A staggering $100 billion partnership between Nvidia and OpenAI, revealed on Monday, marks one of the largest infrastructure deals in the AI world to date. The agreement, involving massive chip purchases and non-voting equity, promises computing power equal to what’s needed by over five million U.S. households. It also deepens the collaboration between two of the most dominant forces in artificial intelligence.
But while major players like Microsoft, Amazon, and Oracle are doubling down with billion-dollar alliances, Google Cloud is charting its own course—targeting emerging AI startups before they become too large to attract.
Francis deSouza, Chief Operating Officer at Google Cloud, brings a unique perspective to this strategy. Formerly CEO of Illumina, a genomics leader, he’s seen firsthand how machine learning revolutionized biotech. As co-founder of Synth Labs, a startup focused on AI alignment and safety, he’s also been close to the ethical challenges of powerful models. Now, in his new role at Google Cloud, he’s leading a bet not just on AI’s present—but on its future.
DeSouza prefers to frame Google Cloud’s AI push with data. In a recent conversation, he emphasized that 9 of the top 10 global AI labs are building on Google’s infrastructure. Nearly every generative AI unicorn is hosted on Google Cloud. And 60% of generative AI startups worldwide have chosen it as their provider. Looking ahead, Google Cloud has secured $58 billion in forward revenue commitments—more than double its current run rate.
When asked how much of that revenue comes from AI customers, deSouza offers a broader view: “AI is redefining the entire cloud market, and we’re leading the shift—especially among startups.”
Meanwhile, the Nvidia-OpenAI partnership underscores the escalating consolidation of the AI infrastructure space. Microsoft’s original $1 billion investment in OpenAI has grown to $14 billion. Amazon has poured $8 billion into Anthropic, securing custom hardware tailored to its infrastructure. Oracle, a relative newcomer to AI leadership, recently closed a $30 billion deal with OpenAI and is preparing for a $300 billion, five-year cloud commitment starting in 2027.
Even Meta, despite developing much of its infrastructure internally, signed a $10 billion cloud agreement with Google and is planning to invest $600 billion in U.S. infrastructure by 2028. Adding to this complex web is the U.S. government-backed “Stargate” initiative, a $500 billion project involving OpenAI, Oracle, and SoftBank.
Against this backdrop, it might appear that Google is being left out of the AI infrastructure land grab. But rather than chasing headline-making deals, Google Cloud is targeting rising startups—like Loveable and Windsurf—as their primary computing partner, often without needing large upfront investments.
According to deSouza, this isn’t just opportunistic—it’s essential. In today’s AI climate, a startup can go from stealth mode to multibillion-dollar valuation in a matter of months. Catching them early could be more valuable than competing for today’s big names.
And it’s not just about infrastructure. Google Cloud is building relationships with these companies by offering up to $350,000 in credits, access to engineering teams, and marketing support via its marketplace. It’s all part of what deSouza calls a “no-compromise AI stack”—including custom chips, foundation models, tools, and applications—delivered with an “open ethos” that gives customers flexibility at every layer.
“Founders value having direct access to our teams and roadmap,” deSouza explains. “They also appreciate that they’re building on enterprise-grade, globally reliable infrastructure.”
Google’s ambitions don’t stop there. The company has quietly expanded its custom chip strategy, placing its TPU (Tensor Processing Unit) chips into third-party data centers for the first time. One such deal, with London-based Fluidstack, includes up to $3.2 billion in support for a data center project in New York.
This dual strategy—competing with AI companies while also powering their infrastructure—requires careful balance. Google Cloud provides TPUs to OpenAI, hosts Anthropic’s Claude models on its Vertex AI platform, and at the same time develops its own Gemini models, which directly rival both. Alphabet, Google’s parent company, also holds a 14% stake in Anthropic.
When asked about this complex relationship, deSouza describes it as a “multi-layered partnership,” quickly redirecting the focus to Google’s open model marketplace, where customers can choose from a range of foundation models—including competitors’.
This balancing act is familiar territory for Google. Its open-source legacy—from Kubernetes to the seminal “Attention Is All You Need” paper that enabled today’s transformer-based AI—has long supported an ecosystem where competitors and collaborators often coexist. Recently, Google published an open-source protocol called Agent-to-Agent (A2A), designed for inter-agent communication—another sign of its ongoing commitment to openness, even in a hyper-competitive landscape.
“We’ve deliberately chosen to stay open at every level of the stack,” says deSouza. “That means others can use our tech to build direct competitors—and that’s fine with us. It’s been happening for decades.”
Google Cloud’s aggressive support of startups comes at a particularly sensitive moment. Just weeks ago, a federal judge delivered a ruling in the government’s long-running antitrust case against Google. While the company avoided the harshest proposed remedies—like being forced to spin off Chrome—the ruling acknowledged concerns that Google’s search monopoly could give it an unfair edge in the AI race.
Regulators worry that Google’s treasure trove of user search data could supercharge its AI systems in ways that others simply can’t match—and that it may use the same playbook that secured its dominance in search to win in AI.
But deSouza sees potential for good. He envisions Google Cloud supporting research into some of the world’s biggest challenges: Alzheimer’s, Parkinson’s, and climate change among them.
“We believe we have the opportunity to unlock insights into diseases and technologies that humanity still doesn’t fully understand,” he says. “And we’re working hard to make sure our technology enables that.”
Still, skepticism persists. By positioning itself as an open, collaborative platform that helps startups rise—rather than control them—Google Cloud may be sending a strategic message to regulators: that it’s not stifling competition, but nurturing it. And those same startups might one day help Google make its case, should antitrust scrutiny intensify further.
Also Read:
Microsoft adds Anthropic’s AI to Copilot
RVAI Global Acquihires TYNYBAY to Develop Next-Generation Ai and Agentic Services Platform
