GPU Clouds and Neoclouds: CoreWeave, Lambda, Nebius — 2026-09-26
This week, Nebius announced GPU and memory price hikes of up to 21% and 41% respectively effective October 1, while JPMorgan upgraded CoreWeave citing a strengthening compute-pricing backdrop — yet neocloud stocks fell broadly as investors sorted through conflicting research notes. Lambda closed a $926 million senior secured Term Loan B backing GPU deployment for an investment-grade customer, and UK neocloud Nscale raised $3.36 billion in convertible financing ahead of its US IPO. Meanwhile, spot H100 rental prices softened in Asian rental markets even as headline neocloud contracts keep pricing power concentrated at the top.
GPU Clouds and Neoclouds: CoreWeave, Lambda, Nebius — 2026-09-26
Top developments
Nebius raises AI cloud prices up to 21%, memory rates up 41%
Nebius raised AI cloud GPU prices by up to 21% and memory rates by 41%, effective October 1, 2026 — its second price hike in three months amid tight capacity. The move extends a pricing trend JPMorgan flagged this week: Nebius has implemented frequent increases while CoreWeave adjusted pricing on roughly 25% of its products in July. For GPU-rental operators with committed capacity, the hikes signal that on-demand pricing power remains strong going into Q4.
JPMorgan upgrade sends Nebius up 6%, but the group splits
A JPMorgan upgrade citing rising compute pricing lifted Nebius 6% on September 24, while CoreWeave treaded water and IREN slid 4% — a divergence that highlights where real pricing power lives in the sector. JPMorgan reset its CoreWeave price target for 2026, saying the pricing backdrop "has strengthened throughout the year." By the next day, however, selling extended across the group: IREN dropped 4%, CoreWeave slid 3% and Nebius eased, as shareholders weighed two respected firms' opposite verdicts.
Lambda closes $926M Term Loan B backed by investment-grade customer
Lambda closed a $926 million senior secured Term Loan B facility on September 22, backing GPU deployment for an investment-grade customer — presumably Microsoft, following its earlier $1B private-debt raise in August. The deal underscores how neocloud debt financing is increasingly premised on top-tier counterparties: lender terms like CoreWeave's six-year straight-line GPU depreciation with no salvage clause show the credit scrutiny now applied to GPU collateral. CoreWeave itself carries roughly $35B in total debt as of June 30, 2026, with quarterly net interest expense of $640M against a $104B revenue backlog.
CoreWeave-Meta contract drives $2.58B revenue surge
CoreWeave's Meta contract helped drive a $2.58 billion revenue quarter, with total revenue of $2.078 billion (+112% YoY per an earlier analysis) and a contracted backlog of $104 billion plus $25 billion added in early Q3. Nebius, by comparison, reported quarterly revenue between $575–582 million, growing more than 450% year-over-year. The Meta anchor deal illustrates how hyperscaler and AI-lab commitments keep neocloud utilization full while spot prices soften.
CoreWeave credit terms: six-year GPU depreciation, no salvage
Analysis of CoreWeave's credit agreements reveals lenders negotiated six-year straight-line depreciation on GPUs with no salvage term — a conservative assumption given generational turnover roughly every two to three GPU steps. The piece examines three cloud rate cards and nine generation steps, noting that one card already prices lower. This matters for the whole sector: if top-tier lenders treat GPU value as fully consumed in six years, debt capacity for Lambda, Nebius and peers depends heavily on contract quality, not just utilization.
Local view
Chinese GPU-rental commentary is flagging a bifurcated market. A Sina/Weibo post reported that NVIDIA rental prices for most cards pulled back, with H100 at $2.85/hr (down 5.63% week-on-week) and H200 down more than 9% weekly. A 什么值得买 summary of Gelonghui quotes on September 26 showed H200 at $5.08/hr (-1.55% daily, +19.81% monthly), H100 at $2.85/hr (-2.40% daily) and A100 flat at $1.02/hr — and noted Nebius planning another ~20% hike, framing Q4 rental strategy around three Ledgers of cost. Another viral take argued that while most cards fell, top-tier compute prices surged — head-card demand remains tight even as older GPUs soften.
Context & numbers
- September's H200 on-demand listings range from $2.09/GPU-hour at Beam to $13.78 at Microsoft Azure — a 6.6x spread across providers.
- H100 rental prices have roughly halved to ~$3.38/hr in 2026, while B200 and H200 command steep premiums amid a memory supply crunch.
- CoreWeave node pricing reference points: $49.24/hr for an 8x H100 node ($6.16/GPU), up to $68.80 for a B200 node, with spot up to 60% off and free egress.
- Silicon Data publishes daily, financial-grade GPU rental indices for H100, H200, A100, B200, B300 and MI300X across neoclouds and hyperscalers.
- SemiAnalysis released ClusterMAX 3.0, its refreshed industry-standard GPU cloud rating system covering reliability, performance, support, pricing, and security.
On the radar
- Nebius price increases take effect October 1, 2026 — watch customer reaction and churn.
- UK neocloud Nscale secured $3.36B in convertible financing (Third Point, Nvidia, others) ahead of a US IPO, amid reports of a going-concern warning in its $35B-valuation IPO filing and $103.4B in contracts (including a $44.6B Anthropic anchor deal) against only $140.6M in first-half 2026 revenue — a circular-financing story worth tracking.
- Anthropic is reportedly in talks to lease up to 1GW of additional compute, with associated investment estimated at $40B+ — potential uplift for whichever neoclouds win allocation.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
