Oracle and the lack of Foresight

Just last month I wrote a piece about the silicon casino, calling out the biggest sucker at the table regarding GenAI hardware [link]. I predicted Oracle was positioning itself to be the ultimate bagholder by stockpiling depreciating chips for a party that might be ending.

I did not think I would get to cash in a prediction receipt this quickly. Frankly, it feels rather prescient now.

If you looked at Oracle’s Q2 earnings report earlier this month, the validation was painted in bright red. To be clear, their other business units are doing fine. The boring stuff pays the bills and the legacy database contracts are sticky. If this were 2019, Larry Ellison would be popping champagne.

But when it comes to the GenAI stuff, that is where the wheels fall off.

Despite the hype, Oracle missed its total revenue targets. The market took one look at the numbers, and more importantly the terrifying amount of money Oracle plans to burn, and the stock tanked nearly 15%. Investors finally asked the question I posed last month: who exactly is paying for all this? Oracle announced they are pouring another $15bn into capital expenditure, effectively maxing out the company credit card to build data centres that have not generated the returns to justify the concrete they are poured from.

The panic is palpable in their messaging. Notice how they have stopped bragging about training massive models and started screaming about "Agentic AI"? That is a pivot born of desperation. They have realised the training era is a finite gold rush and are now desperately trying to convince enterprises that they need H100 superclusters just to run customer service bots.

We do not need a Ferrari to deliver a Domino's pizza. We certainly do not need a massive Nvidia Blackwell cluster just so an "agent" can hallucinate its way through ordering our weekly shopping.

Then there is the $300bn elephant in the room: Project Stargate.

It is genuinely hilarious that Oracle is literally building the infrastructure for OpenAI. They are underwriting the most expensive construction project in tech history for a tenant that could technically pack up and leave for Microsoft or their own custom silicon in a few years.