Artificial Intelligence

Cowen Says AI Growth Will Hit Bottlenecks

Cowen turns 1% more bearish as Imas says agents work, but bottlenecks cap GDP growth.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

3 min read

Tyler Cowen said he is 1% more bearish on AI than a week ago because institutions are not running smoothly, in a discussion published by a16z.

Cowen Says AI Growth Will Hit Bottlenecks - a16z
Cowen Says AI Growth Will Hit Bottlenecks, a16z

Cowen, the economist, pointed to strange moves around Nvidia and prediction markets for the Anthropic IPO not flying above $2 trillion, according to the discussion. That caution matches outside pricing, with Anthropic targeting $70 to $78 per share for $1.26 trillion to $1.41 trillion, below a floated $2 trillion, according to The New York Times. Cowen said Anthropic’s own projected revenue numbers did not fit a super fast takeoff.

Alex Imas did not dispute the capabilities.

Imas, also an economist, argued agentic AI marks a break from the chatbot era because systems now do work instead of answering questions, according to the discussion. He said model releases are more frequent and more powerful, with visible hockey sticks in performance, including the data in the METR report. But he said the key phrase is ill specified tasks, and the economic question is what valuable tasks each iteration unlocks that the last one did not, a link the benchmarks capture imperfectly.

Cowen said faster progress means hitting real bottlenecks sooner. He said he changed his view since August because of math, where models improve with few human constraints and a clearly better model can arrive in three weeks, according to the discussion. He said markets still look normal across real rates and stock prices, and he trusts that signal more than Bay Area revenue forecasts. Imas added that token use is rising exponentially while revenue is flat because frontier labs are cutting prices, citing RAMP data, with Anthropic and OpenAI now competing with Google DeepMind on cost for enterprise workloads.

Both rejected double digit GDP growth as a base case. Imas cited work by Chad Jones and Ben Jones, arguing explosive growth needs increasing returns to ideas with no bottleneck in idea production, but estimates from current data remain below one, according to the discussion. Cowen said Guyana once grew 40% in a year on oil and the United States topped 8% coming out of a deep recession in the 1980s, but a sustained 10% would likely need a war, pandemic or crash plus AI. He said robotics will come slowly, pointing to Waymo still not operating in Washington, D.C., and 40% of the U.S. economy stuck in government, nonprofits, health care and schools where inertia beats regulation as the constraint.

On measurement, Imas said GDP tracks welfare well but misses free digital goods, pointing to research by Erik Brynjolfsson and others on adding survey based welfare to GDP. Cowen said living to 120 would lift GDP in complex ways but add far more in valued years outside it. Neither answered Cowen’s own challenge from the Bay Area: what other sectors will grow fast enough to pay the revenue numbers being claimed for the frontier labs?

Mentioned in this story

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsTyler CowenAlex ImasAnthropicOpenAINvidiaGoogle DeepMindMETRa16z

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.