$100 billion The partnership between Nvidia and Openai, announced on Monday, represents now, the last mega-deal that remodes the panorama of infrastructure AI. The interval includes actions of an agreement linked to enormous purchases of chips and enough calculation power for over 5 million US families, deepening the relationship between two of the most powerful AI players.

Meanwhile, Google Cloud is taking a completely different bet. While the greatest players in the sector are increasingly rigorous partnerships, Google Cloud is Hellbent in capturing the next generation of artificial intelligence companies before becoming too large or short.

Francis Desouza, his coo, saw the IA rehearsal from several points of view. As a form of the genomic giant illuminates, it has seen automatic learning transform the discovery of drugs. As a co-founder of a two-year alignment startup, Synth Labs has faced the safety challenges of increasingly powerful models. Now, after having joined the C-SIEVE to Google Cloud in January, he is orchestrating a large bet on the second wave of AI.

It is a story that you like to tell in number. In a conversation with this publisherNote several times that nine of the first 10 artificial intelligence laboratories use Google infrastructure. It also says that almost all the unicorns to the generative are on Google Cloud, that 60% of all the startups Gen all over the world has chosen the Google Asir Cloud Provider and that the company has lined $ 58 billion in new backgrounds in the next two years more than double the current annual execution rate.

To the question about which percentage of Google Cloud’s revenue come from artificial intelligence companies, it offers intratedas that “Ai is restoring the cloud market and Google Cloud is opening the road, in particular with startups”.

The NVIDIA-OPTAI agreement exemplifies the scope of the vast consolidation infrastructure. The original Openii investment of $ 1 billion in Microsoft grew up to almost $ 14 billion. Amazon has followed $ 8 billion in anthropic investments, ensuring deep hardware customizations that essentially customize training to work better with Amazon’s infrastructure. Oracle was also surprised by the winner, landing has $ 30 billion The cloud treats with Openi and therefore ensuring a stunning jaw $ 300 billion Commitment to five years starting from 2027.

The goal, despite having built its infrastructure, also signed a $ 10 billion With Google Cloud during the planning of $ 600 billion in the US infrastructure expenditure until 2028. The “Stargate” project of $ 500 billion in the Trump administration, which involves SoftBank, Openai and Oracle, adds another level to these interbloc partnerships.

These gigantic agreements may seem threatening for Google, given the partnerships that companies such as Openai and Nvidia seem to cement elsewhere. In fact, it seems that Google is cut by some frantic deal.

But the corporate giant is not exactly sitting on his hands. Instead, Google Cloud is signing sMaller companies As adorable and Windsurf – what Desouza calls the “next generation of companies arriving” – as “primary calculation partner” without important initial investments.

The approach reflects both opportunities and necessary. In a market where companies can go “from being a startup to being a multi-billary company in a very short period of time”, as Desouza says, capturing future unicorns before mature could prove more precious than fighting for today.

The strategy extends beyond the acquisition of customers. Google offers $ 350,000 startups in cloud credits, access to its technical teams and market support through its market. Google Cloud also provides that Whatza Desouza describes as a “no compromise” stack – from chips to models to applications – with an “open ethos” that offers customers the choice in every level.

“Companies love the fact that they can access our stack, they can access our teams to understand where our technologies are going,” says Desouza, bold interview. “They also love that they have access to the Google class infrastructure of corporate degree.”

Google’s game of infrastructure has become even more ambitious, with the reporting that had the maneuver behind the scenes of the company to expand its chip activity to personalized. According to informationGoogle has done business to put its tensor processing units (TPU) for the first time in the other data centers of other cloud providers, including an agreement with Fluidstack based in London which included up to 3.2 billion dollars in financial support for a New York structure.

Compete directly with artificial intelligence companies, providing them at the same time the infrastructure requires – let’s call it – finesse. Google Cloud provides the TPU chips in Openi and hosts the Claude model of Anthropic through its vetex AI platform, even if its Gemini models compete head with both. (The Mother Company of Google Cloud, Alphabet, also has a 14% participation in anthropic, for short documents of the New York Times obtained at the beginning of this year, even if when requested directly on Google’s financial relationship with Anthropic, Desouza defines the relationship with a “multiple levels” redirect me to the market for Google Cloud models – note that customers can access various models Foundation).

But if Google is trying to advance Switzerland, while he has had a lot of practice. The approach has roots in Google’s Open Source contributions, from Kubernetes at the base “Attention is all you need“Document that allowed the architecture of the transformer at the base of the most modern AI. More recently, Google has published an open source protocol called Agent-to-age (A2A) for inter-agent communication in an attempt to demonstrate his continuous commitment to open Innreas.

“Over the years we have made the explicit choice to be open at every level of the stack and we know that this means that companies can absolutely take our technology and use it to build a competitor at the next level”, recognizes Desouza. “It has been happening for decades. It’s something we’re fine with.”

The courtship of Google Cloud’s startups arrives in an interesting interesting moment. Precisely this month, the federal judge Amit Mehta issued a faded sentence in the five -column monopoly box of the government, trying to curb Google’s domain without projecting his artificial intelligence ambitions.

While Google had the most serious sanctions of the Department of Justice, included the forced divestment of his Chrome browser, the regulatory concerns underlined for the company that exploited his research monopoly to dominate the IA. Critics are concerned, understandably, that Google’s vast trunk of research data provides an unfair advantage in the development of artificial intelligence systems and that society could the same monopolistic tactics that have obtained its research domain.

In the conversation, Desouza focuses on much more positive results. “I think we have the opportunity to understand in a fandently so of the main disease that today we wear a good understanding,” says Desouza, for examples, outlines a vision in which Google Cloud helps the research on power in Alzheimer’s technologies, Parkinson’s. “We want to work very hard to make sure we are pioneering of technologies that will allow that job.”

Critics may not be easy. By positioning itself as an open platform it authorizes rather that it controls the next generation of artificial intelligence companies, Google Cloud could show regulators that promotes competition rather than suffocating it, all while creating relationships with startups that could increase the pressure.

For our complete chat with Desouza, take a look at this week Rigorous download podcast; A new episode comes out every Tuesday.