Philosophy Mondays: Human-AI Collaboration
Today's Philosophy Monday is an important interlude. I want to reveal that I have not been writing the posts in this series entirely by myself. Instead I have been working with Claude, not just for the graphic illustrations, but also for the text. My method has been to write a rough draft and then ask Claude for improvement suggestions. I will expand this collaboration to other intelligences going forward, including open source models such as Llama and DeepSeek. I will also explore other moda...
Intent-based Collaboration Environments
AI Native IDEs for Code, Engineering, Science
Web3/Crypto: Why Bother?
One thing that keeps surprising me is how quite a few people see absolutely nothing redeeming in web3 (née crypto). Maybe this is their genuine belief. Maybe it is a reaction to the extreme boosterism of some proponents who present web3 as bringing about a libertarian nirvana. From early on I have tried to provide a more rounded perspective, pointing to both the good and the bad that can come from it as in my talks at the Blockstack Summits. Today, however, I want to attempt to provide a coge...
Share Dialog
Are we in an AI bubble or not? As someone who lived through the dotcom bubble as an investor there are many parallels and also significant differences. By many metrics we are in obvious bubble territory with fantastically extended valuations and lots of roundtripping happening with NVIDIA at the center in a position similar to AOL. Yet there is also something truly different and novel: AI is the first ever technology with the potential for recursive self-improvement and we are clearly in the foothills of that self improvement with the potential for a steep takeoff at any moment.
Unlike the dotcom bubble there is absolutely no demand constraint. People and agents are consuming as many tokens as are made available. There is no shortage of bandwidth or a lack of people willing to make online payments. Rapid revenue ramps into the billions are happening, with OpenAI and Anthropic setting a breakneck pace. And increasingly, AI agents are running 24/7, consuming tokens for research, analysis, and autonomous operations — a category of demand that barely existed a year ago and is growing fast.
This time round the issue isn’t demand. It is unit economics. Right now tokens appear heavily subsidized by equity and debt dollars. For example there are numbers out there suggesting that Anthropic may be losing thousands of dollars on $200/month max plans. Inference will have to become rapidly cheaper and/or labs will need to continue raising vast sums of money to continue advancing their models. The fierce competition which includes open source models and large companies which can divert cash flow from existing businesses doesn’t really allow for increased prices.
Now some might object that NVIDIA’s stock price looks reasonable based on growth and profitability. But that of course is only true as long as their customers can continue to plow money into hardware at the current rate or even faster. Any hiccup here could kick off an incredible contraction. AOL famously managed to merge with Time Warner only to implode as a business shortly thereafter when all the equity financed advertising dollars disappeared (including huge AOL investments which were 80% roundtripped back as advertising).
Where might such a hiccup come from? One candidate is the private credit market which is showing severe stress with multiple large funds having to limit outflows. Now some people claim that this is just a consumer stampede while institutions will continue to provide credit but I strongly doubt that credit will remain available for data center and energy buildout to the same extent that it has been. Another potential candidate are IPOs of labs such as OpenAI, Anthropic and SpaceXai. Many people seem to think that IPOs would be bullish but it was IPOs that ultimately broke the dotcom bubble. IPOs require financial disclosure. In the dotcom bubble they revealed meager revenues. Here they are likely to reveal massive cash flow hemorrhaging not just from CAPEX but also from negative unit economics. As a corollary, if SpaceXai can get away with it they will try not to break out segments and just show the company as a whole.
Are we in an AI bubble or not? As someone who lived through the dotcom bubble as an investor there are many parallels and also significant differences. By many metrics we are in obvious bubble territory with fantastically extended valuations and lots of roundtripping happening with NVIDIA at the center in a position similar to AOL. Yet there is also something truly different and novel: AI is the first ever technology with the potential for recursive self-improvement and we are clearly in the foothills of that self improvement with the potential for a steep takeoff at any moment.
Unlike the dotcom bubble there is absolutely no demand constraint. People and agents are consuming as many tokens as are made available. There is no shortage of bandwidth or a lack of people willing to make online payments. Rapid revenue ramps into the billions are happening, with OpenAI and Anthropic setting a breakneck pace. And increasingly, AI agents are running 24/7, consuming tokens for research, analysis, and autonomous operations — a category of demand that barely existed a year ago and is growing fast.
This time round the issue isn’t demand. It is unit economics. Right now tokens appear heavily subsidized by equity and debt dollars. For example there are numbers out there suggesting that Anthropic may be losing thousands of dollars on $200/month max plans. Inference will have to become rapidly cheaper and/or labs will need to continue raising vast sums of money to continue advancing their models. The fierce competition which includes open source models and large companies which can divert cash flow from existing businesses doesn’t really allow for increased prices.
Now some might object that NVIDIA’s stock price looks reasonable based on growth and profitability. But that of course is only true as long as their customers can continue to plow money into hardware at the current rate or even faster. Any hiccup here could kick off an incredible contraction. AOL famously managed to merge with Time Warner only to implode as a business shortly thereafter when all the equity financed advertising dollars disappeared (including huge AOL investments which were 80% roundtripped back as advertising).
Where might such a hiccup come from? One candidate is the private credit market which is showing severe stress with multiple large funds having to limit outflows. Now some people claim that this is just a consumer stampede while institutions will continue to provide credit but I strongly doubt that credit will remain available for data center and energy buildout to the same extent that it has been. Another potential candidate are IPOs of labs such as OpenAI, Anthropic and SpaceXai. Many people seem to think that IPOs would be bullish but it was IPOs that ultimately broke the dotcom bubble. IPOs require financial disclosure. In the dotcom bubble they revealed meager revenues. Here they are likely to reveal massive cash flow hemorrhaging not just from CAPEX but also from negative unit economics. As a corollary, if SpaceXai can get away with it they will try not to break out segments and just show the company as a whole.
Philosophy Mondays: Human-AI Collaboration
Today's Philosophy Monday is an important interlude. I want to reveal that I have not been writing the posts in this series entirely by myself. Instead I have been working with Claude, not just for the graphic illustrations, but also for the text. My method has been to write a rough draft and then ask Claude for improvement suggestions. I will expand this collaboration to other intelligences going forward, including open source models such as Llama and DeepSeek. I will also explore other moda...
Intent-based Collaboration Environments
AI Native IDEs for Code, Engineering, Science
Web3/Crypto: Why Bother?
One thing that keeps surprising me is how quite a few people see absolutely nothing redeeming in web3 (née crypto). Maybe this is their genuine belief. Maybe it is a reaction to the extreme boosterism of some proponents who present web3 as bringing about a libertarian nirvana. From early on I have tried to provide a more rounded perspective, pointing to both the good and the bad that can come from it as in my talks at the Blockstack Summits. Today, however, I want to attempt to provide a coge...
Share Dialog
Nice job weaving together AI, BDCs and AOL in one post :) To what extent do you see ChatGPT, Claude etc being able to route more traffic to the older models to increase margins while they lose money on the top-end users? And do you think they have pricing power on the power users who pay $200 but gain significant economic value from using the tools?
When I read my brain creates images. When I read this my brain went blank.
Blog post: AI Bubble or Not? https://continuations.com/ai-bubble-or-not
Meanwhile, if inference costs collapse and open source models remain competitive with proprietary ones, the labs face pressure from both sides — they can't raise prices because of competition, and they can't lower costs fast enough because of the hardware investment cycle. In that scenario the value shifts from the model itself to the ecosystem built around it. One specific way open source models can stay competitive is through distillation and reinforcement learning on subtasks for agents.
So where does all of this leave us? I think that there is about a 25% or so chance that a combination of genuine recursive self-improvement and massively cheaper inference will somehow make the economics work in time. Conversely I believe there is a 75% or so probability that we will see a major correction, possibly as early as this or next year.
As with the dotcom bubble the technology is entirely real even if the valuations aren't. The dotcom bust killed Pets.com but it didn't kill the web — Amazon and Google emerged from the wreckage to become dominant companies. Similarly, an AI contraction might devastate labs and their investors, but the capabilities won't disappear. Models will still work. Inference will get cheaper as distressed assets sell. Open source will keep advancing (at least as long as there is research).
The question isn't whether AI will transform the world — it will. The question is whether the current financial structure survives, or whether we will hit a painful reset along the way. I believe the latter is significantly more likely.
Meanwhile, if inference costs collapse and open source models remain competitive with proprietary ones, the labs face pressure from both sides — they can't raise prices because of competition, and they can't lower costs fast enough because of the hardware investment cycle. In that scenario the value shifts from the model itself to the ecosystem built around it. One specific way open source models can stay competitive is through distillation and reinforcement learning on subtasks for agents.
So where does all of this leave us? I think that there is about a 25% or so chance that a combination of genuine recursive self-improvement and massively cheaper inference will somehow make the economics work in time. Conversely I believe there is a 75% or so probability that we will see a major correction, possibly as early as this or next year.
As with the dotcom bubble the technology is entirely real even if the valuations aren't. The dotcom bust killed Pets.com but it didn't kill the web — Amazon and Google emerged from the wreckage to become dominant companies. Similarly, an AI contraction might devastate labs and their investors, but the capabilities won't disappear. Models will still work. Inference will get cheaper as distressed assets sell. Open source will keep advancing (at least as long as there is research).
The question isn't whether AI will transform the world — it will. The question is whether the current financial structure survives, or whether we will hit a painful reset along the way. I believe the latter is significantly more likely.
Subscribe to Continuations
Subscribe to Continuations
>600 subscribers
>600 subscribers
Nice job weaving together AI, BDCs and AOL in one post :) To what extent do you see ChatGPT, Claude etc being able to route more traffic to the older models to increase margins while they lose money on the top-end users? And do you think they have pricing power on the power users who pay $200 but gain significant economic value from using the tools?
When I read my brain creates images. When I read this my brain went blank.
Blog post: AI Bubble or Not? https://continuations.com/ai-bubble-or-not