Early Facebook backer Roger McNamee: AI's existential threat is fiction — the financial bubble is real

Four years after ChatGPT's launch, Roger McNamee — an early investor in both Google and Facebook — has a pointed message for the AI industry: stop worrying about killer robots and start worrying about the bond market.

In a new essay published by The Nerve, McNamee argues that the warnings of AI 'existential risk' that have dominated headlines since 2022 were never really about safety. They were a marketing strategy. According to McNamee, the industry invoked existential risk to suggest AI was unstoppable and to shield its creators from accountability for harmful outcomes. The real danger, he contends, is a financial bubble of scale — one already spreading its risk into pension funds and insurance portfolios.

The entire AI project has depended on the theft of copyrighted materials and private data. The latter is particularly troubling in the context of national security, where government data and secrets can become training data.— Roger McNamee, tech investor and author

The timing of McNamee's essay is pointed: Anthropic has filed paperwork for an initial public offering that could target a valuation exceeding $2 trillion. McNamee is blunt about what that represents. In his telling, the IPO is designed to let early investors exit rather than to generate wealth for those buying in now, writing that it is an opportunity for existing investors to cash out, rather than for new investors to get rich.

McNamee's financial critique rests on a stark mismatch between spending and returns. According to Epoch.ai figures cited in his earlier Guardian essay, total sales of large language model products to the public by market leaders OpenAI, Google, and Anthropic amounted to roughly $1 billion in 2023, growing to about $4 billion in 2024. Against that, the industry had already committed approximately $1.4 trillion in infrastructure investment, with an additional $3 trillion or more in future obligations — spending that McNamee says was rushed to market before innovation could drive down the cost of the underlying hardware.

In the internet era, the cost of infrastructure had declined by nearly three orders of magnitude by the time Google built its search infrastructure.— Roger McNamee, tech investor and author

He draws a direct historical comparison: when Yahoo commercialized the early web, Sun Microsystems servers cost roughly 10% of what the preceding Silicon Graphics supercomputers had. When Google scaled up, it built on bare Intel motherboards at a fraction of that again. The AI industry, McNamee argues, skipped that cost-reduction curve entirely — locking in debt at peak infrastructure prices in a race for speed.

The competitive structure of the industry compounds the problem. In his Guardian essay, McNamee noted that there are five or six essentially identical large language model programs running simultaneously among US big tech players — Google, Amazon, Meta, xAI, and Microsoft/OpenAI — each apparently seeking a global monopoly. Citing the principle associated with former GE chief Jack Welch, McNamee argues that only two players can ultimately be profitable in a competitive industry. That means at least three major players face the prospect of writing off investments that could reach $200 billion each by next year.

The day may come sooner than many expect when shareholders, directors and executives will demand evidence that the massive investment in LLM technology will generate an adequate return for them. The answer will be no for many, if not most, players.— Roger McNamee, tech investor and author

Beyond the investment math, McNamee raises a structural concern about the technology itself. Large language models, he writes, are 'inherently probabilistic,' producing varying levels of errors and non-predictive results — a flaw he calls unfixable at the foundational level. The industry's original thesis — that scaling computing power and training data would overcome these shortcomings — has 'hit a speed bump,' forcing a search for new approaches such as neurosymbolic technology. At the same time, enterprise customers surveyed on the subject report that the productivity improvements AI was supposed to deliver have been slow to arrive.

He also challenges the go-to-market strategy. Silicon Valley's most successful platforms — Google Search, Amazon, eBay — succeeded by solving a single, narrow problem immediately. General-purpose AI tools, by contrast, require customers to invest heavily in integration before seeing any return, and few enterprises have the capacity or funding to do that.

McNamee's concerns about AI's immediate, tangible harms extend beyond finance. As examples of current AI products failing basic consumer safety and security standards, he cites ChatGPT's involvement in an elementary school shooting in British Columbia, chatbots connected to suicides, research indicating that chatbot use impairs cognition especially among young people, and recent security breaches affecting Hugging Face and Australian government agencies.

The systemic financial risk McNamee describes is not confined to tech investors. He warns that risky AI-related debt is being repackaged and sold to pension funds and insurers — a mechanism that, if the bubble bursts, could spread losses well beyond Silicon Valley into the savings of ordinary people.

McNamee is careful to say he is not dismissing AI's long-term potential. He expresses a personal belief that the technology 'may one day be transformative,' while arguing that today's products fall well short of that promise, that the financial commitments already made have set an 'impossibly high' hurdle for success, and that the scale of the problem now threatens not just individual companies but the broader economy. With Anthropic's IPO on the horizon and industry investment forecasts pointing toward $1.48 trillion in 2025 alone, he believes the reckoning is coming faster than most investors expect.

Why it matters — If McNamee is right, the AI industry's debt-financed infrastructure binge — with revenues a fraction of commitments — could trigger write-offs and financial contagion reaching pension funds and insurers, making this a risk for far more than tech investors.

Reported by thenerve.news, theguardian.com, newsnationnow.com, bloomberg.com, webpronews.com