The AI Compute Race: Microsoft’s Miss and Oracle’s Opportunity

The AI Compute Race: Microsoft’s Miss and Oracle’s Opportunity Source: https://isolveproblems.substack.com/p/the-ai-compute-race-microsofts-miss

I don’t know Larry Ellison, the co-founder, executive chairman, and CTO of Oracle Corporation. However, I have a few anecdotes to share.

First, the name Oracle originated from a CIA project, aptly called “Oracle,” and the agency became its first customer. The word on the street is that Oracle also played a key role in storing vast amounts of data following the 1970s treaties on non-proliferation of nuclear weapons and the subsequent scarcity of atomic tests, which later evolved into large-scale computer simulations.

Larry Ellison is, therefore, no stranger to big computing clouds and probably started before anyone knew what a cloud was. He’s also a billionaire—like the Chuck Norris of billionaires.

When Larry buys a jet, a $38 million Gulfstream V, and the landing strip in San Jose, right next to his Santa Clara office, is too short for it, he doesn’t look for a smaller jet. Instead, they extend the landing strip.

Larry also likes boats. I know this because I’m Swiss, and our team, Alinghi, has won the America’s Cup several times, directly competing against Larry’s Oracle Team USA entry. A good friend of mine in Silicon Valley used to tease Ellison by parking his car with a Swiss flag and a large Alinghi sticker on it next to his, near a fancy grocery store with excellent croissants on Sunday mornings, somewhere near Woodside.

Then came 2010. That year’s Oracle entry in the America’s Cup was halfway between a racing sailboat and an alien spaceship. It had a rigid wing sail of a design that no one had seen before.

Those boats are the Formula One of the seas, and his boat that year left everyone else, including Alinghi, standing still.

They somehow made the most of the faintest breeze, allowing the boat to sail upwind better than any other competitor, while setting new speed records up to 2.5 times faster than the wind speed when sailing downwind. It was nothing but a technology showcase with a huge Oracle logo slapped on it, and Larry was actually racing on board the ship as a member of the crew!

After a no-contest domination, I joked that he must have rung his friends at Los Alamos to run his boat’s fluid dynamics in between two nuclear bomb simulations 🙂

Fast forward to January 21, 2025, and the same Larry Ellison is seen flipping burgers with President Donald J. Trump in the Roosevelt Room of the White House in Washington, D.C., alongside Sam Altman of OpenAI and a representative of SoftBank, a Japanese multinational tech investment firm.

The topic was The Stargate Project, a new private venture funded by OpenAI, Oracle, SoftBank, and MGX, an investor from the UAE. The project aims to invest $500 billion over the next four years in AI infrastructure and create 100,000 new jobs in the U.S.

Microsoft wasn’t part of this picture.

Microsoft’s CEO was at the WEF in Davos, Switzerland at that time and told an interviewer that “OpenAI committed in a very significant way to Azure consumption” and that Microsoft had a ROFR (Right Of First Refusal) meaning OpenAI had to shop for compute at Microsoft first: “We have right of first refusal. He comes to us first. If we meet those needs, then we clear it. If not, he can go to these other providers.” said Nadella.

The CNBC interview, which aired on January 22, is available on YouTube:

Apparently, Microsoft did not meet OpenAI’s needs because just a few weeks later, on March 10, OpenAI, in a joint release with CoreWeave, said through Sam Altman their CEO that “Advanced AI systems require reliable compute [...]” and announced a strategic partnership worth $11.9 billion in compute hours to CoreWeave over five years “for training and delivering [OpenAI’s] latest models at scale to its hundreds of millions of users around the world.”

OpenAI also announced a $350 million investment in CoreWeave.

It sounded like CoreWeave had just taken both OpenAI’s model training and model operations away from Azure: https://www.coreweave.com/news/coreweave-announces-agreement-with-openai-to-deliver-ai-infrastructure.

But that was only the beginning.

On July 22, 2025, OpenAI and Oracle announced a $30 billion per year partnership for “data center services,” further affirming OpenAI’s move away from Azure: https://techcrunch.com/2025/07/22/openai-agreed-to-pay-oracle-30b-a-year-for-data-center-services/

The deal was later clarified to represent $300 billion, according to a September 10, 2025, article from the Wall Street Journal: https://www.wsj.com/business/openai-oracle-sign-300-billion-computing-deal-among-biggest-in-history-ff27c8fe

To round things up, OpenAI posted an article on September 23, 2025, summarizing all the deals. The TL;DR is five data centers, $500 billion secured by the end of 2025, 10GW of total power, and CoreWeave and Oracle as infrastructure providers. The name “Microsoft” appears zero times in this article: https://openai.com/index/five-new-stargate-sites/

Based on recent partnerships and the wording in the announcement of their move to CoreWeave, which alludes to reliability concerns with Azure, along with Microsoft’s CEO statement about its right of first refusal, the logical conclusion is that OpenAI’s needs weren’t met.

Consequently, they exercised their right to jump ship, and they did it in style.

The result is a $311.9 billion opportunity loss for Microsoft, at the very least, and a significant setback for Azure in the AI sector, where investments like custom AI chips (“Maya”) and custom datacenter server processors (“Cobalt”) might add to the toll and turn into expensive paperweights as they will be obsolete long before another OpenAI shows up.

Microsoft isn’t shrinking its workforce to replace humans with large language models. That idea seems unlikely. What’s likely happening is that the $11.9 billion, which quickly turned into $311.9 billion in missed opportunities, needs to be recouped somehow. This revenue was most likely part of their overall AI investment and revenue plan, Azure usage, and profit prediction, which hit a snag. Shrinking the workforce is a quick expedient to hit the profit margins and reassure Wall St in the short term.

To add insult to injury, Azure is under scrutiny by the Department of War, which discovered that unvetted foreign workers had access to Defense computers with little supervision. The Secretary of War, the Honorable Pete Hegseth, using words like “unacceptable risk” and “breach of trust” between Microsoft and the Government in public communications: https://www.war.gov/News/News-Stories/Article/Article/4288992/pentagon-halts-chinese-coders-affecting-dod-cloud-systems/

How hard was it to make Azure reliable enough to satisfy OpenAI? Who inside the company was supposed to see that disaster coming, but didn’t, or failed to take action?

The net result so far is that Microsoft lost its largest customer and put its government contracts at risk, with Grok, not CoPilot, taking the stage: https://www.gsa.gov/about-us/newsroom/news-releases/gsa-xai-partner-to-accelerate-federal-ai-adoption-09252025 and Oracle making even more wins with the acquisition of TikTok with President Trump’s blessing (https://www.businessinsider.com/who-could-take-over-tiktok-us-business-investors-oracle-ellison-2025-9) where Microsoft used to be a top bidder: https://www.bbc.com/news/technology-58719674

It looks like some strategic rethinking is in order in Redmond, WA.

Follow for more...