Karpathy says AGI is a decade away — markets barely flinch but tech Twitter melts down

Summary

Andrej Karpathy, a co-founder of OpenAI and former head of Tesla's self-driving program, said on a recent podcast that AGI is likely a decade away. Markets barely moved. Amazon stock rose 1.3%, partly attributed to renewed interest in AWS, while Oracle fell nearly 5%. On X and in tech circles, the reaction was visceral: panic that the AI bubble had popped, calls to rotate into "food, water, shelter, and guns."

Karpathy's ten-year timeline actually aligns with what Sam Altman said in a September 2024 blog post about the "intelligence age." When Altman mentioned superintelligence arriving in "a few thousand days" (roughly ten years), investors cheered rather than dumped holdings. The difference is narrative framing. When a well-funded CEO pitches AGI-as-imminent, capital flows. When a respected researcher outside the funding cycle describes incremental progress, traders panic.

Karpathy's actual argument is more subtle than the headline. He doesn't claim AI progress is slowing. He argues that even if AGI arrives in a decade, GDP growth will likely stay flat at 2% because AI is fundamentally continuous with every other technology—the internet, electricity, railways. Radiologists won't be replaced by AI because the job involves 25 things beyond image recognition. Recursive self-improvement is just researchers using Google, repackaged. The productivity gains from AI are real but not civilization-altering on a near-term basis.

No major AI lab publicly pushed back on the interview. The silence suggests either agreement or a strategic decision to avoid amplifying bearish framing. One researcher comment that surfaced: "RL is pretty mid," acknowledging that reinforcement learning, the current best tool available, remains limited.

Infrastructure bets

Oracle is now on the hook for roughly $300 billion in infrastructure over five years, betting that OpenAI revenue will justify the spend. Co-CEO Clay McGork framed this as Oracle simply mirroring the Google and Amazon playbook from the late 1990s—fast growth, proportional investment. The actual bet is straightforward: OpenAI becomes a multi-trillion-dollar company, or Oracle has massively overspent.

Commerce partnerships

OpenAI has announced partnerships with Etsy and Walmart, notably not Amazon or eBay. The tier-one platforms appear cautious, either still holding leverage or reluctant to cede it yet. But leverage flips if OpenAI becomes a true aggregator. Just as Google pays Apple to be the default search engine, Amazon and eBay may eventually have to partner if OpenAI controls a plurality of user demand. The timeline for that capitulation is probably two to five years.

User growth plateauing

ChatGPT's global daily active users are around 72 million, according to app-tracking data from Apptopia. Growth appeared to plateau in September and October after a period of consistent expansion. The broader question is whether this reflects saturation at scale or a signal that adoption has hit a ceiling. Similar Web data shows Google's Gemini gaining market share against OpenAI, though both are growing in absolute terms. Google has nine products with over 1 billion users but struggles to funnel users from one to another without acquisition or aggressive paid promotion. The challenge for Google is surfacing Gemini as a distinct product without cannibalizing search or looking desperate.

The real story

Public equity investors see infrastructure spending and user growth and treat it as normal hyperscaler scaling. X users see Karpathy's ten-year timeline and interpret it as a death knell for returns. Neither group is fully wrong. AI is genuinely useful and improving, which justifies some capital expenditure. But current spending of $10 to $15 to generate $1 in near-term revenue only works if breakthroughs arrive faster than current physics and data suggest. Francois Chollet noted that over $1 trillion is now committed to this bet. It only makes sense if technology improves dramatically within a three-to-five-year window before data centers depreciate. That's the real deadline, not any researcher's AGI prediction.