Generative AI is getting harder to understand, not easier. Elad Gil, an early-stage investor who called the frontier LLM oligopoly in 2021, has published a stack-level breakdown of every major open question he still cannot answer. His 2021 prediction is largely tracking: frontier models like OpenAI, Google, Anthropic, Grok, Llama, and Mistral are consolidating, while commodity models drop roughly 5x in training cost every two years. The funding math is stark. Microsoft put $10B into OpenAI, which equals about 6 weeks of Azure revenue. Amazon and Google together put $7B into Anthropic. Venture capital is a rounding error by comparison.
The more important argument lives in the numbers Gil surfaces around returns. Azure grew 6 percentage points from AI in Q2 2024 alone, an annualized revenue lift of $5 to $6 billion, which is already 50 percent of Microsoft's total OpenAI investment, per year, in revenue. That is before counting net income or compounding growth. Meta is running a separate playbook, committing a $20B compute budget and pushing Llama as the open-source anchor Gil predicted would need a deep-pocketed sponsor. The structural claim is that nation-states and cloud providers, not VCs, are now the kingmakers for which frontier models survive.
Gil does not pretend to have answers. The value of reading the full piece is the question list itself, organized by stack layer, covering LLMs, infrastructure, applications, and market structure. For anyone building in or investing around AI, this is a rare artifact: a candid accounting of what a knowledgeable insider genuinely does not know, written without a conclusion to sell you.
[READ ORIGINAL →]