AI infra: chip funding and rising rents
Two market signals underline tightening AI infrastructure economics: new funding for open RISC‑V chip designs and a sharp rise in hourly rent for Blackwell GPUs. Parameter reported a $400M raise for SiFive‑backed RISC‑V designs and an industry tracker showed Blackwell GPU rents up to $4.08/hr (≈+48%), while observers note fewer than 10% of US data centres are AI‑ready—creating capacity and cost pressure (parameter.io) (intellectia.ai) (x.com).
Renting Nvidia’s newest Blackwell graphics processors now costs as much as $4.08 an hour, while SiFive has raised $400 million to build more artificial-intelligence chip designs. (techmeme.com) (sifive.com) The rental figure comes from the Ornn Compute Price Index, which the Wall Street Journal cited on April 13 as showing Blackwell rates up 48% from $2.75 two months earlier. SiFive said on April 9 that its Series G round was oversubscribed and valued the company at $3.65 billion. (techmeme.com) (sifive.com) A graphics processor is the workhorse that trains and runs artificial-intelligence models, and a central processor coordinates memory, storage, and networking around it. SiFive sells central-processor designs based on RISC-V, an open instruction set that companies can license instead of using Arm or x86 designs. (sifive.com) (thenextweb.com) SiFive said the new money will speed its high-performance data-center roadmap, with Atreides Management leading the round and Nvidia, Apollo Global Management, Point72 Turion, T. Rowe Price Investment Management, Prosperity7 Ventures, and Sutter Hill Ventures participating. Reuters reported Chief Executive Patrick Little said the company is targeting the data-center central-processor market and expects this to be its last private round before an initial public offering. (sifive.com) (finance.yahoo.com) The price jump for rented Blackwell chips is landing in a market that still lacks enough buildings to host them. Sean Farney of Jones Lang LaSalle told Data Center Knowledge that less than 10% of United States data-center inventory can handle “true AI-dense critical load” today. (datacenterknowledge.com) Farney said many enterprise facilities still rely on traditional air cooling, while “pure-play” graphics-processor clouds are built for much denser power draw and liquid cooling. That leaves older sites poorly matched to the heat and electricity demands of large artificial-intelligence clusters. (datacenterknowledge.com) Construction is also running into bottlenecks outside the server rack. Bloomberg reported on April 1 that shortages of transformers, switchgear, and batteries are delaying United States data-center projects and increasing reliance on imported electrical equipment. (bloomberg.com) Those constraints help explain why investors are funding alternative chip suppliers even as hourly graphics-processor rents climb. One side of the market is trying to add more compute; the other is charging more for scarce capacity that is already online. (sifive.com) (techmeme.com) (datacenterknowledge.com) For artificial-intelligence builders in April 2026, the pressure is showing up in two places at once: on the balance sheet when they rent Blackwell chips, and in the capital markets when suppliers raise money to build around the shortage. (techmeme.com) (sifive.com)
