It’s going to be an interesting period ahead. The market is going to struggle to absorb SpaceX, OpenAI, Anthropic and others. They’re just massive relative to typical IPOs. Maybe one can sneak through but not all. Someone is going to have to have a very bad day and it’s not clear who yet.
Separately there’s a big battle to keep these folks out of the S&P index, because many funds (some of whom are required to buy index stocks) think they’re horribly over valued and will tank once floating.
It's very unlikely, part of the reason why valuations are so high is because there is so much money, not just in the US, but globally, that is desperately seeking a place to park.
You're thinking backwards. The money is parked in these companies. They basically figured out a strategy to maximize return in an environment where the public market is dominated by passive capital.
It's time for Google/Elon/Softbank/sovereign wealth funds, etc to cash out.
> an environment where the public market is dominated by passive capital
While by "the end of May 2024, passive mutual funds and ETF assets" had "grown to nearly 60% of the equity fund market" (emphasis mine) [1], only about a third of American stocks are held by passive capital [2]. (Index funds are about 16%.)
> part of the reason why valuations are so high is because there is so much money
The primary mechanism behind that is liquidity; Passively held assets decrease liquidity, and with there being a lot of passively held assets, liquidity isn't so great.
There are some warning signs that demand for AI companies, especially OpenAI, may not be so high as expected. https://www.bloomberg.com/news/articles/2026-04-01/openai-de... (Though OpenAI asserts that the demand is low because the seller isn't trusted)
On the other hand, it also seems like OpenAI and Anthropic are seeking to have a relatively "small" IPO by not selling too many shares. The market may absorb a few tens of billions. The big questio is how long it'll take for more of those companies to hit the market afterwards; They're burning through their money awfully quickly, and the earlier investors will want out as well.
Yes FOMO and cash looking for a home did help drive up valuations to irrational levels.
Public markets are a different animal all together.
There are some concerned there won’t be enough willing to invest. That is a real risk given people smell a bubble about to pop.
The other ironic risk is if the float is too small it creates an artificially high valuation by letting a thin, momentum-driven market price 100% of the cap table, which then collapses when lockups expire, supply floods in, and the real market discovers the stock is worth far less than the marginal buyers said it was.
For example there’s talk that in that scenario SpaceX could end up with a 5T valuation… for a company that did only $18.5B in revenue last year. Thats beyond irrational even under the most aggressive growth scenarios.
> There are some concerned there won’t be enough willing to invest
I say this as a former public-equity derivatives guy, and now a private capital markets guy: I'm not seeing this seriously argued.
There is a real risk these companies are going out overvalued. But they're already overvalued in the private markets and arguably were even before near-term IPOs were on the horizon.
Markets can go risk off for any reason. If they do between now and these companies' IPOs, that spoils the party. But if they don't, there is no reason to suspect the risk capital driving into all manner of speculative equities, to say nothig of structured products and weird debt creatures, is suddenly going to hold the line at the darlings of the private markets.
> there’s talk that in that scenario SpaceX could end up with a 5T valuation
Liquid cash looking for a home is overwhelmingly white/blue collar retirement money. Billionaire money is almost always a totally undiversified portfolio of their own company. There are billions of people globally who save money and want it to grow, so they hand it over weekly to institutions that do asset management. It's these institutions that are seeking new investments, but the money isn't their own, it's largely everyday people's savings.
Regarding S&P: That is my acute concern. For individual investors, they can be removed by taking an a short position corresponding to the stock's index weight. (I do this for TSLA). More broadly, it exposes people who only use index funds or don't have shorting in their tool box (passive investors for example) to big risk.
1.1% is about 15% of the typical returns for a year in the S&P500, from a single stock. Folks and funds invest here for diversity precisely to limit such impacts.
It’s a big risk and that’s why there’s a big fuss right now to keep these guys out of the index.
The risk is no longer a single company with a dumb CEO tanking.
Now the risk is a massive stockmarket wide collapse. Over a third of the S&P 500 is AI heavy tech stocks. With how reliant Nvidia is on AI compute sales, there's a real risk that at least 7% of the market disappears instantly, and everything else will hurt massively as well.
You can't hedge against that. There's a good chance this will have contagion solely from the sheer size of a dotcom style crash in our current stock market. Pension funds would start selling off their unrelated assets to cover their losses.
Nevermind any secondary concerns; How balls deep private equity and private credit are in AI. How useless the current US administration is at both economic and monetary policy.
Those are great points. Regarding > You can't hedge against that:
Not by shorting individual stocks, but you can by deleveraging in general. Stop using margins or options, hold a larger portion of your money outside the blast zone like account balance (hopefully with interest), bonds, international.
Money wouldn’t just be diverted from other US stocks though.
Foreign money has increasing buying power as USD weakens against certain currencies and the upside of these IPOs is certainly more attractive to global investors than parking money is lack luster real-estate or bond or cash alternatives.
How is that an issue? Many funds that replicate an index are not doing full replication anyway. They do some kind of sampling that includes way fewer stocks than defined in the index.
What would stop a fund from just not including those stocks because of sampling?
Or waiting for time to settle, since even with full physical replication, they are not required to jump in and buy immediately after IPO.
I think that SpaceX and OpenAI will be ok. Ukraine war showed that Starlink is the real deal - all of Ukraine's successes (or reversed losses) in the last few months are mostly due to Starlink cutting off Russia.
OpenAI seem to have the tech on par with Anthropic and world class con artist at the top and access to more compute. So when the tsunami of Chinese models and silicone materializes - Anthropic will be most vulnerable.
These companies are not publicly traded yet so they are not in these indices. Funds are fighting to keep them out so they’re not required to buy stocks they think will tank.
In this case, really? Between Nvidia, Microsoft, AMD, Amazon and everyone else in that S&P 500 there's not >50% of OpenAI already? It's not that indirect.
Hmm, actually I would be interested to get actual number regarding the exposure to OpenAI and Anthropic in S&Py you might be right, completely possible that I underestimated
Anthropic, OpenAI and SpaceX all want to IPO within this year. There's just not enough money in the market to buy all those shares. So people might sell their shares in other companies to buy in at the IPO, then when the next one goes public they might sell the shares they just bought to jump onto the next one and so on. I don't think that there was a situation like this ever before.
The amount of new shares for sale could be very large in those three IPOs:
SpaceX: up to $75B [1]
OpenAI: at least $60B [2]
Anthropic: more than $60B [3]
Together, that would be about $195B+ of IPO shares to buy.
For comparison, all U.S. IPOs together raised $44.0B in 2025 [4].
All IPOs in the world together raised $171.8B in 2025 [5].
So where should the money come from? Either from selling shares in other companies or from loaning money which would only make sense if the Fed brings back ZIRP.
They might be conflating the valuation of these three companies with one of the recent analyses that the presumed valuation of the current “AI Economy” is an order of magnitude over the valuation of the total wealth of the world.
Another scenario would be that they release very little shares but because of their market cap they make up a significant portion of an index so index funds are forced to buy a lot of their shares which would drive prices to insane heights.
I think it's a good thing if the pricing gets more realistic. Both for the customers long-term and for the economy. To evaluate AI tools and how much benefit they provide you need to evaluate the costs as well. And right now nobody truly knows where the costs will end up. That's fine as long as the prices are stable, but they aren't.
In one example I know, a boring company that isn't a pure software company, the Github Copilot pricing change will make it around 15x as expensive as before. It's far from ideal when you cannot rely on pricing to stay somewhat stable.
Copilot isn’t a good example because they had an extremely broken pricing model that was meant to be a loss leader. For a short while if you promoted a certain way you could get a crazy number of tokens out of their billing model.
I hope price increases will mean we see serious moves towards open source and/or on device models. Feels like that's where we should all be headed but the current subsidy is a distraction.
I would self-host if it didn't cost so much for the hardware. But right now that's cost prohibitive for most individuals and small or medium businesses.
How what happened? How everyone stopped having cutting edge graphics cards in all their devices? Most people have never been in a position to be able to do local inference on cutting edge models.
Realistically Apple needs to put a few teams on it, and have built in support for running LLMs on OSX.
I assume if you integrate this on an OS level you might be able to pull off some additional tricks. So far most of the tools for running LLMs locally have been driven by volunteers or very small startups.
Microsoft could also do this, but they're too busy trying to upsell everyone. Microsoft wants to sell you on paying for each token.
Apple has the edge here since they fully control the hardware and can baseline their implementation. It feels like Microsoft's task is much harder because there's a much broader range of hardware and capabilities that they need to support.
Microsoft has spent a lot of time telling people to buy AI ready PCs.
I just don't think it's in their business model to offer on device LLM. They want you to subscribe to Copilot. Which encapsulates the entire enshitication of Windows. Microsoft demands more money. Subscribe to gamepass, subscribe to OneDrive!
I'm aware of the Surface line, but 1) they don't own the hardware stack nearly to the same degree as Apple does, 2) it makes up only a small slice of the ecosystem and is a more premium product as far as PCs go. Windows ecosystem has far more variance in terms of CPU + GPU capabilities.
I can see Microsoft pushing the subscription angle, but I do think there is a place for on-device capabilities since those will always be more responsive for small tasks.
I agree with your points. While I'm just thinking out loud, I can imagine either anthro or open AI releasing hardware for on-prem solutions in the near future.
I believe a few small companies have already specialized in this. But imagine a situation where maybe you pay a small subscription and Open AI allows you to download non open source models guaranteed to work to a degree.
MS is more like assembling the surfaces, it's not the same as having a full vertical integration like Apple is trying to achieve with their contracts with Intel, in-house modem, etc.
There is no subsidy in interference. Serving existing models is a highly profitable business. The proofs are many independent companies serving models for profit.
What is expensive is model development and training, but again, that's nothing inherent to technology. It's just that Anthropic and other western vendors made a business decision to use cheap investors money to brute force a new model development and market grab instead of investing in cost optimizations for model training and inference like Chinese model developers and providers.
As I understand it, those cost optimizations relied on someone else brute forcing new model development. If everyone was trying to do it cheaper, we wouldn't have the models we have today.
Many of the chinese developers are just drafting off of Anthropic, OpenAI, and Gemini though. Distillation is leveraging those models to achieve their leaps in training. It's not obvious to me that Anthropic and do the same. Someone has to build the advanced model to draft off of.
Anthropic have clearly been working towards this since last year when they started focussing more on building products around their models that could be monetised, instead of competing on the most advanced chatbot. IMO this is part of a broader strategy on their part to tap into the enterprise market, because that is where the money is, not in selling subsidised subscriptions to consumers. This is also what the market will want to see: a credible path towards profitability.
I have been sounding the alarm on my team that the current edict to go all-in and not plan features ("we can just throw it away and rebuild it"), not read code ("the agents should do all the coding and review") is something that will end up being a medium term regret since there will be a day of reckoning with pricing.
My sense is that these companies actively need the audience of developers to have their ability to read and write code atrophy and even stunt the growth of early- and mid-career engineers; create the dependency.
It will be interesting to see how this works out because some parts of the equation should get better over time (better algos, better infra), but there are now billions of dollars of investment to recoup and it will continue to be an arms race that requires more money.
Am I correct in thinking it's a race to who launches IPO now? I mean among the SOTA models. Especially since OpenAI has been given the all-clear by the court.
GOOD POINT by my wife: it’s too early in the AI cycle to already have winners especially with sky high valuations. It’s like betting on ibm to win the pc race.
VC’s weren’t going to keep them propped up forever and it’s not like any of these companies have even a semblance of a plan to get into the black, so unfortunately this was always going to be the next stage for any LLM start up not bought out and folded in to a big player.
I'd say it's a mixed bag. Yes, price increases are/were expected. But blocking 3rd party harnesses from their subscription and also moving SDK/claude-p access out of the subscription is blocking innovation and therefore future use of Claude models. What I mean is, while Claude models are SOTA, Claude Code is not. It's full of bugs and shortcomings and new innovative harnesses might make better use of the model, but this won't happen now as they are blocked. Same for all the people building their own scripts/workflows around claude -p or the SDK, they will now stop inventing new stuff on top of Claude.
I think it also doesn’t help that they’re all different flavors of the same tool and it’s very easy to jump between them. None of them actually made a particularly distinct product, and they want us to use their tool to make/justify the nebulous billion dollar product.
As for subscription/token costs, even with increases they’re not even remotely covering costs. If people actually paid what it cost for these companies to even break even, nobody would be using these tools. They simply aren’t that consistently useful despite all the grand claims. They can be useful and in some areas they are very useful, but nobody is going to spend thousands of dollars a month to have something rewrite their emails regularly. And it’s not like these companies are trying to target one industry. They want to target everyone.
I wonder if there's a bit of a standoff situation here where every company feels like they have to burn enormous amounts of cash training/iterating on their models because if they stop, they'll get leapfrogged too quickly. So unless they all agree to just settle where they are more or less and focus on building out functionality and tooling, they all have to keep spending at an unsustainable rate.
After Anthropic achieve's "safe" AGI (A Giant IPO). Those token prices are NOT going down as I am predicting [0]
Why would they?
They need to pay for the increasing costs and the high demand for running Claude and soon it will be reflected in their earnings releases. So every token cost counts and the subsidization era of tokens will eventually end.
The real test is going to be anthropic’s IPO + when Microsoft charges for copilot on June 1. I think a lot of companies are going to learn people are not interested in paying for these tools because they just aren’t consistently useful across-the-board. Like 3-D printing it will thrive in certain industries but will not have nearly the broad appeal and daily use they promise investors.
Well it will always be good in a way, but probably won't become better in the near future. Opus 4.7 was a downgrade in a way that gives Anthropic more control and better margins. And they keep Mythos away from the normies, giving access only to large corporations who pay millions for it.
It’s going to be an interesting period ahead. The market is going to struggle to absorb SpaceX, OpenAI, Anthropic and others. They’re just massive relative to typical IPOs. Maybe one can sneak through but not all. Someone is going to have to have a very bad day and it’s not clear who yet.
Separately there’s a big battle to keep these folks out of the S&P index, because many funds (some of whom are required to buy index stocks) think they’re horribly over valued and will tank once floating.
Get your popcorn ready.
>The market is going to struggle to absorb
It's very unlikely, part of the reason why valuations are so high is because there is so much money, not just in the US, but globally, that is desperately seeking a place to park.
You're thinking backwards. The money is parked in these companies. They basically figured out a strategy to maximize return in an environment where the public market is dominated by passive capital.
It's time for Google/Elon/Softbank/sovereign wealth funds, etc to cash out.
For you, it's time to tilt to bond and value.
> an environment where the public market is dominated by passive capital
While by "the end of May 2024, passive mutual funds and ETF assets" had "grown to nearly 60% of the equity fund market" (emphasis mine) [1], only about a third of American stocks are held by passive capital [2]. (Index funds are about 16%.)
> For you, it's time to tilt to bond and value
Or just buy the market and watch the show.
[1] https://www.americancentury.com/institutional-investors/insi...
[2] https://alexchinco.com/double-what-you-think-it-is.pdf
Fair, that was a flippant response on my part. I'm of the age where sequence of returns is a impactful risk.
> part of the reason why valuations are so high is because there is so much money
The primary mechanism behind that is liquidity; Passively held assets decrease liquidity, and with there being a lot of passively held assets, liquidity isn't so great.
There are some warning signs that demand for AI companies, especially OpenAI, may not be so high as expected. https://www.bloomberg.com/news/articles/2026-04-01/openai-de... (Though OpenAI asserts that the demand is low because the seller isn't trusted)
On the other hand, it also seems like OpenAI and Anthropic are seeking to have a relatively "small" IPO by not selling too many shares. The market may absorb a few tens of billions. The big questio is how long it'll take for more of those companies to hit the market afterwards; They're burning through their money awfully quickly, and the earlier investors will want out as well.
Yes FOMO and cash looking for a home did help drive up valuations to irrational levels.
Public markets are a different animal all together.
There are some concerned there won’t be enough willing to invest. That is a real risk given people smell a bubble about to pop.
The other ironic risk is if the float is too small it creates an artificially high valuation by letting a thin, momentum-driven market price 100% of the cap table, which then collapses when lockups expire, supply floods in, and the real market discovers the stock is worth far less than the marginal buyers said it was.
For example there’s talk that in that scenario SpaceX could end up with a 5T valuation… for a company that did only $18.5B in revenue last year. Thats beyond irrational even under the most aggressive growth scenarios.
> There are some concerned there won’t be enough willing to invest
I say this as a former public-equity derivatives guy, and now a private capital markets guy: I'm not seeing this seriously argued.
There is a real risk these companies are going out overvalued. But they're already overvalued in the private markets and arguably were even before near-term IPOs were on the horizon.
Markets can go risk off for any reason. If they do between now and these companies' IPOs, that spoils the party. But if they don't, there is no reason to suspect the risk capital driving into all manner of speculative equities, to say nothig of structured products and weird debt creatures, is suddenly going to hold the line at the darlings of the private markets.
> there’s talk that in that scenario SpaceX could end up with a 5T valuation
Where?
Curious.
What would you consider a fair value? Would be interested to see your valuation approach
See why it doesn’t trickle down?
It is trickled down money.
Liquid cash looking for a home is overwhelmingly white/blue collar retirement money. Billionaire money is almost always a totally undiversified portfolio of their own company. There are billions of people globally who save money and want it to grow, so they hand it over weekly to institutions that do asset management. It's these institutions that are seeking new investments, but the money isn't their own, it's largely everyday people's savings.
Has anyone done a first-order estimate of the tax windfall about to hit state and federal coffers?
Can you elaborate on what you mean?
Regarding S&P: That is my acute concern. For individual investors, they can be removed by taking an a short position corresponding to the stock's index weight. (I do this for TSLA). More broadly, it exposes people who only use index funds or don't have shorting in their tool box (passive investors for example) to big risk.
Tesla is currently 2.2% of S&P 500. If it halves, losing $800,000,000,000 usd market cap, the index will go down by 1.1%. What big risk?
1.1% is about 15% of the typical returns for a year in the S&P500, from a single stock. Folks and funds invest here for diversity precisely to limit such impacts.
It’s a big risk and that’s why there’s a big fuss right now to keep these guys out of the index.
> What big risk?
Tesla doesn't drag everyone else down.
>50% of the S&P 500 is involved with these AI providers. There will be a cascading effect.
The risk is no longer a single company with a dumb CEO tanking.
Now the risk is a massive stockmarket wide collapse. Over a third of the S&P 500 is AI heavy tech stocks. With how reliant Nvidia is on AI compute sales, there's a real risk that at least 7% of the market disappears instantly, and everything else will hurt massively as well.
You can't hedge against that. There's a good chance this will have contagion solely from the sheer size of a dotcom style crash in our current stock market. Pension funds would start selling off their unrelated assets to cover their losses.
Nevermind any secondary concerns; How balls deep private equity and private credit are in AI. How useless the current US administration is at both economic and monetary policy.
Those are great points. Regarding > You can't hedge against that:
Not by shorting individual stocks, but you can by deleveraging in general. Stop using margins or options, hold a larger portion of your money outside the blast zone like account balance (hopefully with interest), bonds, international.
Interesting yes.
Money wouldn’t just be diverted from other US stocks though.
Foreign money has increasing buying power as USD weakens against certain currencies and the upside of these IPOs is certainly more attractive to global investors than parking money is lack luster real-estate or bond or cash alternatives.
TINA (to US stock market) and all that.
Then Anthropic will gain, people will profit take, and it'll all flow back out to everything else.
How is that an issue? Many funds that replicate an index are not doing full replication anyway. They do some kind of sampling that includes way fewer stocks than defined in the index.
What would stop a fund from just not including those stocks because of sampling?
Or waiting for time to settle, since even with full physical replication, they are not required to jump in and buy immediately after IPO.
Index funds make legal commitments to replicate the entire index. I have no idea why you think they use sampling.
The whole point of an index fund is I'm not paying someone to try to guess what stocks are going to under/over perform the S&P.
And when they are required to buy is not really a mystery either. they have very little discretion.
NASDAQ did changes to be able to fast track SpaceX, S&P is looking at the same as far as I understood: https://finance.yahoo.com/news/new-rule-could-fast-track-spa...
Many Vanguard index funds use sampling, in fact. It's more complicated than you think.
But normally they wait a year for new companies, whereas Sam and friends have managed to buy some officials and have made it a day-one-rug-pull.
I think that SpaceX and OpenAI will be ok. Ukraine war showed that Starlink is the real deal - all of Ukraine's successes (or reversed losses) in the last few months are mostly due to Starlink cutting off Russia.
OpenAI seem to have the tech on par with Anthropic and world class con artist at the top and access to more compute. So when the tsunami of Chinese models and silicone materializes - Anthropic will be most vulnerable.
Huh. Anthropic seems like they are in a significantly better position tech and finance-wise.
> Ukraine war showed that Starlink is the real deal
That held before SpaceX merged X, xAI and friends. Those are bleeding way more money than Starlink can recover.
[dead]
> because many funds (some of whom are required to buy index stocks) think they’re horribly over valued and will tank once floating
Most of S&P 500 have shares in these anyway e.g. Microsoft and OpenAI. Not really making THAT much of a difference.
? No.
These companies are not publicly traded yet so they are not in these indices. Funds are fighting to keep them out so they’re not required to buy stocks they think will tank.
There is indirect exposure to Anthropic/OpenAI through all the incestuous partnerships between hyperscalers and NVDA
True, but that’s tiny relative to these companies going public themselves. Totally different ballgames
lol don’t compare private market pricing where the line only goes up and to the right.
Exactly. Google has SpaceX stock.
To make matters worse all the banks, insurers and investment firms in that S&P list are likely in there somewhere too.
I'm sure I'm missing a whole lot more.
Indirect exposure is just a completely different beast than direct exposure. It’s a different level of risk altogether
> It’s a different level of risk altogether
In this case, really? Between Nvidia, Microsoft, AMD, Amazon and everyone else in that S&P 500 there's not >50% of OpenAI already? It's not that indirect.
Microsoft alone is 27%.
Hmm, actually I would be interested to get actual number regarding the exposure to OpenAI and Anthropic in S&Py you might be right, completely possible that I underestimated
Not to mention Tesla.
Anthropic, OpenAI and SpaceX all want to IPO within this year. There's just not enough money in the market to buy all those shares. So people might sell their shares in other companies to buy in at the IPO, then when the next one goes public they might sell the shares they just bought to jump onto the next one and so on. I don't think that there was a situation like this ever before.
> There's just not enough money in the market to buy all those shares
What are you basing this on?
The amount of new shares for sale could be very large in those three IPOs:
SpaceX: up to $75B [1]
OpenAI: at least $60B [2]
Anthropic: more than $60B [3]
Together, that would be about $195B+ of IPO shares to buy.
For comparison, all U.S. IPOs together raised $44.0B in 2025 [4].
All IPOs in the world together raised $171.8B in 2025 [5].
So where should the money come from? Either from selling shares in other companies or from loaning money which would only make sense if the Fed brings back ZIRP.
[1] https://www.reuters.com/business/aerospace-defense/spacex-ta...
[2] https://www.reuters.com/business/openai-lays-groundwork-jugg...
[3] https://www.investing.com/news/stock-market-news/anthropic-c...
[4] https://www.renaissancecapital.com/review/2025USReview_Publi...
[5] https://www.ey.com/en_ie/newsroom/2026/01/global-ipo-market-...
> Together, that would be about $195B+ of IPO shares to buy...For comparison, all U.S. IPOs together raised $44.0B in 2025
Net buying of corporate equities by American households, trusts, funds and non-profits has averaged $660bn per year for the last few years [1].
[1] https://www.federalreserve.gov/releases/z1/20260319/html/f22... line 26, 2023 to 2025
>So where should the money come from?
Selling shares of other companies. If you think there isn't enough capital headroom, it's not SpaceX/OpenAI/Anthropic you should be worried about.
They might be conflating the valuation of these three companies with one of the recent analyses that the presumed valuation of the current “AI Economy” is an order of magnitude over the valuation of the total wealth of the world.
Another scenario would be that they release very little shares but because of their market cap they make up a significant portion of an index so index funds are forced to buy a lot of their shares which would drive prices to insane heights.
I think it's a good thing if the pricing gets more realistic. Both for the customers long-term and for the economy. To evaluate AI tools and how much benefit they provide you need to evaluate the costs as well. And right now nobody truly knows where the costs will end up. That's fine as long as the prices are stable, but they aren't.
In one example I know, a boring company that isn't a pure software company, the Github Copilot pricing change will make it around 15x as expensive as before. It's far from ideal when you cannot rely on pricing to stay somewhat stable.
Copilot isn’t a good example because they had an extremely broken pricing model that was meant to be a loss leader. For a short while if you promoted a certain way you could get a crazy number of tokens out of their billing model.
I hope price increases will mean we see serious moves towards open source and/or on device models. Feels like that's where we should all be headed but the current subsidy is a distraction.
I would self-host if it didn't cost so much for the hardware. But right now that's cost prohibitive for most individuals and small or medium businesses.
Funny how that happened, isn’t it?
How what happened? How everyone stopped having cutting edge graphics cards in all their devices? Most people have never been in a position to be able to do local inference on cutting edge models.
I agree but with the current hardware pricing it is a big problem
Just be patient for a few years. Even with the end of Moore's Law, compute hardware pricing always declines over the long run.
Yeah I am waiting but:
If anything breaks on my pc, it's going to be really rough.
My kids grow fast,a few years is a lot for them too.
And i have a bunch of games I want to play with my wife that require 2 computers sadly
On device is still meh.
Realistically Apple needs to put a few teams on it, and have built in support for running LLMs on OSX.
I assume if you integrate this on an OS level you might be able to pull off some additional tricks. So far most of the tools for running LLMs locally have been driven by volunteers or very small startups.
Microsoft could also do this, but they're too busy trying to upsell everyone. Microsoft wants to sell you on paying for each token.
Apple has the edge here since they fully control the hardware and can baseline their implementation. It feels like Microsoft's task is much harder because there's a much broader range of hardware and capabilities that they need to support.
Microsoft sells hardware. Surface laptops.
Microsoft has spent a lot of time telling people to buy AI ready PCs.
I just don't think it's in their business model to offer on device LLM. They want you to subscribe to Copilot. Which encapsulates the entire enshitication of Windows. Microsoft demands more money. Subscribe to gamepass, subscribe to OneDrive!
I'm aware of the Surface line, but 1) they don't own the hardware stack nearly to the same degree as Apple does, 2) it makes up only a small slice of the ecosystem and is a more premium product as far as PCs go. Windows ecosystem has far more variance in terms of CPU + GPU capabilities.
I can see Microsoft pushing the subscription angle, but I do think there is a place for on-device capabilities since those will always be more responsive for small tasks.
I agree with your points. While I'm just thinking out loud, I can imagine either anthro or open AI releasing hardware for on-prem solutions in the near future.
I believe a few small companies have already specialized in this. But imagine a situation where maybe you pay a small subscription and Open AI allows you to download non open source models guaranteed to work to a degree.
It's going to be an interesting decade.
MS is more like assembling the surfaces, it's not the same as having a full vertical integration like Apple is trying to achieve with their contracts with Intel, in-house modem, etc.
There is no subsidy in interference. Serving existing models is a highly profitable business. The proofs are many independent companies serving models for profit.
What is expensive is model development and training, but again, that's nothing inherent to technology. It's just that Anthropic and other western vendors made a business decision to use cheap investors money to brute force a new model development and market grab instead of investing in cost optimizations for model training and inference like Chinese model developers and providers.
As I understand it, those cost optimizations relied on someone else brute forcing new model development. If everyone was trying to do it cheaper, we wouldn't have the models we have today.
Many of the chinese developers are just drafting off of Anthropic, OpenAI, and Gemini though. Distillation is leveraging those models to achieve their leaps in training. It's not obvious to me that Anthropic and do the same. Someone has to build the advanced model to draft off of.
Anthropic have clearly been working towards this since last year when they started focussing more on building products around their models that could be monetised, instead of competing on the most advanced chatbot. IMO this is part of a broader strategy on their part to tap into the enterprise market, because that is where the money is, not in selling subsidised subscriptions to consumers. This is also what the market will want to see: a credible path towards profitability.
I have been sounding the alarm on my team that the current edict to go all-in and not plan features ("we can just throw it away and rebuild it"), not read code ("the agents should do all the coding and review") is something that will end up being a medium term regret since there will be a day of reckoning with pricing.
My sense is that these companies actively need the audience of developers to have their ability to read and write code atrophy and even stunt the growth of early- and mid-career engineers; create the dependency.
It will be interesting to see how this works out because some parts of the equation should get better over time (better algos, better infra), but there are now billions of dollars of investment to recoup and it will continue to be an arms race that requires more money.
Am I correct in thinking it's a race to who launches IPO now? I mean among the SOTA models. Especially since OpenAI has been given the all-clear by the court.
So you mean just Anthropic vs OpenAI?
you're right, the other SOTA's are megacorp whic have no interest in spinning off their AI to stand alone.
GOOD POINT by my wife: it’s too early in the AI cycle to already have winners especially with sky high valuations. It’s like betting on ibm to win the pc race.
It took tremendous effort by IBM to lose the PC race. They succeeded eventually - but it took some tries to get there.
That just means there will be legitimate ways to short them.
It will be just fine to see other tools like opencode and pi and other models flourish as subscription benefits get increasingly nerfed.
JCBP; some way to short this!
VC’s weren’t going to keep them propped up forever and it’s not like any of these companies have even a semblance of a plan to get into the black, so unfortunately this was always going to be the next stage for any LLM start up not bought out and folded in to a big player.
I'd say it's a mixed bag. Yes, price increases are/were expected. But blocking 3rd party harnesses from their subscription and also moving SDK/claude-p access out of the subscription is blocking innovation and therefore future use of Claude models. What I mean is, while Claude models are SOTA, Claude Code is not. It's full of bugs and shortcomings and new innovative harnesses might make better use of the model, but this won't happen now as they are blocked. Same for all the people building their own scripts/workflows around claude -p or the SDK, they will now stop inventing new stuff on top of Claude.
If you do not block 3rd party harness, people will discover that a good harness is more important than a good model.
I think it also doesn’t help that they’re all different flavors of the same tool and it’s very easy to jump between them. None of them actually made a particularly distinct product, and they want us to use their tool to make/justify the nebulous billion dollar product.
As for subscription/token costs, even with increases they’re not even remotely covering costs. If people actually paid what it cost for these companies to even break even, nobody would be using these tools. They simply aren’t that consistently useful despite all the grand claims. They can be useful and in some areas they are very useful, but nobody is going to spend thousands of dollars a month to have something rewrite their emails regularly. And it’s not like these companies are trying to target one industry. They want to target everyone.
Yeah I'm worried about that too. Current SOTA models might just be too expensive for most use cases if we had to pay the real costs.
I wonder if there's a bit of a standoff situation here where every company feels like they have to burn enormous amounts of cash training/iterating on their models because if they stop, they'll get leapfrogged too quickly. So unless they all agree to just settle where they are more or less and focus on building out functionality and tooling, they all have to keep spending at an unsustainable rate.
After Anthropic achieve's "safe" AGI (A Giant IPO). Those token prices are NOT going down as I am predicting [0]
Why would they?
They need to pay for the increasing costs and the high demand for running Claude and soon it will be reflected in their earnings releases. So every token cost counts and the subsidization era of tokens will eventually end.
[0] https://news.ycombinator.com/item?id=46886918
The real test is going to be anthropic’s IPO + when Microsoft charges for copilot on June 1. I think a lot of companies are going to learn people are not interested in paying for these tools because they just aren’t consistently useful across-the-board. Like 3-D printing it will thrive in certain industries but will not have nearly the broad appeal and daily use they promise investors.
I like Claude a lot. I hope it continues to be good
Well it will always be good in a way, but probably won't become better in the near future. Opus 4.7 was a downgrade in a way that gives Anthropic more control and better margins. And they keep Mythos away from the normies, giving access only to large corporations who pay millions for it.
> I like Claude a lot. I hope it continues to be good
I like strawberries. And car washing. I hope they stay unaffected by Claude's whims.
May have been better as: "I like stawberrries, and walking to the car wash to clean my car."
Of couse they are preparing for an IPO. They bleed ungodly amounts of money and have no path to approach something that resembles profitability.
The next step of the grift is offloading this into the stock market so that the investors get the exit event.
They will need a lot of bagholders this time around.