Costs, Profits and Losses in the Battle for Superintelligence

By Brian Wang

XAI is releasing Grok 2 in beta. This will be about equal to OpenAI GPT4 artificial intelligence models.

Elon Musk has said XAI is assembled 100,000 Nvidia H100 chips for a total of 400 Exaflops of compute to train Grok 3 by the end of this year. Grok 3 will be at about the power of still unreleased OpenAI GPT 5.

Elon has also said XAI will be setting up 300,000 Nvidia B200 chips in mid-2025 for AI training. This will have 15 times the compute power of the 100,000 Nvidia H100 chips. This will increase another 15 times in 2026-2027 with one million next generation Nvidia Rubin chips.

This increases the cost of the AI data centers. As the cost of the data centers reaches $150 billion this will require revenues that match or exceed the costs.

Revenue would need to be from AI generated software, AI supplied IT or business consulting, AI queries replacing internet search, robotaxi or humanoid bots.

Here, Brian Wang describes the scaling of Grok 4 and 5 using 300,000 twenty Petaflop Nvidia B200 chips in 2025 and going beyond to 1 million Nvidia Rubin chips.