The $100 billion partnership between Nvidia and Openai, announced Monday, represents the latest megadeal reshaping the landscape of AI infrastructure for now. The deal includes non-voting stocks related to large-scale chip purchases and ample computing power for over 5 million US households, deepening the relationship between the two of AI’s most powerful players.
Meanwhile, Google Cloud has a completely different bet. The industry’s biggest players cement more Titer partnerships, but Google is plagued by acquiring the next generation of AI companies before it grows into court.
Its COO, Francis Desouza, has seen the AI revolution from multiple vantage points. As former CEO of Genomics Giant Illumina, he saw discoveries like machine learning transformation. As a co-founder of Synth Labs, an AI alignment startup two years ago, he has been tackling the safety challenges of increasingly powerful models. He currently joins C-Suite on Google Cloud in January and organizes a massive bet in the second wave of AI.
Desouza is a story I like to talk about in numbers. In a conversation with the editor earlier this week, he pointed out several times that nine of the top 10 AI labs use Google’s infrastructure. He also said that almost all generative AI unicorns are running on Google Cloud, with 60% of all GEN AI startups worldwide choosing Google as their cloud provider, and the company has lined up a new revenue commitment of $58 billion over the next two years.
Asked that Google Cloud’s revenue percentage comes from AI companies, it offers instead that “AI has reset the cloud market and Google Cloud is especially leading the way in startups.”
The Nvidia-Openai transaction illustrates the scale of integration that spreads AI infrastructure. Microsoft’s original $1 billion Openai investment grew to nearly $14 billion, essentially restructuring the cloud market. Amazon has secured $8 billion in human investments and ensured deep hardware customization that essentially tunes AI training to work better on Amazon’s infrastructure. Oracle also appeared as a surprise winner, making $30 billion cloud deals with Openai, securing a five-year commitment of $300 billion from 2027.
Even Meta signed a $10 billion contract with Google Cloud, planning to spend $600 billion on US infrastructure until 2028, despite building its own infrastructure.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Given the partnership in which companies like Openai and Nvidia appear to be solidified elsewhere, these huge deals may seem threatening to Google. In fact, it appears that Google is being cut out of a frenzied deal.
But the corporate giant doesn’t exactly sit in its hands. Instead, Google Cloud signs small businesses like Lovable and Windsurf. This is what Desouza calls “the next generation of companies approaching.”
This approach reflects both opportunities and needs. As Desouza says, in a market where companies “turn from startups to billions of dollars,” a market where companies can capture future unicorns before maturity, they can prove to be more valuable than fighting today’s giants.
Strategies extend beyond simple customer acquisition. Google offers AI startups $350,000 in cloud credits, access to technology teams, and market support through the market. Google Cloud describes Desouza as an “uncompromising” AI stack (the “open spirit” that lets you choose your customers across chips, models, applications, and all layers).
“Companies love the fact that they have access to the AI stack. They can access their teams and understand where technology is heading,” Desouza said during an interview. “They also love the ability to access enterprise-grade Google-class infrastructure.”
The benefits of this infrastructure became even more clear this month when I reported that I was operating behind the scenes at Google to expand my custom AI chip business. According to the information, Google has attacked transactions for the first time deploying tensor processing units (TPUs) in data centers of other cloud providers, including an agreement with London-based Fluidstack, which includes up to $3.2 billion in financial assistance for its New York facility.
Competing directly with AI companies and providing infrastructure requires finesse. Google Cloud offers OpenaI a TPU chip and hosts Anthropic’s Claude model through the Vertex AI platform. (Google Cloud’s parent company Alphabet also owns 14% of humanity in each New York Times court document it obtained earlier this year, but when asked directly about Google’s financial relationships with humanity, Desouza calls it a “multi-layered partnership” and can be quickly replicated in Google Cloud’s “Model Garden.”
But if Google is trying to become Switzerland while moving on its own agenda, it’s doing a lot of practice. This approach has roots in Google’s open source contributions, from Kubernetes to basic “careful” papers that allow for the underlying trans-architecture of most modern AI. Recently, Google has published an open source protocol called Agents (A2A) from Agents for interagent communication to demonstrate its ongoing commitment to openness, even in the competitive field.
“We have explicitly chosen to be open in every layer of the stack over the years. We know that this means that companies can absolutely adopt technology and build competitors on the next layer,” admitted Desouza. “It’s been going on for decades. It’s okay for us.”
Startup courtship on Google Cloud comes at a particularly interesting moment. This month, federal judge Amit Mehta handed down a delicate ruling in a government-five year-old search monopoly case, sought to curb Google’s control without hampering AI ambitions.
Google avoided the Department of Justice’s strictest proposed penalties, including the forced sale of its Chrome browser, but the ruling underscored regulatory concerns about companies that are using search monopolies to control AI. Critics are worried that the vast vast number of Google’s search data will provide unfair advantages in developing AI systems and that the company will be able to deploy the same exclusive tactics that ensure search dominance.
In the conversation, Desouza focuses on far more positive outcomes. “I think today we have an opportunity to fundamentally understand some of the major diseases that we don’t fully understand,” Desouza said, for example, that Google Cloud outlined its vision to support power research in Alzheimer’s, Parkinson’s, and climate technology. “We want to work very hard to make sure we are pioneering the technology that makes the job possible.”
Critics can’t be easily tempered. By placing itself as an open platform that empowers rather than control the next generation of AI companies, Google Cloud may show regulators that it will promote competition rather than suppress competition, while building relationships with startups that could support Google’s case if regulators step up pressure.
Check out this week’s StrictlyVC Download Podcast for a full conversation with Desouza. New episodes are announced every Tuesday.
