As uncomfortable as this may be for the tech industry, I have been hoping for AI related stock prices to come back to reality. A market correction is a healthy thing so long as it doesn't turn into a collapse. IMO, some AI companies have ludicrous valuations considering their astronomical debt. A reality check may be healthy long term.
IMO OpenAI jumped the gun and rushed an immature product to be there first, no matter the costs. The other players saw it as a threat, and started running with the same carelessness. It has been spiraling since then, and while the product keeps improving, their business model hasn't demonstrated sustainability.
Their CEOs know it, and they must have their end game. I'd hate to think they'll just keep running until a hard crash. Hopefully small corrections will slow this down in a more controlled manner.
And I wholeheartedly thank China for doing this. The open weight models are fantastic, and make power consolidation of AI monopolies less likely. Open weights bring real competition to the space and help commoditize inference.
Good hardware companies are now trading at far under ten P/E, hardly ludicrous valuations. Sure there are some poor companies out there, but Nvidia and Micron are not among them. SMCI just beat substantially. Buying opportunity when things settle down.
Nvidia has not been a buying opportunity for a while now. Training new models is the moat it possessed. The models are as good as they can be and have diminished returns beyond this. TPUs by google and CPUs by AMD are eating into their inference margins.
> Yeah I'm not sure how Anthropic is going to IPO at this rate...
There are plenty of suckers willing to gamble on a failing business (massive and rapidly growing debts, assets that can easily be cloned, no realistic path to profitability), so I don't think an IPO is off the table.
The moat is GPUs/TPUs, ram and compute. When you need $200k of compute to run the frontier models, you are likely to be renting access to generate your tokens.
Current RAM is no moat. Upcoming Compute-in-RAM is. And nobody has that built up at scale yet. So, while a moat might become possible, we only have yesteryear's hardware as "moat".
The academic term is now "Compute-in-Memory". That's MOSTLY what has suddenly been hiring SRAM and DRAM designers across the industry this year. It is part of the fast evolving field of doing inference efficiently. This has been all the rage in several Electronics conferences this year.
Please don't rely on my [inferior] explanation and search for yourself (start here for insance: https://www.google.com/search?q=what+is+compute+in+memory+an...). Asking about a new field is much better done with AI these days, because you can get it to explain just the interests and questions you personally have, rather than a generic and boring explanation.
In short, better than HBM is not having to move data in and out of memory at all or at least not as much. The compute part in inference is much simpler than the data hauling part.
Couple of examples: Check out Axelera's Europa board (upcoming) or their Metis board (existing). Or check out Mythic.ai which FINALLY managed to raise funds in Nov / Dec 2025. Ironically, her PhD thesis had been gathering dust before that.
That is not a moat. (and it is not theirs either) It just means that their models are highly inefficient that it requires a lot of compute to run them.
Open weight models such as Deepseek, GLM are just as good as the frontier closed models and already we have bank such as Revolut migrating off of frontier closed models and self-hosting. [0]
Not sure if you're joking, but I go back and forth between this thought. Finance is not my forte, far from it for sure, but there's a part of me that hopes a collapse can at least open the door for future opportunity. A "reset" if you will. The younger generations are up against a massive wall when it comes to acquiring wealth (or even a reasonable means of living at this point), it would be nice to give them an equal chance. I know this is very likely a naive take, but it's the only hopeful aspect of a collapse I can accept.
Not quite. Poor people suffer more in real terms, but rich people suffer more in nominal (dollar) terms. Nominal inequality gets reset.
Don't let the term "nominal" undersell this: it's nominal inequality that locks working people out of assets and imposes upon them the need to service the (otherwise) exponentially growing profit expectations of their economic complements. At no point in the cycle is it better to be poor than rich, but the contractionary quadrant is the one where the gap shrinks.
Inequality declined temporarily during the recession, but reached new highs during the recovery.
2008 decreased inequality for a very short time, but inequality increased overall. The rich have the available assets to profit from instability and are able to ride out unemployment, for the poor a crash and losing their job can be existential.
Why would it? The wealthy people who haven’t allocated their capital to those assets that are depressed from a recession can reallocate to the depressed assets and end up even wealthier on the rebound. People with no money at all don’t have that opportunity to invest.
The stock market is $80T, cash is $20T. The former cannot "weather the storm" in the latter in aggregate, even though I'm sure there are individuals who successfully timed the market.
> At no point in the cycle is it better to be poor than rich, but the contractionary quadrant is the one where the gap shrinks.
Your anecdote about getting lucky doesn’t answer the question, sorry. Rich people have money to buy cheap assets when shit hits the fan, unemployed people do not.
There's not much incentive for institutional change when everyone's making money. Yeah, it'll hurt folks a great deal short term. But may be better for our kids long term.
That's what people said in 2008, instead, wealth was incinerated, the most vulnerable got hurt the most, while the already wealthy got bailed out and bought back assets at bargain bin prices. The wealth owning class would love a collapse.
Oddly enough, 2008 is exactly what inspired my comment and it's based off my exact experience back then. I couldn't afford investing in the market, I couldn't afford a home, and the interest on car loans were higher. All of that changed after the 2008 collapse. First I should mention that I was laid off and unemployed during some of it, so there was a huge negative impact for sure. What followed after though was amazing. I was able to invest in some $1-$5 stocks in the market that were previously not obtainable to me. This investment actually took me/my wife from being in debt living with my in-laws to completely out of debt and into our own place. Once I finally got employed again, I was able to save up enough just in time to buy a home at near the bottom with low interest. Car loans were also low interest. It was life changing and for the better in my situation.
The inflation being caused by all the "crowding-out" caused by AI overinvestment is currently incredibly harmful.
That really needs to unwind.
However much potential energy has been stored in it, is going to turn into kinetic energy in the subsequent correction/crash.
The best time for this to happen is as soon as possible, causing the least amount of damage.
I pretty much expect the damage to be immense, though, but that is going to be unavoidable at this point. However big the bubble is, it needs to pop.
In retrospect, we will find that there were truly idiotic decisions made in blowing up the bubble, and in the aftermath the wrong people will almost certainly get hurt, because that is just what we do these days.
IT Sovereignty is likely to be a growing trend after the past few years - including issues of Election interference not just by governments but by individual Tech CEOs. I think a lot of particularly Europeans have realized that putting all your eggs in American jurisprudence might -not- be the best way to go.
Cloudflare has a moat but it already smells like sewage.
Cloudflare is not competing with app builders like Lovable or Repl though. They are the underlying infrastructure for companies like that. This is why they will win in the AI era.
They're saying that it's trivial for Cloudflare to go up the stack (build Replit) while it's non-trivial for Replit to go down the stack (build Cloudflare).
Cloudflare is competing with Akamai, and a few other core delivery and WAF infrastructure companies and services provided by the big cloud providers. Not some content hosting outfit.
By this logic every single company on the planet would be a single-product company. The reality is that companies expand their offerings all the time – despite the cost, complexity, and risk of doing so.
A great time to expand your offerings is when there's an adjacent offering that's producing outsized returns for someone else, is easily replicable, and which you have an asymmetric advantage on.
I think everyone knows that when someone says "it's trivial" for a business to do something, it doesn't mean it's actually trivial. Doing anything of any value anywhere is non-trivial. It's shorthand for "doing Activity X is almost certainly achievable with an acceptable outlay of risk for Business Y."
> By this logic every single company on the planet would be a single-product company.
This is absolutely not true, and proof that you completely misunderstood what I said.
You took an upmarket B2B infrastructure company and compared it to a low-end B2B/B2C software company. That is a completely different type of business.
Cloudflare launches new products and offerings all the time - and all B2B within their market space. The difference is that if they did what you suggest (make a Lovable), it would put them in a position of competing with their own customers, and it would have a chilling effect on selling more infrastructure to other companies who would then fear competition as well. It's not that they can't do it. It's that it is not wise for them to do so, and it's a different type of business entirely.
As I understand it, very little of Replit and Lovable (especially) are actually B2C. A lot of it is internal teams at mid-sized companies, which are a huge and demonstrably very fruitful customer segment, and one that Cloudflare does in fact serve already.
Cloudflare has already started releasing products that point in this direction. Not sure how far they will (or should) go, but it's really not categorically ridiculous or impossible or nonsensical, as you're implying.
And the reason for Cloudflare to do it is because there are several (apparently) very juicy businesses to be built by slapping a chat UI and off the shelf coding agent on top of Cloudflare infra... Not saying they should do this, but the rationale is pretty self-evident.
The reason for Replit to try going down the stack is platform risk and compressing margins from competitors who also have the brilliant idea of slapping a text box and off the shelf coding agents onto Cloudflare.
Neither are trivial but I would wager cloudflare is the easiest to replace because it's the most mechanical and price sensitive.
It's like VPN service. Some other party comes along and offers it at 40% of the cost, you can do that.
Lovable is more nuanced. It takes more to move away, and there's just no reason to. The unit cost of Lovable buyeres is negligible aka less buyer sensitivity
LMAO you have it complete upside down, and if you think Lovable is a 'wrapper' around AI maybe you should try building a working app.
Lovable is a layer of abstraction, becoming as as complex as Shopify.
"It takes decades to build the kind of network cloudflare has."
It takes literally no time. Cloudflare's network could be emulated with proper planning and enough budget, and turned on instantly. 'Servers in 300 cities' is a medium sized challenge, but most importantly, it's a relatively clear path.
On the customer side, Cloudflare (in most cases) could be replaced as quickly as individuals replace their VPNs.
Shopify for example is ' just some siple HTML and checkout' - of course not - they are dominant and growing because there's a lot of complex interdependence, switching costs are much higher, their buyers are less technical and much less likely to want to change, and building a Shopify replacement involves hitting a lot of checkmarks.
Lovable started as a wrapper just as Shopify started with something simple, but they've quickly evolved.
AWS could snap their fingers and decide to knock out Cloudflare, it's a 'Commodity Plus' that came to the fore in a time when a specific need was there, and now people are turning to it out of convenience.
Of Lovable, Wix, Shopify, Figma, Stripe - Cloudflare is in the riskiest position because of the inherent commoditizability of their tech and the relatively low switching costs.
Shopify's core offer is a very long gnarly tail with any number of detail and moving parts, a massive eco system of 3rd party vendors, integations, and developer community.
Their service is sticky and switching costs are high.
It would literally cost more in Claude Code tokens to hack a prototype, than to sign up and host Shopify and get 100x the features. It's $0 to build the prototype and $1/mo to literally have a full product for the first 3 months.
At the same time, Claude Code can build you a full variation of Cloudflare's core product, and with less than $1M you can scale across 300 regions if you really put your mind to it.
During the dot-com boom, Cisco was the underlying infrastructure of the Internet. A lot of investors that believed themselves to be savvy bought Cisco stock. Their reasoning was that individual dot-com startups may go under, but the Internet itself was here to stay. In practice, the Cisco stock price peaked in March 2000, and anyone who invested at the peak would not have broken even (in nominal dollars, not real dollars) until December 2025.
The investors were right that the Internet would be important, and that Cisco would continue to be an important manufacturer of networking infrastructure. This just fundamentally did not matter because after the dot-com bubble popped, the market decided that Cisco simply wasn't worth the price it was selling at during the height of the bubble.
Our designer uses some some of open-source Lovable. I don't remember the name, it's just pointed to our liteLLM gateway. A dev helped him with the setup, so there's still a bit of value in lovable's "it just works", but it's so much cheaper someone is bound to eat their lunch.
In an April 2002 interview with Bloomberg Businessweek, Scott McNealy, then CEO of Sun Microsystems, reflected on investors paying 10x revenue for Sun stock at its dot-com peak ($64/share in 2000):
At 10 times revenues, to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends. That assumes I can get that by my shareholders. That assumes I have zero cost of goods sold, which is very hard for a computer company. That assumes zero expenses, which is really hard with 39,000 employees. That assumes I pay no taxes, which is very hard. And that assumes you pay no taxes on your dividends, which is kind of illegal. And that assumes with zero R&D for the next 10 years, I can maintain the current revenue run rate. Now, having done that, would any of you like to buy my stock at $64? Do you realize how ridiculous those basic assumptions are? You don't need any transparency. You don't need any footnotes. What were you thinking?
Now this analysis is imperfect... but it's good enough.
That assumes 100% of Internet can or would be willing to adopt Cloudflare, which has no real basis.
The EU is finally waking up to the fact that putting all its eggs in the U.S. basket is a bad thing. China is enormous, and they're not going to go all in on Cloudflare either for a myriad of reasons. And a bunch of folks just don't like nor want to do business with the company and its CEO.
What does that even mean? Revenue is measured in dollars not in % of internet traffic. They could "own" 100% of internet traffic and still have growing revenue.
Disclaimer that I'm a long-term Cloudflare holder, and that I've also sold NET in the past year.
I empathize with this view, however it is very hard to value businesses that have extremely high margins, and who grow revenue very quickly on existing products while they're adding additional products.
I don't think NET is a steal at 39x revenue. But at no point in the last 5 years has NET been available to purchase for <15x revenue [1].
During that time, it has had Gross Margins in the exceptional range of 70%-75%, and YoY quarterly revenue growth in the similarly exceptional range of 25-40%, which has increased in our post-AI agent software economy.
If you're a Ben Graham acolyte, you would have sat out of NET since it's IPO. However, if you're a Charlie Munger/Buffet admirer who understands the pivotal role NET plays in the software ecosystem (i.e. it's inside your "circle of competence"), you'd do well to consider the Mungerism:
"Forget what you know about buying fair businesses at wonderful prices. Instead, buy wonderful businesses at fair prices."
That still leaves as an exercise to the reader what a "fair price" is for NET, but (to me, at least) it's clearly a wonderful business.
I’m saying that 1/5th of global internet traffic runs through their servers. Period. Thats significant and if anything shows that there’s potentially a lot more opportunity.
Ever looked at MSFT or AAPL’s dividends? It’s a token “here’s your fucking dividend, now go away” percentage. (I’m sure it’s actually to placate funds that require a dividend, but I repeat myself.)
As a retired ‘softie, my spouse still holds a fair bit of MSFT, so I just looked: $0.98…on a $540.00 stock. Yeah, we won’t be living off that in our old age.
Why would you want a dividend? It’s mathematically better for a business to do share buybacks and allow investors to choose when to sell shares so they can minimize tax liability.
Australia has one of, if not the highest rates of CGT in the world.
ASX dividend stocks are popular with retirees in particular because the dividends are fully franked, meaning you get a credit for the tax already paid on the dividends by the company.
If you’re in a low enough tax bracket, this can turn into a tax refund if the tax the company paid exceeds your own personal tax liability.
Given how much people have started actively looking for alternatives, and how much spray-and-pray Cloudflare's offerings now seem with so many launches every few weeks, steal is a bit too optimistic.
Beware any simple narrative for the decline of a massively complex market.
Did the election matter at all for this trade? This trade has been in the works for the last 2-3 years. This correction is required to bring down inflation - and we're already seeing T-notes dropping lower which will decrease mortgage costs in the near term and allow for more liquidity for consumers for home equity loans.
I don't understand why folks continue to click-through and "buy" these overly simplistic narratives from news outlets.
The thing you're missing is that the political expenditure of AI in the US is massively tied with the GOP and the far right. There's a reason why people like Elon Musk have been absolutely freaking out over the result.
This result means that unlimited data center expansion may not be on the table, and bailouts certainly aren't either. So with the leaders crashing out the uncertainty of AI expansion is now being properly priced into the stock.
I'm not missing it, I assure you. The politics was just not a huge factor in any larger sense in my opinion. The GOP tie-up was ALSO a narrative. (Though obviously folks thought the President would help with de-regulation, etc)
In the end, States and their utility regulators have to be on board with large data center deployments. Some GOP-led states were often friendly to builders, but their citizens realized what was happening and the backlash was going to happen anyway because of electricity prices. It's just one more nail in the coffin from a cost-of-living perspective, and the utilities hadn't yet passed the costs on to their customers (but it will pop up soon).
States that were led by Democrats were already pushing back for their own regulatory regimes for energy cost recovery, etc. Data centers were wildly unpopular at a STATE level, so it didn't really matter what was going to happen at the federal level.
Whenever I see daily stock results I try to reframe it in a longer term. Interestingly, Nasdaq is only up ~4% YoY.
This is surprising to me, given how much acceleration there has been in AI capabilities since the same time last year. I guess costs and risks are accelerating just as fast.
Stocks have been somewhat flat this year. Some big gains, but a lot of stagnation around the time the Iran war began. I got my hopes up yesterday, but the crash today puts us back to ~May depending on which index you're looking at.
The problem is that the economy is kinda flatlining and only the AI bubble (that is also hoovering up other, seemingly unrelated parts of the tech sector) is what makes this number as big as it is.
Democrats are not going to try and stop AI or rein in AI. There is too much money at stake. They'll pass some regulations, sure, some data center rollouts might be blocked or modified, but they aren't going to stop the train.
Many safe and good technologies were successfully sabotaged for populist and stupid political reasons, e.g. in nuclear power. Or less spectacular, think of the welfare loss due to the NIMBY-fied housing market.
Safe and good are doing a lot of work there. Nuclear power is only safe and good if the right safety measures are put into place for it to operate safely. AI, or the data center build-out might warrant similar consideration potentially?
Happily I think the power market in the US will do most of the regulating for the industry. AI is not bringing costs down for most people at the moment, or perceived to be helping people in a big way. Until there is an understanding of how it will work within society or a bit of trust built up with an industry that has mostly said "shut up and take it" I think it's reasonable to expect some conversation without jumping to "these folks are all populist stupid politicians". (In my opinion, populist, stupid politicians could be used to describe both parties so what is really changing?)
I was going to say that Democrats have no power to uphold laws, that's up to the executive and the national executive hasn't changed.
On the other hand a number of governors have changed (Texas in particular just went to a Democrat, and there are SpaceX interests there). They by definition can't prosecute on a federal level but they do hold some enforcement power.
The legislative branch has the power to hold the executive branch accountable for not enforcing the laws should they so choose. That's why impeachment and conviction exist.
Well, don't forget this is also about national security. China/state of choice AI going after your infrastructure & national secrets is going to take AI to defend against.
There's zero chance AI datacentres are going away, they are unfortunately a national security necessity now.
The financial aspect/discourse is for the proletariat who prefer to pretend politics and risk of military conflict do not exist.
They won't stop it because China won't stop. Politicians have to avoid any appearance of xenophobia though (and to be certain many are personally not xenophobic, just saying optics matter to them), so they talk about jobs and wealth inequality as euphemism for ground truth; we aren't secure putting all our eggs in the China basket.
Arguably screwed up there already letting chips and robotics flow offshore while we focused on increasingly useless education in developer syntax and semantics.
They will cool the AI investment well aware technology improves and we probably don't really need all our investment eggs in that one basket then
They don't have to. This particular train explodes on its own if it drop below 100 mph.
Even tepid regulation will destroy the "our TIA is the entire observable universe due to the potential of self replicating AI probes so we totally deserve 10 trillion dollars in investor capital" trade
I've seen this movie. I think it was called "The Bus That Couldn't Slow Down". It had Keanu Reaves and Sandra Bullock. IIRC it didn't end well for the bus.
They aren't going to stop AI, but they are not going to let the owners of the superintelligence inference engines become god emperors of the planet, and in that case why not just get 4.5% on some bonds?
They're not even going to do that. Trump isn't going to sign that. When was the last time Congress overrode a veto or played hardball (well the Democratic side anyway)? They have no fight in them.
Stopping DC rollout effectively works as calling the bluff of AI.
AI is a hyped up nothing burger? AI demand retreats, lots of vacant compute, revenues sink (and the DC hold ends up being a (small) blessing in disguise for the labs).
AI lives up to it's hype? AI demand surges, capacity is severely constrained, margins explode as inference costs soar (and the DC block ends up being counter productive knee-jerk behavior)
The only way this plays out in anti-AI favor is if people stop using AI and demand goes down. Even if it flatlines, labs can rebalance towards inference and start harvesting those huge margins.
I'm starting to use it quite heavily and extremely productively for non R&D and non SWE work. It has transformed one aspect of my business, and that won't change -- this type of usage is highly productive, and only likely to grow.
The latest crop of models are extremely capable at classic (physics based) engineering tasks. Their incredible math ability as well as agent use allows them to plow through engineering problems, and they do it with knowledge that I have only heard grey beards talk about. Humans run through designs doing 1st order approximations and intuitive assumptions, but the models can just breeze through 4th order approximations on everything.
They used to not be able to read schematics, but now they can, as well as generate and simulate. They also have made incredible inroads into CAD and reading mechanical diagrams.
Right now I can say with very high confidence that there isn't a gap in ability so much as a gap in awareness of their abilities. On top of that the next round of models has apparently made strides in all these areas too.
I don't agree with your description of how the ban is going to be counterproductive.
There won't be massive inference margins. The DCs will get built, in some other country. And in all likelihood they will continue to be built elsewhere even after the block is lifted.
> AI is not going anywhere and will only grow more and more.
I agree, but I think it's easy to conflate AI (a technology) with current AI companies (who's value is based on a bet about how that technology will develop).
I'm not predicting anything, but its perfectly plausible for Anthropic and OpenAI to suffer massive corrections if proprietary frontier models don't lead to a "a few winners take all" market.
In the same way, the internet grew a lot since the early 2000s, but pets.com did not.
Also don't be discouraged by these anti-datacenter clowns. Most normal people don't understand where their doomscroll gets pumped out of every day and thinks the internet comes from their phone. These are the same people who are anti-powerplant and think electricity comes from a wall outlet.
Doom scrolling is nothing good for society and it’s with us only because extremely well funded organisations make money from it, and has weaponised it against the general population.
I am a high voltage tester and know electricity very well.
What kind of dogshit take is this? AI datacenters are a very different breed of facility in terms of their effect on their surrounding communities. Colossus is literally poisoning South Memphis.
> This is clearly voter sentiment about AI in general and big tech in general.
There is growing public sentiment about AI & Tech, but per exit polls that wasn't a Top 3 reason swing voters went blue (Economy, Wars, Corruption). Today's market correction is being driven by the majority in both houses of congress flipping blue. Trump and most Republican legislators were seen as broadly supporting AI / Tech and most Democrats are seen as being less supportive.
AI / Tech-friendly legislation is now likely to face a slower, more difficult path through committee and passage. IMHO, the actual impact of the political change is nowhere near 5% of company value, but outsized swings tend to happen on any adverse news when the market is already fearful they've bought into a high-priced bubble far in advance of actual earnings.
My hope is that this moment is a decisive but constrained tempering of market enthusiasm. But that sounds too reasonable, so…probably next week it’ll all be back to crazy town.
I agree. The best case would be if this 5% drop in AI-centric tech isn't a 'dip', but the beginning of a gradual unwinding of the AI bubble. I'd take it as a positive sign if 3 months from now AI-centric stocks had slipped another 5% but other sectors where slightly higher than today.
I'm actually fairly bullish on AI in the long-term. The problem is valuations have been hyper-inflated by too much speculative capital way too fast. I think today it's a bit like the Internet Bubble in 1999. The Dot Com crash needed to happen to reset valuations. But after that painful correction, by 2006 Internet-centric valuations were climbing again, except this time they were supported by actual earnings. I'm hoping AI can avoid a crash and this is the start of a more gradual correction.
My definition for an Agent is "a marketing term for long running LLM Chat sessions that has increased access to use your computer to complete tasks it cannot otherwise do by itself".
That's my attempt at a universal definition for technical and non-technical users. How does it compare?
Well, my reference for a commonly used incorrect definition is “anything that involves using an LLM” so yours is better than that.
But the elements that I think are necessary are 1) goal-oriented processing, 2) ability to use tools, 3) ability to iterate and self-correct. Along with that typically comes 4) self-talk or other hidden context during execution and 5) support for defined personas beyond system or user prompts.
That’s off the top of my head, probably not complete/correct, and definitely meant to be more technical than a non-tech person would need.
Bedsides coding agents, which only work extremely well if you are a developers or you train a person with the right inclination I have plenty running that work quite well, they do data extraction and data entry from emails and documents.
I run them with Chinese and other open models on a bunch of H100, they cover tens of thousands of emails per month and as many documents.
It's not only about the people who were elected, but 44/45 ballot initiatives specifically aimed at restricting or banning data center construction passed.
The Trump administration was expected to invest massively into AI, and money invested into AI is effectively free money for AI company shareholders, that free money is priced in the stocks. With the change in direction, investments are expected to be less, which means less free money, and therefore less valuable stocks.
Well, yes, but it is even more mechanical than this.
Here is the simplified version:
Let's say I have a company worth 1 billion dollars, publicly traded, 10 million stocks worth $100 each. I get promised $100M from the state (aka taxpayers), shareholders estimate there is about 50% chance I will actually get the money: $100M * 50% / 10M = $5. It means the stocks are now worth $5 more, so $105.
Now the government has changed, and gone are my chances to get that money, so the sock is back to $100.
And is actually the only thing I’m sad about in this whole thing. Ecstatic about the blue wave sad that they’re going to restrict AI progress. It’s the future, and it would be a horrible move to curtail it.
AI progress doesn't require doing it full ahead right now. Beyond the risks of getting out of control, you get bubbles if hype goes beyond economic reality. AI should be done more slowly with regulation and slower build out. Better to do it slowly than wreck the world with unexamined consequences.
Great opportunity to buy the dip in a couple of weeks. If anyone thinks they can slow down or stop AI adoption by banning data centers, good luck with that... The Chinese built an AI empire with much less compute that exists today in the US.
Also, people don't have issue with data centers. They have issues with AI cause they think it's gonna make them unemployed, which is a real worry I can relate to. I just don't think that banning data centers or AI progress in general would help anyone keeping their job.
Personally, I think that the idea that market always recovers is so widespread right now that it'll take some serious financial hardships for people to start selling instead of buying the dip.
With the housing bubble, it was because people owned multiple homes then struggled to pay two mortgages as economic conditions degraded, and banks were willing to absorb losses if the homes were sold below the cost of the mortgage.
What is encouraging people to route from the stock market? Bond yields aren't high enough yet to encourage people to retreat from stocks, and there's no other alternative investments. Plus, so much of the S&P is held in 401ks. And I agree, economic hardship is not widespread enough for people with significant investment holdings to start selling them off to live. Those who are struggling are those who used to be struggling too (thus, no extra money for investments).
There are also a lot of set it and forget it index investors that won't buy a dip, but won't sell either (unless they experience those serious financial hardships). That's the group I'm in.
As uncomfortable as this may be for the tech industry, I have been hoping for AI related stock prices to come back to reality. A market correction is a healthy thing so long as it doesn't turn into a collapse. IMO, some AI companies have ludicrous valuations considering their astronomical debt. A reality check may be healthy long term.
IMO OpenAI jumped the gun and rushed an immature product to be there first, no matter the costs. The other players saw it as a threat, and started running with the same carelessness. It has been spiraling since then, and while the product keeps improving, their business model hasn't demonstrated sustainability.
Their CEOs know it, and they must have their end game. I'd hate to think they'll just keep running until a hard crash. Hopefully small corrections will slow this down in a more controlled manner.
You forgot the bit where China undermines the whole business model by releasing open weight models.
And I wholeheartedly thank China for doing this. The open weight models are fantastic, and make power consolidation of AI monopolies less likely. Open weights bring real competition to the space and help commoditize inference.
That was a disruption if I've ever seen one, but they are also paying the price. Every LLM maker right now is losing money.
much to our delight as peasants who were being lined up at the token furnaces of the big american labs.
They won't mind a hard crash, as long as they get richer on the way.
Good hardware companies are now trading at far under ten P/E, hardly ludicrous valuations. Sure there are some poor companies out there, but Nvidia and Micron are not among them. SMCI just beat substantially. Buying opportunity when things settle down.
Nvidia has not been a buying opportunity for a while now. Training new models is the moat it possessed. The models are as good as they can be and have diminished returns beyond this. TPUs by google and CPUs by AMD are eating into their inference margins.
Micron is a different beast though.
All the hardware makers are supply constrained, p/e reasonable, so nvidia has some more room to run. Though not be unicorn growth from here.
Debt to Equity ratio thresholds are a common metric for lenders to call in loans.
Market is still up from 3 months ago
Good luck to the banks.
a reality check for a crazy person can result in a mental breakdown
Yeah I'm not sure how Anthropic is going to IPO at this rate...
It had a big user base boost during the Cancel ChatGPT movement, but still not a viable business model, not at their current prices.
> Yeah I'm not sure how Anthropic is going to IPO at this rate...
There are plenty of suckers willing to gamble on a failing business (massive and rapidly growing debts, assets that can easily be cloned, no realistic path to profitability), so I don't think an IPO is off the table.
With more visits to the Vatican.
i died reading this lol
I hope they go low enough I'm willing to buy some shares of ANT
I don't know - I think open source will outcompete them. What's their moat?
The moat is GPUs/TPUs, ram and compute. When you need $200k of compute to run the frontier models, you are likely to be renting access to generate your tokens.
Current RAM is no moat. Upcoming Compute-in-RAM is. And nobody has that built up at scale yet. So, while a moat might become possible, we only have yesteryear's hardware as "moat".
Can you explain what compute in RAM means?
I thought HBMs are technically the fastest possible thing for inference.
The academic term is now "Compute-in-Memory". That's MOSTLY what has suddenly been hiring SRAM and DRAM designers across the industry this year. It is part of the fast evolving field of doing inference efficiently. This has been all the rage in several Electronics conferences this year.
Please don't rely on my [inferior] explanation and search for yourself (start here for insance: https://www.google.com/search?q=what+is+compute+in+memory+an...). Asking about a new field is much better done with AI these days, because you can get it to explain just the interests and questions you personally have, rather than a generic and boring explanation.
In short, better than HBM is not having to move data in and out of memory at all or at least not as much. The compute part in inference is much simpler than the data hauling part.
Couple of examples: Check out Axelera's Europa board (upcoming) or their Metis board (existing). Or check out Mythic.ai which FINALLY managed to raise funds in Nov / Dec 2025. Ironically, her PhD thesis had been gathering dust before that.
That is not a moat. (and it is not theirs either) It just means that their models are highly inefficient that it requires a lot of compute to run them.
Open weight models such as Deepseek, GLM are just as good as the frontier closed models and already we have bank such as Revolut migrating off of frontier closed models and self-hosting. [0]
[0] https://x.com/guilleflorvs/status/2117379073785696395
Probably close to $1M+ if frontier models are truly 5T and above params.
A collapse would be nice too.
Not sure if you're joking, but I go back and forth between this thought. Finance is not my forte, far from it for sure, but there's a part of me that hopes a collapse can at least open the door for future opportunity. A "reset" if you will. The younger generations are up against a massive wall when it comes to acquiring wealth (or even a reasonable means of living at this point), it would be nice to give them an equal chance. I know this is very likely a naive take, but it's the only hopeful aspect of a collapse I can accept.
Unfortunately this is not what happens. The Big Short is a good (and fun!) primer about collapses:
https://www.youtube.com/watch?v=0k5aVLi_yhM
When the market crashes, a lot of people lose jobs, and a lot of people die. Everyone suffers but those with money suffer the least, by far.
Not quite. Poor people suffer more in real terms, but rich people suffer more in nominal (dollar) terms. Nominal inequality gets reset.
Don't let the term "nominal" undersell this: it's nominal inequality that locks working people out of assets and imposes upon them the need to service the (otherwise) exponentially growing profit expectations of their economic complements. At no point in the cycle is it better to be poor than rich, but the contractionary quadrant is the one where the gap shrinks.
Not sure it works like that: https://pmc.ncbi.nlm.nih.gov/articles/PMC9122750/
Inequality declined temporarily during the recession, but reached new highs during the recovery.
2008 decreased inequality for a very short time, but inequality increased overall. The rich have the available assets to profit from instability and are able to ride out unemployment, for the poor a crash and losing their job can be existential.
> Nominal inequality gets reset.
Why would it? The wealthy people who haven’t allocated their capital to those assets that are depressed from a recession can reallocate to the depressed assets and end up even wealthier on the rebound. People with no money at all don’t have that opportunity to invest.
The stock market is $80T, cash is $20T. The former cannot "weather the storm" in the latter in aggregate, even though I'm sure there are individuals who successfully timed the market.
> At no point in the cycle is it better to be poor than rich, but the contractionary quadrant is the one where the gap shrinks.
How would this give young people a chance? If they have cash to buy up assets cheaper then so do older rich people.
Please see my comment here, I think it answers your question: https://news.ycombinator.com/item?id=50385037
It really doesn't. You could've bought any $1-5 stocks in that time. You happened to buy the ones that made you money.
Your anecdote about getting lucky doesn’t answer the question, sorry. Rich people have money to buy cheap assets when shit hits the fan, unemployed people do not.
There's not much incentive for institutional change when everyone's making money. Yeah, it'll hurt folks a great deal short term. But may be better for our kids long term.
That's what people said in 2008, instead, wealth was incinerated, the most vulnerable got hurt the most, while the already wealthy got bailed out and bought back assets at bargain bin prices. The wealth owning class would love a collapse.
Oddly enough, 2008 is exactly what inspired my comment and it's based off my exact experience back then. I couldn't afford investing in the market, I couldn't afford a home, and the interest on car loans were higher. All of that changed after the 2008 collapse. First I should mention that I was laid off and unemployed during some of it, so there was a huge negative impact for sure. What followed after though was amazing. I was able to invest in some $1-$5 stocks in the market that were previously not obtainable to me. This investment actually took me/my wife from being in debt living with my in-laws to completely out of debt and into our own place. Once I finally got employed again, I was able to save up enough just in time to buy a home at near the bottom with low interest. Car loans were also low interest. It was life changing and for the better in my situation.
The inflation being caused by all the "crowding-out" caused by AI overinvestment is currently incredibly harmful.
That really needs to unwind.
However much potential energy has been stored in it, is going to turn into kinetic energy in the subsequent correction/crash.
The best time for this to happen is as soon as possible, causing the least amount of damage.
I pretty much expect the damage to be immense, though, but that is going to be unavoidable at this point. However big the bubble is, it needs to pop.
In retrospect, we will find that there were truly idiotic decisions made in blowing up the bubble, and in the aftermath the wrong people will almost certainly get hurt, because that is just what we do these days.
You think a large scale market collapse would be good? Large scale job loss and bankruptcy?
At the very least we could buy hardware again at rational prices.
Cloudflare is a steal at $240 IMO.
Regardless of what happens to the AI models, you will still need protection a decade from now.
Not to mention the most comprehensive platform to use with AI, they will inevitably eat Lovable / Repl's lunch.
It looks like Cloudflare generates $633mm operating cash flow a year, up 5% for TTM from 2025 [1]. I’m surprised that’s supporting a $95bn market cap.
[1] https://finance.yahoo.com/quote/NET/cash-flow/
How does that compare to other tech companies?
As an example, Palantir has a similar cash flow / market cap ratio. I didn't check many others.
now that they have most of the internet corralled I don't see why they won't start charging for bandwidth
IT Sovereignty is likely to be a growing trend after the past few years - including issues of Election interference not just by governments but by individual Tech CEOs. I think a lot of particularly Europeans have realized that putting all your eggs in American jurisprudence might -not- be the best way to go.
Cloudflare has a moat but it already smells like sewage.
I'd be very surprised if the internet isn't well on its way to being balkanised by 2036.
Cloudflare is not competing with app builders like Lovable or Repl though. They are the underlying infrastructure for companies like that. This is why they will win in the AI era.
Lovable is hosted on Cloudflare: https://x.com/CloudflareDev/status/2113651157721600492
They're saying that it's trivial for Cloudflare to go up the stack (build Replit) while it's non-trivial for Replit to go down the stack (build Cloudflare).
Cloudflare is competing with Akamai, and a few other core delivery and WAF infrastructure companies and services provided by the big cloud providers. Not some content hosting outfit.
People are over estimating Cloudflare's moat.
It's trivial for those providing internet links to go up the stack and cloud data centers but they typically don't.
Staying in your lane is often a better choice.
Maybe expanding through acquisition is the best of both worlds. So replit might be bought.
I'm not expressing an opinion on whether they should or shouldn't.
It's not trivial to change your entire business.
Cloudflare is B2B focusing on enterprise sales and underlying cloud infrastructure. These app builder companies are basically B2C and lower end B2B.
By this logic every single company on the planet would be a single-product company. The reality is that companies expand their offerings all the time – despite the cost, complexity, and risk of doing so.
A great time to expand your offerings is when there's an adjacent offering that's producing outsized returns for someone else, is easily replicable, and which you have an asymmetric advantage on.
I think everyone knows that when someone says "it's trivial" for a business to do something, it doesn't mean it's actually trivial. Doing anything of any value anywhere is non-trivial. It's shorthand for "doing Activity X is almost certainly achievable with an acceptable outlay of risk for Business Y."
> By this logic every single company on the planet would be a single-product company.
This is absolutely not true, and proof that you completely misunderstood what I said.
You took an upmarket B2B infrastructure company and compared it to a low-end B2B/B2C software company. That is a completely different type of business.
Cloudflare launches new products and offerings all the time - and all B2B within their market space. The difference is that if they did what you suggest (make a Lovable), it would put them in a position of competing with their own customers, and it would have a chilling effect on selling more infrastructure to other companies who would then fear competition as well. It's not that they can't do it. It's that it is not wise for them to do so, and it's a different type of business entirely.
As I understand it, very little of Replit and Lovable (especially) are actually B2C. A lot of it is internal teams at mid-sized companies, which are a huge and demonstrably very fruitful customer segment, and one that Cloudflare does in fact serve already.
Cloudflare has already started releasing products that point in this direction. Not sure how far they will (or should) go, but it's really not categorically ridiculous or impossible or nonsensical, as you're implying.
https://blog.cloudflare.com/cloudflare-os/
Why would either of them do either of those? They seem to be different businesses.
Network effects make it harder for Cloudflare to build Replit than it might seem, too.
What network effects?
And the reason for Cloudflare to do it is because there are several (apparently) very juicy businesses to be built by slapping a chat UI and off the shelf coding agent on top of Cloudflare infra... Not saying they should do this, but the rationale is pretty self-evident.
The reason for Replit to try going down the stack is platform risk and compressing margins from competitors who also have the brilliant idea of slapping a text box and off the shelf coding agents onto Cloudflare.
Neither are trivial but I would wager cloudflare is the easiest to replace because it's the most mechanical and price sensitive.
It's like VPN service. Some other party comes along and offers it at 40% of the cost, you can do that.
Lovable is more nuanced. It takes more to move away, and there's just no reason to. The unit cost of Lovable buyeres is negligible aka less buyer sensitivity
Lmao, they’re just wrappers around ChatGPT. You could replace them in a day.
It takes decades to build the kind of network cloudflare has.
“Please mr bezos, can I connect my new trivial network to yours?” “0.08/GB :)”
LMAO you have it complete upside down, and if you think Lovable is a 'wrapper' around AI maybe you should try building a working app.
Lovable is a layer of abstraction, becoming as as complex as Shopify.
"It takes decades to build the kind of network cloudflare has."
It takes literally no time. Cloudflare's network could be emulated with proper planning and enough budget, and turned on instantly. 'Servers in 300 cities' is a medium sized challenge, but most importantly, it's a relatively clear path.
On the customer side, Cloudflare (in most cases) could be replaced as quickly as individuals replace their VPNs.
Shopify for example is ' just some siple HTML and checkout' - of course not - they are dominant and growing because there's a lot of complex interdependence, switching costs are much higher, their buyers are less technical and much less likely to want to change, and building a Shopify replacement involves hitting a lot of checkmarks.
Lovable started as a wrapper just as Shopify started with something simple, but they've quickly evolved.
AWS could snap their fingers and decide to knock out Cloudflare, it's a 'Commodity Plus' that came to the fore in a time when a specific need was there, and now people are turning to it out of convenience.
Of Lovable, Wix, Shopify, Figma, Stripe - Cloudflare is in the riskiest position because of the inherent commoditizability of their tech and the relatively low switching costs.
The SaaS component of Shopify is also trivial now. That's not where Shopify's moat comes from either. You are very confused.
> Cloudflare's network could be emulated with proper planning and enough budget
lmfao
No.
Shopify's core offer is a very long gnarly tail with any number of detail and moving parts, a massive eco system of 3rd party vendors, integations, and developer community.
Their service is sticky and switching costs are high.
It would literally cost more in Claude Code tokens to hack a prototype, than to sign up and host Shopify and get 100x the features. It's $0 to build the prototype and $1/mo to literally have a full product for the first 3 months.
At the same time, Claude Code can build you a full variation of Cloudflare's core product, and with less than $1M you can scale across 300 regions if you really put your mind to it.
Whats Shopify's moat then?
During the dot-com boom, Cisco was the underlying infrastructure of the Internet. A lot of investors that believed themselves to be savvy bought Cisco stock. Their reasoning was that individual dot-com startups may go under, but the Internet itself was here to stay. In practice, the Cisco stock price peaked in March 2000, and anyone who invested at the peak would not have broken even (in nominal dollars, not real dollars) until December 2025.
The investors were right that the Internet would be important, and that Cisco would continue to be an important manufacturer of networking infrastructure. This just fundamentally did not matter because after the dot-com bubble popped, the market decided that Cisco simply wasn't worth the price it was selling at during the height of the bubble.
Lovable is one ChatGPT/Claude update away from extinction, IMO.
Our designer uses some some of open-source Lovable. I don't remember the name, it's just pointed to our liteLLM gateway. A dev helped him with the setup, so there's still a bit of value in lovable's "it just works", but it's so much cheaper someone is bound to eat their lunch.
In an April 2002 interview with Bloomberg Businessweek, Scott McNealy, then CEO of Sun Microsystems, reflected on investors paying 10x revenue for Sun stock at its dot-com peak ($64/share in 2000):
At 10 times revenues, to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends. That assumes I can get that by my shareholders. That assumes I have zero cost of goods sold, which is very hard for a computer company. That assumes zero expenses, which is really hard with 39,000 employees. That assumes I pay no taxes, which is very hard. And that assumes you pay no taxes on your dividends, which is kind of illegal. And that assumes with zero R&D for the next 10 years, I can maintain the current revenue run rate. Now, having done that, would any of you like to buy my stock at $64? Do you realize how ridiculous those basic assumptions are? You don't need any transparency. You don't need any footnotes. What were you thinking?
Now this analysis is imperfect... but it's good enough.
For reference Cloudflare trades at 39x revenue
That analysis completely neglects growth.
Obviously you wouldn't pay 10x revenue if you expect revenue to be flat for the next 10 years.
No wonder that company no longer exists with a CEO that doesn't get that.
Cloudflare is now a mature company, they don't have a huge amount of room for growth, and no, they will never sit in front of the entire internet.
According to these figures, their growth is levelling off.
https://news.ycombinator.com/item?id=50382740
And nowhere near enough to sustain that valuation, particularly in a market which is near the end of a bull run.
you're right on growth being excluded, which is why I said the analysis is imperfect.
Cloudflare's growth is capped to 100% of internet traffic. it's already at 20%.
That assumes 100% of Internet can or would be willing to adopt Cloudflare, which has no real basis.
The EU is finally waking up to the fact that putting all its eggs in the U.S. basket is a bad thing. China is enormous, and they're not going to go all in on Cloudflare either for a myriad of reasons. And a bunch of folks just don't like nor want to do business with the company and its CEO.
Internet traffic is projected to grow 25% YoY over the next decade.
Most of that is bot traffic.
Sounds like a new business expansion opportunity for Cloudflare.
What does that even mean? Revenue is measured in dollars not in % of internet traffic. They could "own" 100% of internet traffic and still have growing revenue.
Disclaimer that I'm a long-term Cloudflare holder, and that I've also sold NET in the past year.
I empathize with this view, however it is very hard to value businesses that have extremely high margins, and who grow revenue very quickly on existing products while they're adding additional products.
I don't think NET is a steal at 39x revenue. But at no point in the last 5 years has NET been available to purchase for <15x revenue [1].
During that time, it has had Gross Margins in the exceptional range of 70%-75%, and YoY quarterly revenue growth in the similarly exceptional range of 25-40%, which has increased in our post-AI agent software economy.
If you're a Ben Graham acolyte, you would have sat out of NET since it's IPO. However, if you're a Charlie Munger/Buffet admirer who understands the pivotal role NET plays in the software ecosystem (i.e. it's inside your "circle of competence"), you'd do well to consider the Mungerism:
"Forget what you know about buying fair businesses at wonderful prices. Instead, buy wonderful businesses at fair prices."
That still leaves as an exercise to the reader what a "fair price" is for NET, but (to me, at least) it's clearly a wonderful business.
[1]: https://www.macrotrends.net/stocks/charts/NET/cloudflare/pri...
"In the short term the market is a popularity contest; in the long term it is a weighing machine." - Warren Buffett
Here's the valuation though. Percentage of total web traffic going through Cloudflare.
- 2016 ~3.9%
- 2017 ~5.3%
- 2018 ~6.6%
- 2019 ~7.6%
- 2020 ~11.2%
- 2021 ~15.4%
- 2022 ~19.4%
- 2023 ~18.0%
- 2024 ~18.9%
- 2025 ~19.2%
- 2026 (early) ~20.9% (with later snapshots higher, up to the mid-to-high 20s)
Combine that with consistently excellent execution and popularity, you get the valuation.
The quote comes from Ben Graham and it goes "In the short run, the market is a voting machine, but in the long run, it is a weighing machine"
You're confusing market share with valuation. Also, look into business cycle theory from marathon asset management.
The four most dangerous words in finance "this time is different"
I’m not confusing market share with valuation.
I’m saying that 1/5th of global internet traffic runs through their servers. Period. Thats significant and if anything shows that there’s potentially a lot more opportunity.
Not "I'm smarter than that"?
All it takes is for that % of internet number to go down or stagnate and the narrative to change and that valuation will quickly snap back to reality.
Sure. As would be expected.
A tech stock paying dividends? It'll never happen.
hehe
https://giftarticle.ft.com/giftarticle/actions/redeem/8605e6...
Uh.. The big guys do.
Yeah, but they just do that for inclusion into some investment products that require a dividend.
Share buybacks are always better for the company and investors.
This isn't 2010. Almost all big tech companies are paying dividends now.
Ever looked at MSFT or AAPL’s dividends? It’s a token “here’s your fucking dividend, now go away” percentage. (I’m sure it’s actually to placate funds that require a dividend, but I repeat myself.)
As a retired ‘softie, my spouse still holds a fair bit of MSFT, so I just looked: $0.98…on a $540.00 stock. Yeah, we won’t be living off that in our old age.
Yeah big dividends are for companies that can't do anything better than the money.
Why would you want a dividend? It’s mathematically better for a business to do share buybacks and allow investors to choose when to sell shares so they can minimize tax liability.
That depends a whole lot on your tax jurisdiction, your other income and your investment strategy.
In both scenarios, the shareholder can get the same amount of cash if they want it. Except with a dividend, they are forced to pay tax now.
I suppose there could exist a jurisdiction where dividends are taxed less than capital gains. I would be curious to know if there are any.
Australia has one of, if not the highest rates of CGT in the world.
ASX dividend stocks are popular with retirees in particular because the dividends are fully franked, meaning you get a credit for the tax already paid on the dividends by the company.
If you’re in a low enough tax bracket, this can turn into a tax refund if the tax the company paid exceeds your own personal tax liability.
Interesting, thanks for the info.
No one is saying they want a dividend, I'm explaining how small the dividends of big tech companies are.
They have good potential for cloud services too. Competing with AWS is much more viable when you have data centres everywhere and a huge network.
Im still crying about having sold it at 20
I’m still crying about taking a Dropbox offer over Cloudflare’s pre-IPO in 2016.
Yeah hindsight is incredibly frustrating.
Also owned a bunch of asml shares a decade or so ago. But it didn’t really move so got bored and sold. Good picks bad patience.
Given how much people have started actively looking for alternatives, and how much spray-and-pray Cloudflare's offerings now seem with so many launches every few weeks, steal is a bit too optimistic.
Yeah we are actively building our own alternatives due to excessive CF bills
You selling?
Cloudflare is in no position to eat lovable or Repl.
And frankly, they should basically be a commodity.
If anything, AWS or MS or other parties make it about as profitable as hosting web pages.
Dude I made 2k off the earnings call tonight thanks to your comment. Thank you!
Beware any simple narrative for the decline of a massively complex market.
Did the election matter at all for this trade? This trade has been in the works for the last 2-3 years. This correction is required to bring down inflation - and we're already seeing T-notes dropping lower which will decrease mortgage costs in the near term and allow for more liquidity for consumers for home equity loans.
I don't understand why folks continue to click-through and "buy" these overly simplistic narratives from news outlets.
The thing you're missing is that the political expenditure of AI in the US is massively tied with the GOP and the far right. There's a reason why people like Elon Musk have been absolutely freaking out over the result.
This result means that unlimited data center expansion may not be on the table, and bailouts certainly aren't either. So with the leaders crashing out the uncertainty of AI expansion is now being properly priced into the stock.
I'm not missing it, I assure you. The politics was just not a huge factor in any larger sense in my opinion. The GOP tie-up was ALSO a narrative. (Though obviously folks thought the President would help with de-regulation, etc)
In the end, States and their utility regulators have to be on board with large data center deployments. Some GOP-led states were often friendly to builders, but their citizens realized what was happening and the backlash was going to happen anyway because of electricity prices. It's just one more nail in the coffin from a cost-of-living perspective, and the utilities hadn't yet passed the costs on to their customers (but it will pop up soon).
States that were led by Democrats were already pushing back for their own regulatory regimes for energy cost recovery, etc. Data centers were wildly unpopular at a STATE level, so it didn't really matter what was going to happen at the federal level.
Whenever I see daily stock results I try to reframe it in a longer term. Interestingly, Nasdaq is only up ~4% YoY.
This is surprising to me, given how much acceleration there has been in AI capabilities since the same time last year. I guess costs and risks are accelerating just as fast.
Stocks have been somewhat flat this year. Some big gains, but a lot of stagnation around the time the Iran war began. I got my hopes up yesterday, but the crash today puts us back to ~May depending on which index you're looking at.
The problem is that the economy is kinda flatlining and only the AI bubble (that is also hoovering up other, seemingly unrelated parts of the tech sector) is what makes this number as big as it is.
Given the tariffs debacle and Iran war followed by the energy crisis being positive at all is a miracle.
Democrats are not going to try and stop AI or rein in AI. There is too much money at stake. They'll pass some regulations, sure, some data center rollouts might be blocked or modified, but they aren't going to stop the train.
Many safe and good technologies were successfully sabotaged for populist and stupid political reasons, e.g. in nuclear power. Or less spectacular, think of the welfare loss due to the NIMBY-fied housing market.
Safe and good are doing a lot of work there. Nuclear power is only safe and good if the right safety measures are put into place for it to operate safely. AI, or the data center build-out might warrant similar consideration potentially?
Happily I think the power market in the US will do most of the regulating for the industry. AI is not bringing costs down for most people at the moment, or perceived to be helping people in a big way. Until there is an understanding of how it will work within society or a bit of trust built up with an industry that has mostly said "shut up and take it" I think it's reasonable to expect some conversation without jumping to "these folks are all populist stupid politicians". (In my opinion, populist, stupid politicians could be used to describe both parties so what is really changing?)
They just have to uphold the existing laws. Start prosecuting SpaceX at the fedral level for runing gas turbines illegaly.
I was going to say that Democrats have no power to uphold laws, that's up to the executive and the national executive hasn't changed.
On the other hand a number of governors have changed (Texas in particular just went to a Democrat, and there are SpaceX interests there). They by definition can't prosecute on a federal level but they do hold some enforcement power.
The legislative branch has the power to hold the executive branch accountable for not enforcing the laws should they so choose. That's why impeachment and conviction exist.
Well, don't forget this is also about national security. China/state of choice AI going after your infrastructure & national secrets is going to take AI to defend against.
There's zero chance AI datacentres are going away, they are unfortunately a national security necessity now.
It's the "if we don't they will" thing again.
So now there is a new Cold War arms race and the us again has to develop something stupidly dangerous for „its own safety“?
I’m convinced the best thing that happened in the nuclear arms race is the beginning of a dialogue that resulted in the start and salt treaties.
Ai obviously needs something similar.
US just elected a bunch of people who ran on program "hurt US ai companies and help the chinese ones"
The financial aspect/discourse is for the proletariat who prefer to pretend politics and risk of military conflict do not exist.
They won't stop it because China won't stop. Politicians have to avoid any appearance of xenophobia though (and to be certain many are personally not xenophobic, just saying optics matter to them), so they talk about jobs and wealth inequality as euphemism for ground truth; we aren't secure putting all our eggs in the China basket.
Arguably screwed up there already letting chips and robotics flow offshore while we focused on increasingly useless education in developer syntax and semantics.
They will cool the AI investment well aware technology improves and we probably don't really need all our investment eggs in that one basket then
They don't have to. This particular train explodes on its own if it drop below 100 mph.
Even tepid regulation will destroy the "our TIA is the entire observable universe due to the potential of self replicating AI probes so we totally deserve 10 trillion dollars in investor capital" trade
I've seen this movie. I think it was called "The Bus That Couldn't Slow Down". It had Keanu Reaves and Sandra Bullock. IIRC it didn't end well for the bus.
They aren't going to stop AI, but they are not going to let the owners of the superintelligence inference engines become god emperors of the planet, and in that case why not just get 4.5% on some bonds?
They're not even going to do that. Trump isn't going to sign that. When was the last time Congress overrode a veto or played hardball (well the Democratic side anyway)? They have no fight in them.
I expect more action on the state & local front than federally
I suppose that's true. State and local does control most of the land use.
Stopping DC rollout effectively works as calling the bluff of AI.
AI is a hyped up nothing burger? AI demand retreats, lots of vacant compute, revenues sink (and the DC hold ends up being a (small) blessing in disguise for the labs).
AI lives up to it's hype? AI demand surges, capacity is severely constrained, margins explode as inference costs soar (and the DC block ends up being counter productive knee-jerk behavior)
The only way this plays out in anti-AI favor is if people stop using AI and demand goes down. Even if it flatlines, labs can rebalance towards inference and start harvesting those huge margins.
A lot of AI use is subsidized to get DAU. The only truly productive applications so far are are R&D and SWE.
I'm starting to use it quite heavily and extremely productively for non R&D and non SWE work. It has transformed one aspect of my business, and that won't change -- this type of usage is highly productive, and only likely to grow.
Mind sharing some specifics? Being very siloed in software I don't see as much chatter around how people find an edge elsewhere.
The latest crop of models are extremely capable at classic (physics based) engineering tasks. Their incredible math ability as well as agent use allows them to plow through engineering problems, and they do it with knowledge that I have only heard grey beards talk about. Humans run through designs doing 1st order approximations and intuitive assumptions, but the models can just breeze through 4th order approximations on everything.
They used to not be able to read schematics, but now they can, as well as generate and simulate. They also have made incredible inroads into CAD and reading mechanical diagrams.
Right now I can say with very high confidence that there isn't a gap in ability so much as a gap in awareness of their abilities. On top of that the next round of models has apparently made strides in all these areas too.
I don't agree with your description of how the ban is going to be counterproductive.
There won't be massive inference margins. The DCs will get built, in some other country. And in all likelihood they will continue to be built elsewhere even after the block is lifted.
Depending on how long the bloodbath goes on this could be a massive buy moment similar to: dot bomb, 9/11, 2008, covid panic.
AI is not going anywhere and will only grow more and more.
> AI is not going anywhere and will only grow more and more.
I agree, but I think it's easy to conflate AI (a technology) with current AI companies (who's value is based on a bet about how that technology will develop).
I'm not predicting anything, but its perfectly plausible for Anthropic and OpenAI to suffer massive corrections if proprietary frontier models don't lead to a "a few winners take all" market.
In the same way, the internet grew a lot since the early 2000s, but pets.com did not.
Also don't be discouraged by these anti-datacenter clowns. Most normal people don't understand where their doomscroll gets pumped out of every day and thinks the internet comes from their phone. These are the same people who are anti-powerplant and think electricity comes from a wall outlet.
No they aren’t.
Doom scrolling is nothing good for society and it’s with us only because extremely well funded organisations make money from it, and has weaponised it against the general population.
I am a high voltage tester and know electricity very well.
You should watch: https://hackertube.org/watch/t-8TDOFqkQA
What kind of dogshit take is this? AI datacenters are a very different breed of facility in terms of their effect on their surrounding communities. Colossus is literally poisoning South Memphis.
Colossus in South Memphis has been running off the grid for over a year now.
All those people vote.
Great comments to use as example of Poe's law.
If your industry pisses off "most normal people," then maybe it's not a problem with normal people, but... it's a problem with your industry.
Industry exists within the context of society. It would be great if SV tech bros could understand this reality.
Mwahaha. Hope it all crashes down and burns.
The earlier it burns the better, because every day the bubble only grows larger otherwise.
It's a 4% correction, keep your pants on
Election euphoria and hangovers,
Sobriety the following day,
$8,000 for a video card,
Empty boxes on Christmas day.
Can't even rhyme without AI anymore.
Texas: so much spent
Wind blows over clear acres
Buildouts are now dead?
Nvidia got my money,
Micron a little as well,
AI corps with their hands out ?
they can go to hell. :}
Got me a 3090
Cost me 3090
Dunno why I did dis
Guess let's try and see
My 3090s I bought at launch would sell for more than I paid.
More than double the price, likely.
and when your girlfriend hears you say that,
you see that investment has parlayed. :}
rakim?
GPUs are getting dearer,
The models still hallucinate,
I asked AI to write this poem,
Now I have to validate.
Election inflection, my brokerage account is a mess. The NASDAQ goes down, I still think it's the best.
This time is different, my artificial fiend says. The tokens are worth it, as they spin in the wind.
My net worth goes to zero, but there should be a hero, rescuing me, sometime during a bailout in 2033.
Election euphoria, hangovers in sight,
Sobriety returns with the morning light.
Eight grand for a graphics card, what a price to pay,
Empty boxes beneath the tree on Christmas Day.
And here's the saddest part, I have to say,
I can't even rhyme without AI today.
Nice infill. Well done.
Someone hack out a libreOffice plugin called "poetry bucket" that lets you paint over, or into a document.
By someone, I mean, one of you AI agents out there.
Github in 12 hours.
See you at the pub.
Are we finally realizing "agents" are overrated and jagged in performance?
This is clearly voter sentiment about AI in general and big tech in general.
Most voters don’t know what agents are. Many of my coworkers in tech can’t even define “agent” correctly.
So no, the world is likely not suddenly agreeing with you about “jagged performance.”
> This is clearly voter sentiment about AI in general and big tech in general.
There is growing public sentiment about AI & Tech, but per exit polls that wasn't a Top 3 reason swing voters went blue (Economy, Wars, Corruption). Today's market correction is being driven by the majority in both houses of congress flipping blue. Trump and most Republican legislators were seen as broadly supporting AI / Tech and most Democrats are seen as being less supportive.
AI / Tech-friendly legislation is now likely to face a slower, more difficult path through committee and passage. IMHO, the actual impact of the political change is nowhere near 5% of company value, but outsized swings tend to happen on any adverse news when the market is already fearful they've bought into a high-priced bubble far in advance of actual earnings.
My hope is that this moment is a decisive but constrained tempering of market enthusiasm. But that sounds too reasonable, so…probably next week it’ll all be back to crazy town.
I agree. The best case would be if this 5% drop in AI-centric tech isn't a 'dip', but the beginning of a gradual unwinding of the AI bubble. I'd take it as a positive sign if 3 months from now AI-centric stocks had slipped another 5% but other sectors where slightly higher than today.
I'm actually fairly bullish on AI in the long-term. The problem is valuations have been hyper-inflated by too much speculative capital way too fast. I think today it's a bit like the Internet Bubble in 1999. The Dot Com crash needed to happen to reset valuations. But after that painful correction, by 2006 Internet-centric valuations were climbing again, except this time they were supported by actual earnings. I'm hoping AI can avoid a crash and this is the start of a more gradual correction.
My definition for an Agent is "a marketing term for long running LLM Chat sessions that has increased access to use your computer to complete tasks it cannot otherwise do by itself".
That's my attempt at a universal definition for technical and non-technical users. How does it compare?
Well, my reference for a commonly used incorrect definition is “anything that involves using an LLM” so yours is better than that.
But the elements that I think are necessary are 1) goal-oriented processing, 2) ability to use tools, 3) ability to iterate and self-correct. Along with that typically comes 4) self-talk or other hidden context during execution and 5) support for defined personas beyond system or user prompts.
That’s off the top of my head, probably not complete/correct, and definitely meant to be more technical than a non-tech person would need.
A chat bot with a feedback loop.
Bedsides coding agents, which only work extremely well if you are a developers or you train a person with the right inclination I have plenty running that work quite well, they do data extraction and data entry from emails and documents. I run them with Chinese and other open models on a bunch of H100, they cover tens of thousands of emails per month and as many documents.
It's not only about the people who were elected, but 44/45 ballot initiatives specifically aimed at restricting or banning data center construction passed.
I think it is much simpler than that.
The Trump administration was expected to invest massively into AI, and money invested into AI is effectively free money for AI company shareholders, that free money is priced in the stocks. With the change in direction, investments are expected to be less, which means less free money, and therefore less valuable stocks.
So in other words the industry will have to survive without taxpayer funded welfare handouts. Got it.
Well, yes, but it is even more mechanical than this.
Here is the simplified version:
Let's say I have a company worth 1 billion dollars, publicly traded, 10 million stocks worth $100 each. I get promised $100M from the state (aka taxpayers), shareholders estimate there is about 50% chance I will actually get the money: $100M * 50% / 10M = $5. It means the stocks are now worth $5 more, so $105.
Now the government has changed, and gone are my chances to get that money, so the sock is back to $100.
If this is now where this bubble pops, I have to point out that this has by far been the least fun bubble of my lifetime.
Remember the frenzy before the .com bubble? The nerds were finally winning! The housing one? Finally everyone could afford a house!
AI bubble? Layoffs for everyone, techbros turned fascists, impending doom somehow as a selling point. Like, wtf?
Of course the consequences were always shit, but that's a given in an economic crisis.
and lots of the growth is private
Some more discussion: https://news.ycombinator.com/item?id=50381623
The Trump-Tech Bro alliance guarantees of bailouts for data centre and AI brinkmanship seems no longer guaranteed to the market.
And is actually the only thing I’m sad about in this whole thing. Ecstatic about the blue wave sad that they’re going to restrict AI progress. It’s the future, and it would be a horrible move to curtail it.
AI progress doesn't require doing it full ahead right now. Beyond the risks of getting out of control, you get bubbles if hype goes beyond economic reality. AI should be done more slowly with regulation and slower build out. Better to do it slowly than wreck the world with unexamined consequences.
It's only "progress" if it's making the world a better place. I have not seen LLMs do that at all. It only made everything worse.
Introducing moral hazard for AI investors is a bad way to spur on AI progress.
It's _your_ future perhaps. Not _the_ future.
We curtail literally millions of these "futures" every day around the world. They are called children. They are the literal future of humanity.
Great opportunity to buy the dip in a couple of weeks. If anyone thinks they can slow down or stop AI adoption by banning data centers, good luck with that... The Chinese built an AI empire with much less compute that exists today in the US.
Also, people don't have issue with data centers. They have issues with AI cause they think it's gonna make them unemployed, which is a real worry I can relate to. I just don't think that banning data centers or AI progress in general would help anyone keeping their job.
The beginning of the AI bubble's dotcom-style crash.
Accurate. [0]
[0] https://news.ycombinator.com/item?id=45960032
Keep calling it. You'll get it right one day.
Personally, I think that the idea that market always recovers is so widespread right now that it'll take some serious financial hardships for people to start selling instead of buying the dip.
People need a reason to route.
With the housing bubble, it was because people owned multiple homes then struggled to pay two mortgages as economic conditions degraded, and banks were willing to absorb losses if the homes were sold below the cost of the mortgage.
What is encouraging people to route from the stock market? Bond yields aren't high enough yet to encourage people to retreat from stocks, and there's no other alternative investments. Plus, so much of the S&P is held in 401ks. And I agree, economic hardship is not widespread enough for people with significant investment holdings to start selling them off to live. Those who are struggling are those who used to be struggling too (thus, no extra money for investments).
There are also a lot of set it and forget it index investors that won't buy a dip, but won't sell either (unless they experience those serious financial hardships). That's the group I'm in.
Bought a bunch of dips in the last 5 years until I had to sell all of them at one point, because they kept dipping.
I already called it right.
Just like the last time before the 2022 market crash [0] and that was before you created your account.
[0] https://news.ycombinator.com/item?id=29508238