DeepSeek V4 Just Broke the Math on $500 Billion in AI Capex
Two game theories, one disruptor, and an investment thesis the frontier labs cannot solve their way out of.
Last week I switched my OpenClaw supervisor off the frontier model I’d been paying premium rates for and pointed it at DeepSeek V4. Same prompts, same context windows, same tasks — the results were at least as good as the model I’d swapped out, and on the coding-heavy paths measurably better. Cleaner reasoning traces. Faster on agentic loops.
The week’s API bill came in at a fraction of what I’d budgeted.
I did not expect that. I had mentally categorized DeepSeek as a so...
Sign In to Continue
This article is available to all EverVests members with a Basic subscription or higher.
Sign in to your account or start a membership to access this content.
Already a member?