For the past several years, the story of artificial intelligence has largely been told through the lens of the models. But the real power in the AI era was never going to be decided by whose chatbot writes better poetry. It was always going to be decided by whoever controls the infrastructure underneath all of it. The chips. The data centers. The cloud computing contracts that determine which models get built, how fast they get built, and who gets to build them. And on that front, Google has spent the past week making moves that suggest it understands exactly where the real competition is being fought.
The first move was a multi-billion dollar deal with Thinking Machines Lab, the AI startup founded by Mira Murati after she left her position as chief technology officer at OpenAI in one of the most closely watched departures in the history of the technology industry. The deal, valued in the single digit billions according to sources familiar with the matter, gives Thinking Machines Lab access to Google Cloud’s AI infrastructure, including systems powered by Nvidia’s latest GB300 chips alongside a full suite of cloud services covering storage, databases and deployment. It is Google’s latest in a string of aggressive moves to lock in fast-growing frontier AI labs before its competitors can get to them. Anthropic signed a deal with Google and Broadcom for multiple gigawatts of tensor processing unit capacity this month, and separately secured up to five gigawatts of capacity from Amazon. The race to secure compute is happening at a scale and speed that would have seemed extraordinary just two years ago.
But the more significant long-term move may be the one that received less immediate attention. Google is developing its own new AI chips specifically designed for inference, the process by which AI models actually generate their responses after they have been trained, and those chips are being positioned as a direct challenge to Nvidia’s near-total dominance of the AI hardware market. This matters enormously because inference is where the real ongoing cost of running AI lives. Training a model is expensive but it happens once. Inference happens billions of times a day across every AI product in existence, and whoever makes the most efficient chips for that process stands to capture an almost incomprehensible share of the value being created in this industry. Nvidia has owned that space for years. Google is now signaling that it intends to compete for it seriously.
The context for all of this is an AI investment cycle that has refused to slow down despite everything the world has thrown at it in recent months. A war in the Middle East, an oil price shock, a broad market selloff in March, none of it has dented the capital commitments flowing into AI infrastructure. S&P 500 tech sector earnings are expected to significantly outpace the rest of the market in the first quarter of 2026. OpenAI has surpassed $25 billion in annualized revenue and is reportedly eyeing a public listing as soon as late this year. Anthropic is approaching $19 billion. The money is not slowing down. If anything it is accelerating, and Google’s moves this week are a clear signal that the company intends to be at the center of where it all flows.
What is emerging is a picture of an AI industry that is consolidating around a small number of very large players with the infrastructure, the capital and the cloud relationships to shape what gets built and who gets to build it. Nvidia has been the undisputed king of that world for the past three years. Google is not trying to dethrone it overnight. It is doing something more patient and potentially more dangerous. It is making itself impossible to avoid. Every frontier lab that signs a cloud deal with Google, every chip that runs on Google infrastructure, every model that deploys through Google’s stack is another thread in a web that is being woven quietly and deliberately around the entire industry. By the time anyone decides they want out, they may find that Google is not just a vendor. It is the architecture.
Sources: TechCrunch, CNBC, Al Jazeera
