Keep pulling the thread on Dylan Patel.
If AI model capabilities do not continue to improve, the U.S. economy will enter a recession due to the massive capital expenditures being made.
OpenAI risks becoming competitively irrelevant if it cannot secure enough compute to keep pace with tech giants like Meta and Google.
Meta's Mark Zuckerberg is pursuing a deal with Apollo for $30 billion to build a mega data center in Louisiana.
The annual rental cost for AI data center capacity is between $10 billion and $15 billion per gigawatt.
OpenAI CEO Sam Altman has stated that the company needs more than 10 gigawatts of data center capacity.
OpenAI is estimated to have an annual recurring revenue of $15-16 billion, which is projected to reach $20 billion by the end of the year.
The speaker believes that AI scaling laws, which show performance improving with more compute, will continue to hold true as they have over more than 10 orders of magnitude.
Anthropic has experienced one of the fastest revenue ramps ever, growing from under $1 billion to a $7-8 billion run rate, driven primarily by code generation use cases.
As part of a deal for 10 gigawatts of compute capacity, NVIDIA will make a $100 billion equity investment into OpenAI.
The total capital expenditure to build one gigawatt of AI data center capacity is approximately $50 billion, of which NVIDIA captures about $35 billion.
The strategic partnership between NVIDIA and OpenAI allows NVIDIA to effectively offer a discount to a key customer by taking equity, thereby securing future demand without a public price reduction.
The demand for AI inference tokens is reportedly doubling every two months.