Google is getting even closer to the companies building the physical machinery behind artificial intelligence. A new agreement with Marvell Technology could eventually give Google a multibillion-dollar ownership stake in the chipmaker while dramatically expanding Marvell’s role inside Google’s AI infrastructure. The arrangement is worth paying attention to for a reason that goes beyond the headline number.
Big Tech companies are no longer simply buying chips. They are designing them, financing suppliers, locking in capacity and, increasingly, blurring the line between customer and investor.
Google Could Become One of Marvell’s Largest Shareholders
Under the agreement, Marvell granted Google a warrant allowing it to acquire as many as 58.97 million Marvell shares at $206.58 per share.
If Google eventually exercises the full warrant, those shares would be valued at roughly $12.18 billion. The position could make Google Marvell’s fifth-largest investor, according to reporting cited by Competition Policy International / PYMNTS. But the stock warrant is tied to something much larger: custom silicon.
Marvell will work with Google on chips and supporting technologies used throughout Google’s artificial intelligence infrastructure. The potential commercial payoff for Marvell is enormous. If Google reaches the targets associated with the agreement, Marvell could generate around $120 billion in revenue through fiscal 2033. That is not the economics of an ordinary supplier contract.
Google Wants More Control Over the AI Hardware Stack
Google has spent years building its own Tensor Processing Units, better known as TPUs.
Those chips give the company an alternative to relying entirely on Nvidia GPUs for artificial intelligence workloads. They can also be optimized specifically for Google’s cloud infrastructure and AI models. The Marvell agreement reaches across several parts of that system.
According to the reporting, Marvell’s involvement includes processors used for AI inference as well as technology related to storage and networking. That matters because the AI chip race is no longer only about the accelerator sitting inside a server.
Memory. Networking. Data movement. Storage. Interconnects. Every one of those pieces becomes more important when data centers are running enormous clusters of AI processors. Google appears to be treating the whole stack as strategic infrastructure.
Broadcom Suddenly Has More Company
Broadcom has been one of Google’s most important custom-chip partners. Investors noticed the Marvell announcement immediately.
Broadcom shares fell more than 5% after the agreement became public, while Marvell climbed nearly 8%. Alphabet shares were comparatively steady. The immediate assumption might be that Marvell is replacing Broadcom.
That interpretation is probably too simple. Morningstar analyst William Kerwin told Reuters that the deal looked more like Google was expanding its supplier base rather than shifting its entire custom-chip business away from Broadcom.
For Google, having multiple major suppliers makes sense. AI infrastructure demand is exploding. Depending too heavily on one chip partner introduces cost, capacity and supply-chain risks that become harder to ignore as spending moves into the tens of billions of dollars.
The AI Supply Chain Is Becoming Financially Interconnected
The stranger part of the deal is not that Google wants custom chips. It is the ownership structure. Google could simultaneously become one of Marvell’s biggest customers and one of its biggest shareholders.
That creates a relationship considerably more complicated than buyer and seller. And Google is hardly alone. AI companies have been experimenting with increasingly elaborate arrangements involving chip purchases, equity options, infrastructure financing and long-term capacity commitments.
Competition Policy International points to OpenAI’s agreement with AMD, which included an option for OpenAI to acquire roughly 10% of the chipmaker. Nvidia has also participated in major financing arrangements connected to AI data-center projects.
That broader financing shift is already visible in massive AI infrastructure projects such as Oracle’s Project Jupiter, where computing, energy and long-term capital are becoming tightly connected.
- The money is beginning to move in circles.
- Chipmakers supply AI companies.
- AI companies invest in chipmakers.
Infrastructure providers finance data centers used by the same companies buying their hardware.
That does not automatically mean something anti-competitive is happening. It does mean regulators are getting a much messier market to examine.
Regulators May Eventually Have Questions
There is currently no indication that competition regulators have opened an antitrust investigation specifically into the Google-Marvell agreement.
Still, the structure illustrates an emerging problem. If the largest AI infrastructure buyers also hold meaningful financial stakes in the companies supplying them, traditional market relationships become harder to separate.
One argument is that these arrangements increase competition. Marvell gaining more Google business could create a stronger alternative to Broadcom. Custom AI chips could also reduce Google’s dependence on Nvidia. The other side is less comfortable.
When customers, suppliers and investors become financially tied together, incentives can change. Companies that would normally compete independently may have reasons to protect relationships that are worth billions of dollars. Competition authorities are likely to spend much more time untangling that question as AI infrastructure expands.
Custom Chips Are Becoming Central to Google’s AI Strategy
The Google-Marvell AI chip deal also says something about where Google believes the AI market is heading. Running cutting-edge models at massive scale is brutally expensive. The biggest AI companies are therefore looking for every possible advantage: cheaper inference, specialized processors, better networking and tighter integration between software and hardware.
Google happens to have something most AI companies do not. It owns the cloud platform, develops major AI models, designs its own processors and operates enormous data centers.
That vertical integration also supports Google’s increasingly aggressive model roadmap, including its push toward faster and cheaper Gemini models for coding and agentic workloads. Marvell gives Google another partner inside that vertical stack.
For Marvell, meanwhile, landing a deeper role in Google’s TPU ecosystem could reshape the company. Potential revenue approaching $120 billion through 2033 would make Google’s AI expansion one of the biggest commercial opportunities in Marvell’s history.
The Bigger AI Chip Battle Is Starting to Look Different
Nvidia still dominates the conversation around AI computing. But underneath that dominance, hyperscalers are quietly building alternatives. Google has TPUs. Amazon has Trainium and Inferentia. Microsoft has developed its own AI accelerators. Meta is working on custom silicon too. The endgame may not be a single company replacing Nvidia.
It could be something less dramatic and more consequential: the largest technology companies gradually controlling more of their own AI hardware. That shift is happening alongside Nvidia’s own expansion beyond chips into AI models, agent software and local inference infrastructure.
Google’s expanded partnership with Marvell fits neatly into that picture. This deal is about chips, yes. It is also about leverage, supply security and who controls the infrastructure underneath the next generation of artificial intelligence. And $12.2 billion is a fairly loud way of making that point.
