Oracle goes all in: $50 billion for AI cloud

Larry Ellison has never been known for modesty, but his latest appetite is making even Silicon Valley swallow nervously. Oracle intends to raise up to 50 billion dollars in 2026. For scale, that's the budget of a mid-sized European country, or the cost of several ambitious space programs. The money will go toward a single goal: building cloud infrastructure that can absorb AI's endless demand. In an era when compute has become the new oil, Oracle wants to own the most wells.

Just a few years ago, Oracle was seen as a respectable but somewhat sluggish database giant. Many wrote the company off in the cloud race while Amazon, Microsoft, and Google carved up the market. But Ellison realized early that AI isn't just clever algorithms; it's also physical space filled with servers, cables, and highly complex cooling systems. Oracle bet on its so-called "second-generation" architecture (Gen 2 Cloud), which lets it deploy capacity faster and more flexibly than its competitors. Now that demand for training large language models has exceeded every expectation, that strategy is starting to pay off.

The 50 billion will come from a mix of debt and stock sales. It's a risky move given current economic instability and interest rates, but in AI, delay costs more than any loan interest. The company needs to build new data centers right now, because the queue of LLM developers stretches months ahead. A provider without free GPUs and space to house them simply drops out of the game. Oracle doesn't just want to take part; it wants to set the terms.

It's fascinating to watch a software company turn into one of the biggest developers of digital real estate. Oracle no longer just sells software licenses; it's laying the foundation for the entire future industry. Its partnership with Nvidia and tight integration with Nvidia's chips have already given Oracle an edge that helped it lure some major customers away from Azure and AWS. Investors see this not merely as expansion but as an attempt to seize leadership in a critical segment. Oracle is betting that its cloud will become the standard for training the heaviest models.

This financial maneuver also confirms that the era of "cheap" AI ended before it ever began. Training next-generation models takes not millions but tens of billions of dollars. Oracle is betting that the tech boom won't burst in the next couple of years and will instead become the backbone of a new global economy. While critics debate whether the sector is overvalued, Ellison is ordering concrete, power systems, and server racks years in advance, effectively reserving the future for himself.

The company's main challenge now isn't finding money but spending it effectively. Building a huge hangar and packing it with chips is only half the job. Oracle also has to secure massive amounts of power, which is becoming scarce even in developed countries. Oracle has already announced plans to use small modular nuclear reactors for its data centers. If the 50 billion helps solve not only the space problem but the energy problem too, the company will become nearly untouchable by competitors for decades.

We're watching the cloud market turn, once and for all, into a capital expenditure market. The winner won't be whoever writes the best code, but whoever has the most transformers, liquid cooling systems, and grid access. Oracle spent a long time in the shadow of the "Big Three," but now it's stepping into the light with a check that's impossible to ignore. This isn't just a business plan; it's a declaration of intent to dominate the physical world where artificial intelligence lives.

The big question: Will competitors have the nerve and the cash to answer Oracle's aggressive challenge in 2026?