If it seems like Google Cloud is ubiquitous in AI, that’s because they are.
Amid competitors that have been touting mega-alliances and headline-making hardware commitments, Google Cloud seems to be flooding the zone from below, weaving a thick tapestry of relationships with the startups most likely to grow into tomorrow’s platform companies.
Inside the Startup Land Grab at Google Cloud
Within Google Cloud, the premise is straightforward: win companies that do not yet exist at scale. Approximately 60% of generative AI startups around the world opt to use Google’s cloud, while nine of the top 10 artificial intelligence labs operate on Google infrastructure, said Chief Operating Officer Francis deSouza. He also highlights nearly every gen AI unicorn running on Google Cloud and $58B in new revenue commitments booked over the next two years. The numbers are meant to be a message — this is where the next wave is forming.
The playbook combines a lavish bailout with technical gravity. Early-stage AI companies can access up to $350,000 in cloud credits, hands-on access to engineering teams, and go-to-market support through the marketplace. Rather than expensive equity arrangements, Google is inking deals as the “primary computing” supplier to up-and-coming names like Loveable and Windsurf, betting that speed and proximity matter more than splashy checks when workloads get an overnight pace boost.
DeSouza’s background informs the push. As Illumina’s CEO, he saw machine learning reconfigure drug discovery, then co-founded Synth Labs to tackle AI alignment. That blend of enterprise scale and frontier research leaves him unusually fluent in what startups need on day one, and what they break when they hit month twelve.
Playing Switzerland While Selling Chips
Google is threading a tricky needle, providing infrastructure to the world’s top AI researchers while building more and more of its own capabilities that could be turned into a threat. Anthropic’s Claude is based on Vertex AI, OpenAI has employed Google’s TPU chips, and Google’s own Gemini family directly competes with both. Alphabet’s minority stake in Anthropic — a detail described in court records and reported by The New York Times — points to this new world of tech trends we happen to live in, where “coopetition” increasingly characterizes the AI stack.
To allay worries about lock-in, Google leans heavily on an open posture. Its history with Kubernetes and that seminal “Attention Is All You Need” paper is well documented. More recently, a protocol from agent to agent for inter-agent communication was published by the researchers. The logic is clear: openness at every layer — even if it enables customers to build rivals at the top of Google’s stack — grows the whole pie and keeps developers near.
Infrastructure Bets Outside Google’s Walls
The flood is more than just customers; it’s also silicon and geography. New reporting from The Information suggests that Google has begun plugging TPU systems at other providers’ data centers, a deal with London-based Fluidstack that comes with as much as $3.2 billion in support for a New York site. If true, it shows a willingness to pursue compute demand where and as it happens — and that TPUs are first-class citizens in multi-cloud strategies.
That comes on the heels of unprecedented consolidation at the highest reaches of the market for AI infrastructure. The chip maker Nvidia and the artificial intelligence company OpenAI have created a $100 billion partnership out of thin air, with much of it focused on chip supply and compute. Go deeper: Microsoft’s investment in OpenAI has grown to around $14 billion. Amazon has also put $8 billion into Anthropic to optimize workloads across its stack. Oracle signed a $30 billion cloud pact with OpenAI — and later locked in a five-year contract that would be worth $300 billion starting from 2027. Even Meta, which constructs a lot of its own infra, has apparently inked a $10 billion deal with Google Cloud. A separate multibillion-dollar “Stargate” idea involving SoftBank, OpenAI and Oracle has also been making the rounds. Against that landscape, Google’s startup-first embrace seems less like a consolation prize and more like a long-tail power grab.
Regulatory Optics With Competitive Upside
Regulators are paying attention to how the power of dominant platforms is being expanded with AI. A U.S. district judge’s recent ruling in the long-running search case underscored just how sensitive regulators have become about using search scale to get a leg up in adjacent markets. Google’s response in cloud has been to accentuate choice — many models in a marketplace, portable tooling, and universal infrastructure — while courting startup ecosystems that can credibly assert independence.
There’s also a substance pitch behind the optics. DeSouza contends that AI at the cloud scale can help speed research on difficult-to-crack problems in health-care and climate. The company cites work in genomics data analysis, modeling neurodegenerative diseases, and emissions forecasting as areas where access to specialized chips, large datasets, and model tooling can cut years off of discovery timelines.
What to Watch as the AI Infrastructure Zone Fills Up
Three metrics will tell us if the strategy is sticking.
- Conversion: How many credit-laden pilots convert to long-lasting, high-margin workloads.
- Cost curves: Do TPUs provide an economically meaningful advantage for both training and inference versus Nvidia’s offering as models scale?
- Portability: How many startups will select Google as their control platform in a multi-cloud world (and not just overflow capacity).
The gravity of the AI infrastructure center is still moving, but one thing seems clear. As the titans race to sign eye-popping checks, Google Cloud is flooding the field with chips, credits, and choice. If only a sliver of the early bets convert into the next wave of AI platforms, today’s startup explosion may be tomorrow’s dominant market share.
