All roads lead to Neocloud: GPU renters evolving into energy, aggregation and custom-chip inference sellers
FinanceYF5 · x · 2026-09-08
The author argues all paths in AI compute ultimately lead to the Neocloud model, which now goes far beyond "renting GPUs".
Extensions include starting from energy and data centers, aggregating compute from multiple providers, selling compute to research and RL customers, building custom chips and then selling inference, and running inference across chips.
Why: AI companies' compute demand keeps rising; long-term leases are too expensive, and after buying GPUs they must raise utilization — so they naturally start selling compute externally or hosting workloads.
Related event: All Roads Lead to Neocloud: Seven Types of Compute Companies Converge(2 posts)→
More from Infra
- AUR llama.cpp-cuda removal caused 10x slowdown; manual rebuild restored 1800 t/s prefill — MrHall · 2026-09-08
- Longsys lists in Hong Kong, raising $910M as AI-driven memory boom fuels record profits — 创业邦 · 2026-09-08
- H100 rental price up 22% in a month to $3.28/hr, vindicating NVIDIA's depreciation call — iScienceLuvr · 2026-09-08
- Anthropic Reportedly Abandons ~$6B Acquisition of Israeli AI Startup Decart Weeks Before Its IPO — ns123abc · 2026-09-08
- VDN-H3 video generation running on 24GB VRAM: 30-second clips on an RTX 3090 Ti — NoMouse9610 · 2026-09-08
- NeoCloud is brutal: one failed rack can cost operators six months of rent — rohanpaul_ai · 2026-09-08