The chain of partnerships keeps on increasing. Is this a collaboration or a consolidation of power?
The surface reading of the Google-Meta chip deal is straightforward. Meta has agreed to rent Google’s tensor processing units through Google Cloud to train and run its next-generation large language models, in a multi-year agreement worth billions of dollars. Google gets a major enterprise customer. Meta gets another compute supplier. Clean transaction. Except nothing about this is clean.
Meta’s AI infrastructure spending is projected to reach $135 billion in 2026. The company has 30 data centres planned, 26 of them in the United States. That is not a company with a compute problem. That is a company executing a deliberate strategy to ensure no single supplier can hold it hostage. This week alone, Meta has deals running simultaneously with Nvidia, AMD, and now Google. Morningstar analysts are calling it a multipronged silicon strategy. We would call it something blunter: leverage, bought in advance and at scale.
The timing is not incidental. Meta has been running into serious problems with the AI chips it is designing internally, scrapping its most advanced in-house training chip last week. When your own silicon program hits a wall, you move fast, and you move wide. The Google deal is partly a diversification. It is also insurance.
For Google, the stakes are different but equally structural. Google first developed TPUs more than a decade ago for internal workloads. The Meta deal represents a major expansion of its TPU commercialisation strategy, which previously kept the chips largely inside Google’s own infrastructure. Google is now forming a joint venture to lease TPUs to other AI customers, with some Cloud executives estimating the expansion could capture as much as 10 percent of Nvidia’s annual revenue. In October, Anthropic signed a deal for access to up to one million TPUs. Meta follows. The pattern is becoming a market.
So is this a consolidation of power, an expansion of it, or something that resembles creative collaboration? We think the honest answer is: it is none of those things individually. It is two dominant players using each other to reduce their dependence on a third dominant player. NVIDIA sits at the centre of the AI race in a way that makes every other company in the ecosystem uncomfortable, and the deals being signed this week are the market’s response to that discomfort.
The partnership economy does not always produce partners. Sometimes it produces mutual hedges dressed in press release language. This is one of those.
