A datacenter this size is an absurdity to begin with. Using Grok 5 ambitions, this would support training a 62t parameter model. 12-15x larger than what Anthropic currently can train. This means 12-15x slower (before considering that no western cluster can hold a model this size for inference, and slowdown is well over 100x) and 12-15x (100x here too) more expensive. So the concept is completely worthless to begin with.

Oracle’s force majeure absolutely means the project is delayed (despite statements in article) because they are delaying payments because its delayed. Probably by 2 years, unless after mid terms, all of a sudden, permits get rush approved despite political candidate consensus against Oracle’s BS.

If a 2 year delay, Oracle should cancel Nvidia GPU order. With 50% likely price increase path, it would be a $150B order that costs $30B in interest over 2 years at bad Oracle credit rating, and another $45B minimum depreciation over 2 years. This total is more than Oracle’s book value. This cancellation would be enough to ease significantly rampocalypse, and reverse Nvidia price increases.

The shell SPV companies that own the datacenters are an absurd scam. $18B in losses from lenders are at stake if project gets cancelled, with smaller losses for delays. They were all stupid to lend the money at low interest rates, and all other future projects will have expensive financing, and be unfavourable. This end of stupid is end of uninvestable AI fraud bubble. Again, end of rampocalypse.

Oracle is already a joke of stupidity in its AI pursuit. This makes it imppossible for it to ever raise more money for more stupidity. And so one of the first hyperscaler nvidia clients to tap out, as the start of domino chain