Google's AI boss admits that 'no moat' memo was real — but disagrees

A few months ago, a leaked memo sent shockwaves across the tech world by claiming that neither Google nor OpenAI had any real moat against open-source AI. Now, we finally have official confirmation that the document was real, straight from the top.

In a recent interview with Decoder, DeepMind CEO Demis Hassabis acknowledged that the memo was written by a Google engineer, though he strongly pushed back against its dire conclusions.

Validating the leak while denying the threat

Hassabis explained that engineers inside Google write various internal documents all the time, and occasionally one leaks and catches fire online.

While admitting it was legitimate, Hassabis made it clear he does not share the author's pessimism. Although he noted he had not read that specific memo in full detail, he openly disagreed with its central thesis that open source will ultimately outpace closed frontier labs.

The document in question, originally circulated by SemiAnalysis after being discovered on a public Discord server, argued that a third faction was quietly eating Google's lunch. The researcher claimed that open-source models were faster, more customizable, far more private, and pound-for-pound more capable.

Doubling down on frontier research

Hassabis is betting the house on his newly consolidated team. With the recent merger of Google Brain and Google DeepMind under his leadership, he believes the company's internal competitive drive and research talent will keep it ahead.

He pointed to the history of breakthrough innovations coming out of both teams over the past decade, arguing that the track record speaks for itself. According to Hassabis, he would bet on Google producing the next decisive breakthroughs over the coming ten years just as they did in the past.

Why it matters

Having Google's AI boss publicly address internal dissent shows just how serious the open-source pressure has become. For developers and founders, the memo confirmed what many already feel: open-source alternatives are becoming remarkably competent, highly customizable, and vastly more cost-effective for everyday applications without vendor lock-in.

For the broader industry, this pushes closed-source giants to prove their worth. If proprietary models cannot maintain a massive capability gap, charging premium subscription and API fees becomes a much harder sell.

We think Hassabis's response is classic DeepMind: confident in foundational science and sheer compute power. But banking on past architectural breakthroughs might overlook the sheer speed at which the global developer community optimizes existing models. The battle between centralized frontier research and decentralized open collaboration is only heating up.