All paths lead to neocloud: seven flavors of companies now selling AI compute
vaibhavbetter · x · 2026-09-07
Jessie Dong maps out why 'neocloud' now means many things: AI companies need ever more compute, renting gets expensive, and long-term contracts win, so everyone ends up selling compute.
She sorts players into seven categories: renting own GPUs (CoreWeave, Nebius, Lambda), starting from power and data centers (Crusoe, IREN, Nscale), aggregating compute (RunPod, Vast.ai, Akash), doing research/RL while selling compute (Prime Intellect, Together), building own chips and selling inference (Groq, Cerebras), and running inference across chips (Gimlet Labs). Her takeaway: at the 2026 pace, terminology changes every 60 days.
More from Venture
- Hirevire open-sources its metrics: $15,103 MRR, up 10.34% MoM in year 4 — lee_stott · 2026-09-07
- Instinct's playbook: free product, costly compute, VC cash subsidizing agent habit formation, Uber-style — AccBalanced · 2026-09-07
- Nvidia reportedly negotiating $2.5B investment into Mira Murati's Thinking Machines Lab — eyishazyer · 2026-09-07
- Baidu's HK Shares Join Shenzhen and Shanghai Stock Connect, Effective Immediately — Baidu_Inc · 2026-09-07
- Town CEO on $100B AI Assistant Race and $75K AI Tools Spend Per Engineer — 20VC · 2026-09-07
- AfterQuery hits $3.2B valuation 18 months after founding, up 10x in 5 months — 创业邦 · 2026-09-07