Modal CEO on who wins inference, open-source peak, and RL's real limits
20VC · rss · 2026-11-02
20VC interviews Modal CEO Erik Bernhardsson (Modal: $1.2B raised, $750M Series D led by Accel at $15.75B valuation, $1B+ ARR growing 15x YoY):
- Inference market: who wins in 5 years; whether open-source LLM inference has peaked; if model routing is a defensible business.
- Compute economics: GPUs as an "open-air drug market"; whether GPU prices collapse and data centers are overbuilt; custom silicon vs NVIDIA; is CUDA really the moat; NeoClouds (Nebius vs CoreWeave) as partners or rivals.
- Business models: are 90% software gross margins dead in AI; should enterprises own their own intelligence; the gold rush selling data environments to labs.
- Technical limits: the real limits of post-training and RL.
- Startup lore: raising $1B with no pitch deck, the danger of raising too much, Modal's GPU burn and monthly coding-agent spend, $20M+ packages for exceptional new grads.
More from Venture
- Right about everything: Amazon beat Walmart on fundamentals yet returned 80 points less since 2023 — FinanceYF5 · 2026-11-02
- Grady Booch Jokes He'll Move Retirement Funds Into NFTs Amid AI Stock Volatility — Grady_Booch · 2026-11-02
- AI game site reveals referral leaderboard: top sharer's 20,932 visits earn just $20.30 — ezshine · 2026-11-02
- AI safety super PACs outspend rivals $44.4M to $27.9M in 2026 midterms — xeophon · 2026-11-02
- Nvidia's 27-year market cap history: from $800M to $5T, with 70% crashes along the way — mitchdeg · 2026-11-02
- Sequoia's $1T thesis: sell the work, not the software — and why copilots may be the wrong bet — basedjensen · 2026-11-02