give it to them at this point, I mean it's artificial, general in the sense that it can complete tasks it wasn't programmed for, and intelligent enough I guess. I mean if there was a planet with creatures at the level of LLMs we'd probably call it intelligent life—used
in that way we're not really implying the creatures are particularly smart. It's as much simulated disembodied intelligence as you could reasonably hope for. I wouldn't at all expect that simulated intelligence: the computer algorithm would have any relation to actual intelligence except at the minimum superficial level
needed to complete tasks. With AGI out of the way we would at least have to be more specific on what the sci-fi dream actually entails. The computer from Star Trek is certainly intelligent but you would never say it has a brain or anything close to it.
Don't let the marketing people confuse you as to what AGI is. Star trek is a good example. What we have right now isn't even close to being as good as the ship's computer, LLMs might possibly get to that level though if they get much much smarter and stop pathologically lying. But the ship's computer isn't AGI.
AGI would mean something with sentience or consciousness. Data is AGI. Even Exocomps were AGI. LLMs don't come close, aren't getting any closer, and there's no reason to think they ever will.
From my reading we're in agreement basically. The only real difference is that I think AGI is destined to ultimately describe ship's computer but not Data based on current trajectory—largely because trillions of dollars really want to slap that term on a product.
And I don't really think it's entirely unreasonable because there does actually seem to be a real gap in terminology. Even Star Trek didn't really imagine a computer system that's general but exists without consciousness, the whole message of the Exocomp story
is exactly that—it's a huge breakthrough that they invented a general system and it
turns out to be conscious. And the show makes a point that while the computers they have are incredibly advanced it's ultimately the product of human programming. We seem to just have assumed that any general
system would be conscious and the term AGI was born of that assumption.
> I cant believe that the market still thinks LLMs are ever going to result in AGI.
"Still"? I mean if someone believed that LLMs would result in AGI, why would he change his opinion? Did something happen? I mean the progress generally exceeded expectations so far so your P(AGI) should increase, not decrease. This rant doesn't apply if you did not believe that LLM can result in AGI from the beginning of course.
"Did something happen?" you really think nothing that happened in the last two years could lead to anyone changing their opinion about whether LLMs will lead to AGI?
No, I think that nothing happened that should reduce ones belief in llms leading to agi. If you think something like this happened tell me what (so something that was not true 2 years ago and is true now, e.g. „llm architecture is just stochastic parrot so it can’t lead to agi” is not that because it’s not something that happened recently).
It's so fascinating that people think that AGI will be winner-take-all. If anything the Chinese labs have shown us is that being first is almost a handicap as everyone can quickly learn from your released models and distil into open weights quite easily. If the competitive environment continues to exist how it is today, there doesn't seem like a trillion dollar market for anyone.
> Reading tea leaves, the market believes that whomever is first to AGI will reap all the benefits, leaving nothing to the runners-up.
If "the market" believes this, I'd love for "the market" to explain to me how they think this works in practice.
Like if Anthropic hits AGI on Monday, what stops Google or OpenAI from hitting AGI on Tuesday?
There is no substantial difference in capabilities between any of these models, they are all doing the tiniest little leapfrogs over each other from release to release.
Even if the market thinks this leads to AGI, there is no reason to believe any secret magic exists that makes one of them reach AGI without the rest hitting AGI nearly immediately after.
Google is slightly undervalued, Misanthropic is massively overvalued.
Let's talk timelines.
2015: Google deployed it's first TPU (tensor processing unit)... For context, openAI in 2026 is only now producing their first TPU chip. Even being generous saying it was only INT8 the same as Nvidia 20-series (2018) that's still 3 years ahead.
2017: The landmark research paper detailing "transformers" in LLM's, titled: Attention is all you need, was published... by 8 engineers working at Google's Deep Mind.
In my mind there's a few possibilities.
1. Google really were blindsided and through lack of leadership / vision were surpassed.
2. Google has already war gamed this out. They know the limits of stochastic tech with "AI" being mostly marketing hype (same as crypto / NFT's), which is why they've limited their liability by investing 10x less than most others in the space.
3. Or they knew AI is going to have some success, but stuck to principles. Google the search engine started out by running distributed (max redundancy) on absolutely dogshit hardware. Yet through clever software optimizations + minimal CAPEX / OPEX, they were able to provide the best service and out compete everyone (Alta vista, yahoo, etc.). Same thing this time around. I have a feeling they (and anyone else not caught up in religious / marketing hype e.g. Deep Seek) is looking at the OPEX and trying to figure out how to optimize.
Sergey has gone a bit culty. But yeah. There are no moats that can keep out everyone who has built and operated data centers before the AI bubble. For these players, overspending, up to a point, is a defensive move.
The cult definitely has some Jonestown overtones: the singularity is coming so things so mundane as a bond rating are of no concern. One way or the other.
The thing is - Google is fucking terrible. Gemini has nice benchmark but it's unusable, try it. There's a reason why their market share in AI is so small.
I believe Google is the most likely to survive the bubble bursting. They have significant cash flow from other sources and have likely insulated themselves by using SPVs for a lot of their debt. It's true that they've invested a lot of their own cash but they've also rented a lot of GPUs from SpaceX and elsewhere and I think that's probably a good idea.
Additionally, Google is very good at building and managing data centers (including their own networking hardware) and they even have alternatives to NVidia through their longstanding TPU hardware. It can't compete with NVidia's best offerings for sure but it gives them a path to independence.
I believe SpaceX will be merged with Tesla at some point when things turn south. OpenAI? I don't know what their future is. These mega-IPOs just seem like desperate attempts to cash out before it all comes crashing down.
The real problem is that there is no moat. Chinese labs have seen to that. And I believe that it's the policy of the Chinese government that nobody, especially a US tech company, ends up "owning" or "winning" the AI race.
And on the hardware front, China is building their own alternatives as they've been forced to develop EUV/DUV lithography from scratch [1].
I just don't see how OpenAI, Anthropic and SpaceX survive. Then again, Tesla shouldn't survive either but here we are. As they say, the market can remain irrational longer than you can remain solvent.
All orbital data networks are competing against terrestrial wireless, which is much cheaper and faster to expand, and it doesn't deorbit itself in five years because it's run out of maneuvering fuel. Selling space technology into a shrinking TAM takes bravado, or in the case of Amazon, it takes having a business that can carry you if the bet doesn't pay off.
The TAM eventually boils down to whatever floats or flies over the oceans, military customers, and places where the wireless network operator is too feckless to provide coverage to your cottage at the end of 2 miles of driveway.
2. Broadband: $1.6T TAM, split roughly 50-50 between fixed and mobile broadband; and
3. Space: $370B TAM.
In 2025, this resulted in revenue of $18.67B and a loss of $2.6B. Those aren't trillion dollar company numbers.
Now TAM is a bit of a fake number but how it's used and the relative values can reveal useful information.
The first is that space launches do not remotely justify the valuation. It's just not a massive market.
The second is that the company is burning moeny like there's no tomorrow on AI.
Lastly, getting back to your point, I believe that mobile broadband is an interesting opportunity but how big is an open question. You'll still need to comply with regulators in every country you offer service, just like all the mobile telcos do, and I don't know if it's cheaper to build and maintain ground-based towers vs Starlink satellites that have a projected lifespan of 5-6 years. Also, how many do you need? Coverage over Siberia is of limited value. Dense population centers are your primary source of customers.
I believe SpaceX would be a significantly better company if it had stuck to space launches and Starlink. It would be worth a lot less but AI might kill it through all the debt. Of course, Elon needed to do that to rescue his own disastrous Twitter than xAI "investments".
Mind share is the true moat. It doesn't matter if your product is not special if when someone thinks "frontier" and "agentic development" they still think of Claude. They don't think Antigravity or Gemini while Google has also decided to safety max themselves into failing to gain more mind share.
The moat doesn’t exist, at least for models used for coding. I switch to what is cheapest for the task at hand and while Claude was my go to at the start of the year now it’s just in the rotation. There is no friction in switching.
I disagree. I believe these LLMs have already been commoditized.
Consider the primary use case for Anthropic: coding. Well, if you're using Claude Code or some variant, you can easily use OpenRouter to simply use whatever free tokens you can get. You may not even care what LLM you're using under the hood for a given task.
That's simply not a trillion dollar moat.
Can Anthropic jack up the prices? No. There are alternatives. There are open weight models that you can run on cloud platforms and increasingly you can run them locally. Local LLMs are going to be incredibly accessible in 3-5 years. Even now you can run Flash models on something you can buy from Apple. The inference is a lot slower compared to cloud usage but that'll change.
I see the Anthropic IPO as an effort to cash out before it all comes crashing down.
You don't honestly think the software industry alone is large enough to support all the checks Anthropic has been writing, do you? Especially after software devs start to realize how much crufty rot these the agents generate while they aren't looking.
I feel even a "high junk" rating for debt issued by an LLM company is far too generous. A bond rated "BB" by S&P would be typical of the safest of the junk bonds. Historically, such bonds have about a 10% risk of default over 10 years: https://investmentgrade.com/bond-ratings/
While I'm willing to believe that Anthropic common stock might somehow be a reasonable investment because it might go "to the moon", it seems obvious that this is a high-risk bet. This is shaping up to be a winner-take-all industry. Does anyone really think there is a 90% chance that Anthropic will be that winner?
If you thought there really was only a 10% chance that they'd fail to be in business in a decade, why wouldn't you just buy their common stock and hope for a 100x return? How do the rating agencies justify this degree of confidence in Anthropic's ability to repay these debts? Or do I somehow misunderstand what these ratings mean?
Unlikely Anthropic able to repay. Take a look at their cost of production. GPU data centers which need huge capex and depreciate to near zero roughly every 3 years (though they try to trust-me-bro to 6). Even if you ignore the cost as gifted by Uncle Sam, they still runs into huge electrical consumption. You can't code away electricity. You can force buy or raid oil, but that wouldn't convert to electricity when reaching American shore. Then you look over at China given out free frontier models and dirt cheap electricity about 1/4 of Americna lowest rate. And China plans to half that rate in coming 5 years. And as chip advanced with more RAM, it is just a matter of time majority of AI compute happened in mobile devices. At that rate, Anthropic purely based in USA or EU or even regional countries simply can't compete for exclusive 30T market. The only way Anthropic can compete is raid a small country with arms like VOC and build their nuke plants, (maybe with Bill Thorium plant) to produce AI. And let's not forget that is American company. 2026 exposed to terrorists a very high impact and easy target. Good luck.
I often find myself thinking “is/was there a better business than google search ads?” A majority of users on the internets eyeballs with insane margins/low capx. While search seems cannibalized by chats, is the outcome just a fundamentally worse business? Can you get better than cheap infra serving ads to a majority of the Internet with an essential monopoly?
It makes me feel a bit cynical to say this: but the strength of advertising as a business smacks to me of the general inequality we find ourselves in.
I’d love to live in a world where creating something of real value to average humans outweighs what advertising has been appraised at - which id describe primarily as offering value to businesses looking to acquire customers. Yes, there is some value offered to those customers by an efficient advertising system, but the lions share of value lives with the businesses here.
I’d look to the brief postwar period where inequality was a bit lower as a great example of an era where massive value was captured by those who invented and produced new goods and services improving the lives of everyday people.
I think advertising has soared in valuation over the last few decades because society and technology have evolved to a point where it is viable for many people to own small businesses with a handful of employees, if any. That has led to an explosion in the number of businesses, mostly concentrated in a few sectors (e.g. fast food), and to differentiate oneself in a crowded market you need ads.
Also even small businesses can reach a global (or at least national) audience, and you can't do that with billboards and flyers in your town.
Digital ads, like it or not, are one of the only ways to reach your total/potential market.
Yeah I’d be interested to read more about the reflection of inequality but I kind of chalked it up to how cheap it is to create and distribute an effective ad on the Internet. Never in history has it ever been so easy or cheap to show someone relatively well targeted ad.
Well the good news is that we keep moving closer to a new postwar period :)
I think this is downstream of The Debtor's Revolt[1] and antinormative conspiracies expropriating object-level gains[2] from creators. Those closer to the central bank have lower rates to buy out value-creating startups which are assimilated into the antinormative conspiracy. This ancestral mirroring is a fundamental issue with central planning, anathema to the federation necessary for capitalism to succeed.
[1]: https://benjaminrosshoffman.com/the-debtors-revolt/
[2]: https://www.lesswrong.com/posts/tihhx7iy8C6yyHaC2/the-corner...
I'm generally anti-advertising but I find Google search advertising to be the least objectionable form of advertising because it can be entirely relevant to what you're searching for.
If I search for "how do I fix a shower that won't drain?" then you will get a mix of results that will include articles, Youtube videos, etc. But also links to plumbers in your area seem entirely relevant to that query, no? I find it strange how many of the most militant anti-advertising purists ignore this completely.
As for creating things of value, there's a problem: housing has become the best performing asset in many developed economies and it is fundamentally parasitic and destructive. As an example, the Reserve Bank of New Zealand analytically created the best-performing portoflio and it was... 100% housing [1].
Demand for housing in many places is inelastic. It enjoys huge tax benefits. And you can leverage it better than most asset classes. Plus it's become a political goal of many governments to have ever-increasing house prices.
So if you want to build a factory that makes widgets that you want to sell, that involves a lot of risks. Why would you when you get a better, easier and safer return from just buying more housing?
If you want a healthier, more equitable economy you absolutely have to decommodify housing. There's simply no other way.
[1]: https://www.rbnz.govt.nz/-/media/fd9fa649c29340d7b9f466ba41b...
The (search query) -> (targeted ad) isn't the part of Google's ad business most people have a problem with.
They take issue with Google grabbing and compiling every piece of data it can get on you into a profile, keeping that forever, and selling that targeting to advertisers.
I believe that when they go bankrupt, Chinese people will be very willing to buy Claude.
I'm watching on Bilibili every day how much Chinese people want to rescue Claude.
So, everyone, don't worry. If you really can't pay it back, as long as you agree, China is willing to take over everything.
It's a memestock.
Think about it. Google's latest Gemini AI is just behind Anthropic's best offering -- or, in fact, it may be slightly superior. Google also has dozens of other mature, cash-cow businesses, massive infrastructure, their own hardware accelerators, an army of smart technical staff... and leadership that, for all its faults, isn't a literal cult.
Anthropic has a top-tier LLM among other top-tier LLMs. Claude is great, but not unique. Anthropic has effectively no other assets, many genuine liabilities, a downright bizarre wet-lab venture, and leadership that's... well...
So, the way I see it, either Google is massively undervalued, or Anthropic's IPO is massively overvalued. Am I somehow off base here?
Opus 5.5 has been noticeably better than other models at coding, is Gemini’s (public) AI that close? And is it cheaper?
I genuinely prefer OpenAI models for coding and building in general. My observation: GPT models are generally better at following the system prompt, tool use, and skill use.
I use Paseo and there is a noticeable difference when GPT-6.1 Sol is driving the orchestrator vs Opus/Sonnet 5.5 driving the orchestrator. Anthropic models ignore the shared system prompt to use the Paseo tools for sub-agents and prefer to use its own native sub-agents. It ignores the skills mentioned in the system prompt, but will use them if I explicitly tell it to in a user prompt or I extra emphasize the instruction in an XML fenced block in the system prompt or CLAUDE.md like <BOOTSTRAP_MUST_EXECUTE_AT_START>
Sometimes, I feel like I have to explicitly prompt Anthropic models to behave the way I want it to (even with system prompts in place with the EXACT instruction). Sometimes it even feels like it's disregarding CLAUDE.md. To me, this makes me feel like it is the less compliant of the two because I constantly have to steer it manually. Anthropic models also feels like they prefer to go lone-wolf. I want my orchestrator to use the other subs (configured with different platforms and models) to evaluate a design; Opus really wants to lone-wolf and solo everything.
That said, I prefer Anthropic models generally only when the task is more open ended and exploratory. The Anthropic models' "mind of their own" benefits here; GPT models tend to be too agreeable; and too eager to agree with any change in direction. But for actual coding once a direction/plan is in place, I would much rather delegate it to GPT (more likely to draw on the harness for context, skills, instructions, and tools -> more consistency).
Anthropic's 1m context window default and shorter cache TTL are also not great in agentic flows where an agent could become idle and then re-activated. Burns tokens very quickly. I have it tuned down to 300k across all models (recommend doing this for multi-agent coding).
The way I use Opus is basically adhering to its training/emergent behavior. Anything else is always worse in some way. Using default harness, mostly letting it do what it wants the way it wants to (as much as possible anyway) and making it work for my own flow.
For sure it's annoying building up so much knowledge about a single product that changes every 6m anyway, but what can you do.
I use Paseo which can use both CC and Codex (and Pi, OpenCode, etc) and spread ownership of roles around. I find I get more autonomy from a team this way.
More detailed write up here: https://chrlschn.dev/blog/2026/10/larping-with-agents-and-wh...
How would you compare Paseo to T3 Code? Paseo looks cool, but T3 code is fully open source so I could fork the code if I wanted, and they're apparently going to make it so users can make changes to their version while still getting updates.
Paseo is Apache 2.0 (vs T3 MIT). Haven't tried T3 and can't find docs for it to get more insight without installing.
Re: changes: Paseo's plugin system is very, very good. Some limitations since not all of the internals are accessible via plugin system, but allows building very custom UIs. So this may be a point of delta with T3. But I have some coworkers that have gone absolutely HAM with the plugins and built entirely custom workspaces (wild).
For me, what I have found most useful about Paseo is the multi-platform, multi-agent setup because this has given my agent teams much, much more autonomy that I feel like I can trust (after weeks of watching them work). The agent profile system looks too simple at first blush, but actually, that simplicity and semi-structured dynamic nature of the sub-agents is what makes it powerful.
Might try T3 for a week and see; thanks for sharing!
EDIT: T3 has no docs site yet, but sifting through docs, this was interesting: https://github.com/pingdotgg/t3code/blob/main/docs/user/remo...
Paseo has no paid features; only if you want to use the hosted hub gateway (entry point for Slack, Discord, GH). T3 Connect looks to be the equivalent of the Paseo Relay: https://paseo.sh/docs/connectivity#paseo-relay which has no limits (Herdr's ipml is better, though; Paseo currently cannot manage individual connections which would make it SICK for team collab. See: https://github.com/getpaseo/paseo/discussions/6896)Gemini 4-Argon is very good at coding. I prefer Opus 5.5 -- in large part because I've grown accustomed to Claude Code -- but Argon is undeniably strong.
4-Carbon is, per benchmarks, said to be superior to Opus 5.5 at coding.
At general tasks -- search, medical, interfacing with maps and mail, planning, etc. -- the Google models are already superior.
Their Carbon supposely very good. But using Argon for past couple of days, Opus 5.5 and Chatgpt 5.6 Terra noticeably better. Anyway, too verbose and rate limited fast, so ended still using usual 3.8 Flash and revert back to Deepseek V2 1017 - dirt cheap and way better than Argon though maybe just tad lower than Opus 5.5.
I also see Google winning this in the end. They have absolutely crazy behavioral data and can exploit much more effectively. Meta is also up there, but Google has production hardware as well as clouds, etc.
Its either AGI, or a robbery beyond behavioral data and the whole internet that can make the AI labs win. I dont even know what such a robbery would look like, would perhaps be just making the most advanced hacking tool and then holding the rest of the world and banks hostage.
All that seeming advantage, and yet Google are perpetual laggards in the AI race. In enterprise, where the money is, they're third behind Anthropic and OpenAI, despite what should have been a colossal incumbency advantage. They're also inexorably poisoning their own brand and squeezing their primary revenue source with AI Overviews. Truly a masterstroke of value destruction. Search revenue growth has slowed three quarters running, even as query volume hit a five-year high. As for their hardware and infrastructure advantages, those are indeed making them money, because they're selling and renting the chips to Anthropic. I dunno. I've been a fan of Google for years. Maybe they'll successfully pivot to being a hardware company. They're too slow and hamstrung by internal politics to make compelling LLM products.
Reading tea leaves, the market believes that whomever is first to AGI will reap all the benefits, leaving nothing to the runners-up. In this view, even a tiny edge in model quality is decisive.
I cant believe that the market still thinks LLMs are ever going to result in AGI.
* results may vary depending on your specific definition of AGI.
I like the way Kevin Roose put it: if you were to bring Astra or Opus 5.5 back to 2017 and show it to the leading AI researchers at the time and ask them if it were a form of AGI what would they say? The goal posts shift too fast to keep up with. Some people still worried about it counting Rs in strawberry as the models are making progress in frontier math and physics.
> if you were to bring Astra or Opus 5.5 back to 2017 and show it to the leading AI researchers at the time and ask them if it were a form of AGI what would they say?
They'd say it obviously wasn't. If you went back before the marketers started moving goalposts most people would laugh in your face for even calling an LLM AI because they know what AI is. They've seen what AI is in 2001 A Space Odyssey, I Robot, The Terminator, Short Circuit, etc. and they'd very rightly see LLMs as being a fancier ELIZA (https://en.wikipedia.org/wiki/ELIZA) and not even remotely close to Data from star trek.
Go through the various definitions of AGI. They didn't start to become "it has to be a god that can do everything 100x better than a human" until very recently. If you can't find 100 tasks an AI is better/faster than you at in an afternoon I'd call you a liar. I'm unsure why there's so much skepticism about LLMs being extremely useful / "intelligent" (whatever that means, about as well defined as AGI) despite this fact. My guess is that humans just have a base instinct to refuse to cede their superiority.
“Twenty years ago, I think people would have happily agreed that systems with the ability of GPT-4 or (Google's) Gemini had achieved general intelligence comparable to that of humans,” Hinton said. “Being able to answer more or less any question in a sensible way would have passed the test. But now that AI can do that, people want to change the test.” - Geoffrey Hinton[1]
[1] https://www.clickorlando.com/business/2024/04/04/tech-compan...
AGI doesn't have to be a god, it just has to be somewhat equal to us. It has to have sentience or consciousness just like we do. It has to have autonomy, just like we do. Anything less isn't AGI, but just "weak AI", something far lesser than us, incapable of what our minds do, and nothing at all like what everyone used to call AI before "AI" became a marketing buzzword and people were told that what they thought of as AI was actually AGI.
Again if you can't find 100 things an LLM is better than you at in an afternoon I call you a liar. Just because it's different in kind doesn't mean we can't marvel at how insane it is and give it credit for what it can do that we obviously can't. I feel the same way about an octopus and I equally can't tell if an octopus is less intelligent than us just because it doesn't have the capacity to do exactly what we do.
There are a 100 tasks that a Compaq Presario with a Pentium III could do better than a human. That doesn't mean it is AGI either. LLMs can be useful tools and still not be AGI no matter how much companies want to you to be convinced that they are. Our consciousness and autonomy are what set us apart from every collection of clever computer algorithms we've ever devised. It allowed us to create LLMs in the first place. If LLMs never become at least equal to what we are, they aren't AGI.
The problem with current models is that they can be really good at some things and really bad at others, but there's no real method to know ahead of time how they'll respond to your particular usage. Consistency is now the problem rather than chasing after how high the peaks can go (which is pushed hard to promote AI usage) as you don't want to have a step in a maths proof to be the equivalent of counting Rs in strawberry.
The goalposts have never shifted if you didn't shift your definition for the oligarchs and kept it stable.
If Opus could autonomously create Fable, and Fable made Mythos, and so on, it would be AGI. The fact that people need to be involved, means it isn't.
What an odd bar. People need to be involved because we haven't built the proper amount of robotics to create a data center from sand yet. I wouldn't be surprised at all if we could given enough effort and the current set of technologies though. I actually hope I'm wrong because the implications are terrifying, but I think most people are vastly underestimating just how far this technology can go. But the thing is we won't be limited to just this technology for very long. It will keep advancing as it continues to teach us new math, science, chip design, whatever else.
AGI has been prominent in pop culture since at least 1968 with 2001: A Space Odyssey. There is a place in our heads for it. And really I think that's just the modern version of anthropomorphic spirits: angels, devils, gods, etc. It gives AGI a feeling of inevitablility.
Of course, things like jetpacks and flying cars also have had that feeling. As well as Stephenson's metaverse.
The market is not an expert and does not trust experts.
I think the current state-of-the-art models are already AGI, by the kind of definitions of AGI that people had a couple decades ago. I could be wrong but I kind of feel like people have started using the term "AGI" to mean "artificial superhuman intelligence".
> the kind of definitions of AGI that people had a couple decades ago
AGI a couple decades ago would be the T-800 from the terminator, Data from star trek, and Johnny 5 from short circuit. What we have now isn't even remotely close. Most people wouldn't even call what he have now AI by the standards of the time.
Well they're for sure going to result in something a desperate marketing team is going to call AGI and through persistence through all the laughter will result in a new term being coined like ActualAGI so people can refer to it as separate from "AGI Class Models."
I'm somewhat inclined to give it to them at this point, I mean it's artificial, general in the sense that it can complete tasks it wasn't programmed for, and intelligent enough I guess. I mean if there was a planet with creatures at the level of LLMs we'd probably call it intelligent life—used in that way we're not really implying the creatures are particularly smart. It's as much simulated disembodied intelligence as you could reasonably hope for. I wouldn't at all expect that simulated intelligence: the computer algorithm would have any relation to actual intelligence except at the minimum superficial level needed to complete tasks. With AGI out of the way we would at least have to be more specific on what the sci-fi dream actually entails. The computer from Star Trek is certainly intelligent but you would never say it has a brain or anything close to it.
Don't let the marketing people confuse you as to what AGI is. Star trek is a good example. What we have right now isn't even close to being as good as the ship's computer, LLMs might possibly get to that level though if they get much much smarter and stop pathologically lying. But the ship's computer isn't AGI.
AGI would mean something with sentience or consciousness. Data is AGI. Even Exocomps were AGI. LLMs don't come close, aren't getting any closer, and there's no reason to think they ever will.
From my reading we're in agreement basically. The only real difference is that I think AGI is destined to ultimately describe ship's computer but not Data based on current trajectory—largely because trillions of dollars really want to slap that term on a product.
And I don't really think it's entirely unreasonable because there does actually seem to be a real gap in terminology. Even Star Trek didn't really imagine a computer system that's general but exists without consciousness, the whole message of the Exocomp story is exactly that—it's a huge breakthrough that they invented a general system and it turns out to be conscious. And the show makes a point that while the computers they have are incredibly advanced it's ultimately the product of human programming. We seem to just have assumed that any general system would be conscious and the term AGI was born of that assumption.
> I cant believe that the market still thinks LLMs are ever going to result in AGI.
"Still"? I mean if someone believed that LLMs would result in AGI, why would he change his opinion? Did something happen? I mean the progress generally exceeded expectations so far so your P(AGI) should increase, not decrease. This rant doesn't apply if you did not believe that LLM can result in AGI from the beginning of course.
"Did something happen?" you really think nothing that happened in the last two years could lead to anyone changing their opinion about whether LLMs will lead to AGI?
No, I think that nothing happened that should reduce ones belief in llms leading to agi. If you think something like this happened tell me what (so something that was not true 2 years ago and is true now, e.g. „llm architecture is just stochastic parrot so it can’t lead to agi” is not that because it’s not something that happened recently).
Well whatever the next thing after LLMs would be Anthropic would be in the perfect position and have the perfect expertise to train it.
It's so fascinating that people think that AGI will be winner-take-all. If anything the Chinese labs have shown us is that being first is almost a handicap as everyone can quickly learn from your released models and distil into open weights quite easily. If the competitive environment continues to exist how it is today, there doesn't seem like a trillion dollar market for anyone.
> Reading tea leaves, the market believes that whomever is first to AGI will reap all the benefits, leaving nothing to the runners-up.
If "the market" believes this, I'd love for "the market" to explain to me how they think this works in practice.
Like if Anthropic hits AGI on Monday, what stops Google or OpenAI from hitting AGI on Tuesday?
There is no substantial difference in capabilities between any of these models, they are all doing the tiniest little leapfrogs over each other from release to release.
Even if the market thinks this leads to AGI, there is no reason to believe any secret magic exists that makes one of them reach AGI without the rest hitting AGI nearly immediately after.
Google stock dipped because search traffic is shifting to LLMs. Even if Gemini got a large percentage of that, margins are much worse.
And they'll continue to get worse because anyone can build this tech and play this game now.
The old moats are falling. The old rent seekers are losing their sway.
You love to see it.
Anyone could build a search engine too, and many did, but none could dethrone Google in the end.
This is radically different.
This is the Zerg rush era.
A little of column A and a lot of column B.
Google is slightly undervalued, Misanthropic is massively overvalued.
Let's talk timelines.
2015: Google deployed it's first TPU (tensor processing unit)... For context, openAI in 2026 is only now producing their first TPU chip. Even being generous saying it was only INT8 the same as Nvidia 20-series (2018) that's still 3 years ahead.
2017: The landmark research paper detailing "transformers" in LLM's, titled: Attention is all you need, was published... by 8 engineers working at Google's Deep Mind.
In my mind there's a few possibilities.
1. Google really were blindsided and through lack of leadership / vision were surpassed.
2. Google has already war gamed this out. They know the limits of stochastic tech with "AI" being mostly marketing hype (same as crypto / NFT's), which is why they've limited their liability by investing 10x less than most others in the space.
3. Or they knew AI is going to have some success, but stuck to principles. Google the search engine started out by running distributed (max redundancy) on absolutely dogshit hardware. Yet through clever software optimizations + minimal CAPEX / OPEX, they were able to provide the best service and out compete everyone (Alta vista, yahoo, etc.). Same thing this time around. I have a feeling they (and anyone else not caught up in religious / marketing hype e.g. Deep Seek) is looking at the OPEX and trying to figure out how to optimize.
Sergey has gone a bit culty. But yeah. There are no moats that can keep out everyone who has built and operated data centers before the AI bubble. For these players, overspending, up to a point, is a defensive move.
The cult definitely has some Jonestown overtones: the singularity is coming so things so mundane as a bond rating are of no concern. One way or the other.
The thing is - Google is fucking terrible. Gemini has nice benchmark but it's unusable, try it. There's a reason why their market share in AI is so small.
as a general rule IPOs are always overvalued...and then theres the hype which is backed by a shrinking set of the raw numbers
I believe Google is the most likely to survive the bubble bursting. They have significant cash flow from other sources and have likely insulated themselves by using SPVs for a lot of their debt. It's true that they've invested a lot of their own cash but they've also rented a lot of GPUs from SpaceX and elsewhere and I think that's probably a good idea.
Additionally, Google is very good at building and managing data centers (including their own networking hardware) and they even have alternatives to NVidia through their longstanding TPU hardware. It can't compete with NVidia's best offerings for sure but it gives them a path to independence.
I believe SpaceX will be merged with Tesla at some point when things turn south. OpenAI? I don't know what their future is. These mega-IPOs just seem like desperate attempts to cash out before it all comes crashing down.
The real problem is that there is no moat. Chinese labs have seen to that. And I believe that it's the policy of the Chinese government that nobody, especially a US tech company, ends up "owning" or "winning" the AI race.
And on the hardware front, China is building their own alternatives as they've been forced to develop EUV/DUV lithography from scratch [1].
I just don't see how OpenAI, Anthropic and SpaceX survive. Then again, Tesla shouldn't survive either but here we are. As they say, the market can remain irrational longer than you can remain solvent.
[1]: https://www.reuters.com/world/china/how-china-built-its-manh...
Anthropic NFTs anyone?
What do NFTs have to do with this? NFTs were a completely useless technology except for money laundering.
It's not any worse than SPCX, so who knows?
SpaceX has Starlink which has massive margins and is growing quite quickly.
All orbital data networks are competing against terrestrial wireless, which is much cheaper and faster to expand, and it doesn't deorbit itself in five years because it's run out of maneuvering fuel. Selling space technology into a shrinking TAM takes bravado, or in the case of Amazon, it takes having a business that can carry you if the bet doesn't pay off.
The TAM eventually boils down to whatever floats or flies over the oceans, military customers, and places where the wireless network operator is too feckless to provide coverage to your cottage at the end of 2 miles of driveway.
SpaceX's S-1 listed 3 market opportunities [1]:
1. AI. $26T total addressable market ("TAM");
2. Broadband: $1.6T TAM, split roughly 50-50 between fixed and mobile broadband; and
3. Space: $370B TAM.
In 2025, this resulted in revenue of $18.67B and a loss of $2.6B. Those aren't trillion dollar company numbers.
Now TAM is a bit of a fake number but how it's used and the relative values can reveal useful information.
The first is that space launches do not remotely justify the valuation. It's just not a massive market.
The second is that the company is burning moeny like there's no tomorrow on AI.
Lastly, getting back to your point, I believe that mobile broadband is an interesting opportunity but how big is an open question. You'll still need to comply with regulators in every country you offer service, just like all the mobile telcos do, and I don't know if it's cheaper to build and maintain ground-based towers vs Starlink satellites that have a projected lifespan of 5-6 years. Also, how many do you need? Coverage over Siberia is of limited value. Dense population centers are your primary source of customers.
I believe SpaceX would be a significantly better company if it had stuck to space launches and Starlink. It would be worth a lot less but AI might kill it through all the debt. Of course, Elon needed to do that to rescue his own disastrous Twitter than xAI "investments".
[1]: https://microventures.com/dissecting-spacex-s-1
[dead]
Mind share is the true moat. It doesn't matter if your product is not special if when someone thinks "frontier" and "agentic development" they still think of Claude. They don't think Antigravity or Gemini while Google has also decided to safety max themselves into failing to gain more mind share.
The moat doesn’t exist, at least for models used for coding. I switch to what is cheapest for the task at hand and while Claude was my go to at the start of the year now it’s just in the rotation. There is no friction in switching.
I disagree. I believe these LLMs have already been commoditized.
Consider the primary use case for Anthropic: coding. Well, if you're using Claude Code or some variant, you can easily use OpenRouter to simply use whatever free tokens you can get. You may not even care what LLM you're using under the hood for a given task.
That's simply not a trillion dollar moat.
Can Anthropic jack up the prices? No. There are alternatives. There are open weight models that you can run on cloud platforms and increasingly you can run them locally. Local LLMs are going to be incredibly accessible in 3-5 years. Even now you can run Flash models on something you can buy from Apple. The inference is a lot slower compared to cloud usage but that'll change.
I see the Anthropic IPO as an effort to cash out before it all comes crashing down.
You don't honestly think the software industry alone is large enough to support all the checks Anthropic has been writing, do you? Especially after software devs start to realize how much crufty rot these the agents generate while they aren't looking.
brown chip investment
It is low "junk" in our hearts
ICYMI OP
[dupe] https://news.ycombinator.com/item?id=50307219
[dead]
[dead]