AGI arrives in 36 months. The AI labs building it will still face bankruptcy?
This is basically what Anthropic's CEO says. With the current approach, he predicts AGI (artificial general intelligence) in 1-3 years. Yet, he also says that he does not want to purchase the $1 Trillion in compute required to do exactly that.
In other words: Today, the technical scaling of AI is outpacing economic diffusion. Which is of course a massive structural risk.
The AI models are improving EXPONENTIALLY while encountering enterprise procurement cycles LINEARLY.
This creates a temporal cash-flow trap.
If a frontier lab overestimates corporate adoption velocity by 12 months, they have $100bn in CapEx that will force them immediately into insolvency.
Which means AI labs currently (must or should) under-procure compute relative to technological progress.
Which creates an artificial constraint on how fast AI could actually progress.
The first question now is:
If a handful of companies are making trillion-dollar bets on a technology that could reshape civilization, but their survival hinges on perfectly timing sluggish enterprise adoption in an unpredictable macro environment, who actually bears the cost if their forecasts are wrong? The executives and investors who placed the bets, or taxpayers who will be told it's too dangerous to let them collapse?
And from a different perspective:
Should the AI companies who are building the most consequential technology in human history be able to go bankrupt? Or does that risk force us to accept a new and different arrangement?
This is basically what Anthropic's CEO says. With the current approach, he predicts AGI (artificial general intelligence) in 1-3 years. Yet, he also says that he does not want to purchase the $1 Trillion in compute required to do exactly that.
In other words: Today, the technical scaling of AI is outpacing economic diffusion. Which is of course a massive structural risk.
The AI models are improving EXPONENTIALLY while encountering enterprise procurement cycles LINEARLY.
This creates a temporal cash-flow trap.
If a frontier lab overestimates corporate adoption velocity by 12 months, they have $100bn in CapEx that will force them immediately into insolvency.
Which means AI labs currently (must or should) under-procure compute relative to technological progress.
Which creates an artificial constraint on how fast AI could actually progress.
The first question now is:
If a handful of companies are making trillion-dollar bets on a technology that could reshape civilization, but their survival hinges on perfectly timing sluggish enterprise adoption in an unpredictable macro environment, who actually bears the cost if their forecasts are wrong? The executives and investors who placed the bets, or taxpayers who will be told it's too dangerous to let them collapse?
And from a different perspective:
Should the AI companies who are building the most consequential technology in human history be able to go bankrupt? Or does that risk force us to accept a new and different arrangement?