The $100 billion alliance between Nvidia and OpenAI, revealed on Monday, marks the most recent major transaction transforming the AI infrastructure sector. This arrangement features non-voting equity linked to significant chip acquisitions and provides enough computational resources to support over 5 million American homes, further strengthening the collaboration between two of the most influential entities in artificial intelligence.
At the same time, Google Cloud is pursuing a distinctly different strategy. While industry giants are forging increasingly close alliances, Google Cloud is determined to attract the next wave of AI startups before they grow too large to engage.
Francis deSouza, the company’s COO, has observed the AI transformation from several perspectives. As Illumina’s former CEO, he witnessed machine learning revolutionize drug development. As a co-founder of Synth Labs, an AI alignment startup launched two years ago, he tackled the safety issues posed by advanced models. Now, after joining Google Cloud’s executive team in January, he is leading a significant bet on the next phase of AI.
deSouza often illustrates his points with statistics. In a discussion with this editor, he repeatedly highlights that nine of the top ten AI labs rely on Google’s infrastructure. He also mentions that nearly every generative AI unicorn operates on Google Cloud, that 60% of generative AI startups globally have selected Google as their cloud provider, and that the company has secured $58 billion in new revenue commitments for the next two years—more than twice its current annual revenue pace.
When asked about the proportion of Google Cloud’s revenue derived from AI firms, he responds instead that “AI is transforming the cloud industry, and Google Cloud is at the forefront, particularly with startups.”
The Nvidia-OpenAI partnership highlights the extent of consolidation occurring in AI infrastructure. Microsoft’s initial $1 billion investment in OpenAI has ballooned to nearly $14 billion. Amazon has invested $8 billion in Anthropic, securing extensive hardware customizations that optimize AI training for Amazon’s systems. Oracle has also become a notable player, landing a $30 billion cloud contract with OpenAI and later obtaining an astonishing $300 billion, five-year commitment beginning in 2027.
Even Meta, despite developing its own infrastructure, entered a $10 billion agreement with Google Cloud and is planning $600 billion in U.S. infrastructure investments through 2028. The Trump administration’s $500 billion “Stargate” initiative, which includes SoftBank, OpenAI, and Oracle, adds yet another dimension to these interconnected deals.
These enormous agreements might appear to sideline Google, as companies like OpenAI and Nvidia seem to be forming exclusive partnerships elsewhere. It can seem as though Google is being left out of the current wave of high-stakes deals.
The Google logo is displayed during a meeting between Alphabet and Google CEO Sundar Pichai and Polish Prime Minister Donald Tusk at Google for Startups in Warsaw, Poland, on February 13, 2025. (Photo by Klaudia Radecka/NurPhoto via Getty Images)
Image Credits:Klaudia Radecka/NurPhoto / Getty Images
However, the tech giant is far from passive. Instead, Google Cloud is partnering with smaller firms such as Loveable and Windsurf—what deSouza refers to as the “up-and-coming generation of companies”—as their main computing partners, without requiring large initial investments.
This strategy is driven by both necessity and opportunity. In a landscape where, as deSouza notes, companies can “grow from startups to multi-billion dollar enterprises in a remarkably short time,” securing relationships with future unicorns before they mature could be more advantageous than competing for today’s industry leaders.
Google’s approach goes beyond simply acquiring new customers. The company provides AI startups with $350,000 in cloud credits, access to its engineering teams, and support for bringing products to market via its marketplace. Google Cloud also delivers what deSouza calls a “no-compromise” AI stack—from hardware to models to applications—embracing an “open philosophy” that allows clients flexibility at every level.
“Companies appreciate that they can utilize our AI stack and consult with our teams to learn about our technology roadmap,” deSouza explains in our interview. “They also value having access to Google’s enterprise-grade infrastructure.”
Google’s ambitions in infrastructure have grown recently, with reports indicating the company is expanding its custom AI chip business behind the scenes. According to The Information, Google has reached agreements to install its tensor processing units (TPUs) in other cloud providers’ data centers for the first time, including a deal with London-based Fluidstack that involves up to $3.2 billion in funding for a New York data center.
Balancing direct competition with AI companies while also supplying them with infrastructure is a delicate act. Google Cloud supplies TPU chips to OpenAI and hosts Anthropic’s Claude model on its Vertex AI platform, even as its own Gemini models compete directly with both. (Alphabet, Google Cloud’s parent company, reportedly holds a 14% stake in Anthropic, according to New York Times court documents from earlier this year. When asked about Google’s financial ties to Anthropic, deSouza describes it as a “multi-layered partnership” and quickly shifts focus to Google Cloud’s model marketplace, highlighting that customers can access a variety of foundational models.)
If Google is aiming to be a neutral party while advancing its interests, it has plenty of experience. This strategy is rooted in Google’s history of open-source contributions, from Kubernetes to the influential “Attention is All You Need” paper that introduced the transformer architecture now central to modern AI. More recently, Google released an open-source protocol called Agent-to-Agent (A2A) for inter-agent communication, aiming to reinforce its commitment to openness even in competitive fields.
“We have deliberately chosen to be open at every layer of our stack, fully aware that this enables others to use our technology to build competing products,” deSouza admits. “This has been happening for decades, and we’re comfortable with that.”
Google Cloud’s outreach to startups comes at a particularly notable time. Just this month, federal judge Amit Mehta issued a nuanced decision in the government’s long-running search monopoly case, seeking to limit Google’s dominance without hindering its AI pursuits.
Although Google avoided the harshest penalties proposed by the Justice Department, such as being forced to sell its Chrome browser, the ruling highlighted ongoing regulatory concerns about the company using its search monopoly to gain an edge in AI. Critics worry that Google’s vast search data gives it an unfair advantage in developing AI technologies and that it could use similar tactics to those that established its search dominance.
In conversation, deSouza prefers to focus on the positive possibilities. “I believe we have a real chance to gain a deep understanding of major diseases that currently remain mysterious,” he says, envisioning Google Cloud supporting research into Alzheimer’s, Parkinson’s, and climate solutions. “We are committed to pioneering the technologies that will make this research possible.”
Skeptics may remain unconvinced. By positioning itself as an open platform that empowers rather than controls the next wave of AI companies, Google Cloud could be signaling to regulators that it encourages competition, all while building relationships with startups that may benefit Google if regulatory scrutiny intensifies.
To hear our complete conversation with deSouza, listen to this week’s StrictlyVC Download podcast; new episodes are released every Tuesday.
