Demis Hassabis is one of the godfathers of modern AI — the man who gave the world DeepMind, and created some of the algorithmic underpinnings of the present wave of pre-trained transformers. In the past four years, he has led Google’s remarkable turnaround in the space, going from a midweight AI to one of the leaders by computation if not yet by market scale. When he speaks, many listen.

So when Hassabis sat down with the Financial Times for an interview published over the weekend, he was posed the usual first question of intelligent journalists: is it a bubble? What he had to say was mostly practiced blandishments: that companies with no visible product, backed by venture capital, suggest a market that is somewhat overheating — but this does not apply to the big players.

Then he gave the game away: “Of course, Google has deep enough pockets to ride this one out.” That could be the sharpest call on the coming market structure for the post-bubble world.

As talk rumbles on about overheating, OpenAI has begun to work up its ad strategy, even though CEO Sam Altman said he was totally opposed to the idea just a year ago. Elsewhere, as the processor crunch bites, Amazon is suing the Oregon state energy board for withholding power that the state protests it needs to keep the lights on. Increasingly, there are signs that the crazed railway-building era of AI will end in much blood on the carpet.

Hassabis is pointing towards what comes next. What if the reach of the challenger brands outruns their grasp, and they are forced into cut ‘n’ shut mergers? If the ad revenue model holds together for the next couple of years, Google — holding reserves of cash seldom seen in human history — could be the last man standing in a winner-takes-all world.

In that space, OpenAI ends up as AltaVista, Anthropic as Yahoo. Google could continue to run its systems at the kinds of $5 billion-a-year losses that OpenAI is clocking up. Sweep a few of the peripheral players into its net, and either expand or shutter them.

But it might not result in a clean monopoly. Duopoly is also a common feature of tech. Stood in the background is another titan, flush with cash, which predates the web itself. Microsoft already managed to engineer a stake in OpenAI, during the failed anti-Altman putsch of late 2023. If the company were to outrun the cash reserves its soaring stock valuation has spun for it, Microsoft might in turn be happy to help OpenAI become CoPilot Chat.

That outcome would be dismal for many. Economists will tell you that monopolies are seldom as permanent as they appear. Of the top hundred companies on the NYSE in 1926, few survive. But here, because of the super-normal profit structures, even bold challenges to the ancien regime, backed by gimlet-eyed investors, end up recycling the same capital.

The last decade produced few new challengers, bar the Chinese semi-government entity TikTok. In fact, it has chiefly meant the purchase of challengers. Google bought YouTube and Waze; Facebook took over WhatsApp and Instagram; never mind Microsoft’s infamous shutdowns of early rivals such as Visio, the Mosaic web browser, and Skype.

As the infrastructure grid expands and matures, it may become cheap enough to produce and run AI. But if they don’t fall quickly enough, then Google and Microsoft stand ready to burn holes in their own pockets, on the assumption that something this valuable must turn a profit eventually. As Hassabis implied, we may have to wait years for that.

---

Gavin Haynes is a journalist and former editor-at-large at Vice.

@gavhaynes

---