The Great GPU Shortage – Rental Capacity – Launching our H100 1 Year Rental Price Index

Daniel Nishball · SemiAnalysis · April 02, 2026 at 04:25 · ⏱ 23 min read  | Read on Substack ↗
Summary
GPU rental pricing is surging ~40% as demand from AI labs, agentic workloads, and open models outstrips supply, while memory and server component costs are rising parabolically. This contradicts market skepticism about GPU terminal value and suggests sustained pricing power for Neocloud providers, with upward pressure on supply-chain names like TSMC and Micron.
  • H100 1-year GPU rental contract pricing rose 40% from $1.70/hr in Oct 2025 to $2.35/hr by Mar 2026.
  • Anthropic's ARR tripled from $9B to over $30B in a single quarter, driving GPU demand.
  • March 2026 saw complete sell-out of on-demand GPU capacity across all types; lead times for new Blackwell deployments extend to June-July.
  • Memory pricing (DRAM, NAND) went 'parabolic' in early 2026, with LPDDR5 and DDR5 contract prices tracking ~4x-5x YoY increases.
  • OEMs repriced AI servers significantly above underlying component cost increases, compressing project returns and slowing supply additions.
  • AI Labs are locking up 4-5 year offtake deals with high prepays (20%+), while also renewing H100/H200 clusters to prevent capacity re-entering shorter-term markets.
  • Despite tight supply and rising prices, public market sentiment on Neoclouds (CoreWeave, Nebius, IREN) remains negative, with share prices near 6-12 month lows.
  • The author identifies three checkpoints: GB300 cluster ramp, silicon shortage severity, and AI lab ARR growth — all pointing to continued upward price pressure.
Read time 23 min
Length 23,204 chars
Category finance
Trade Ideas
Daniel Nishball Substack author, SemiAnalysis
The article describes hyperscalers like AWS (Amazon) backstopping Neocloud offtake deals, collecting a slice of project revenue without expanding their balance sheet. AWS also benefits from tight on-d
The article describes hyperscalers like AWS (Amazon) backstopping Neocloud offtake deals, collecting a slice of project revenue without expanding their balance sheet. AWS also benefits from tight on-demand pricing (p6-b200 spot instances at $14/hr). Amazon is an indirect beneficiary of the GPU rental scarcity. Risk: AWS's own AI chip investments (Trainium, Inferentia) could shift demand away from NVIDIA GPUs, but near-term tightness still favors hyperscalers.
Daniel Nishball Substack author, SemiAnalysis
The article directly names CoreWeave as a Neocloud whose shares are at the low end of trading range despite clear evidence of tightening supply and rising prices — conditions that benefit Neoclouds th
The article directly names CoreWeave as a Neocloud whose shares are at the low end of trading range despite clear evidence of tightening supply and rising prices — conditions that benefit Neoclouds through margin expansion. The author argues the market is anchored to an oversupply narrative that contradicts ground reality. Risk: Public market sentiment may take time to reverse; execution risk on capacity additions.
Daniel Nishball Substack author, SemiAnalysis
The article states memory pricing (DRAM and NAND) went 'completely parabolic' with LPDDR5 and DDR5 tracking ~4x-5x YoY increases in Q1 2026. This directly benefits memory manufacturers like Micron (MU
The article states memory pricing (DRAM and NAND) went 'completely parabolic' with LPDDR5 and DDR5 tracking ~4x-5x YoY increases in Q1 2026. This directly benefits memory manufacturers like Micron (MU) as a key supplier of HBM, DRAM, and NAND for AI servers. Risk: Demand could soften if AI server deployments slow; memory cycle historically volatile.
Daniel Nishball Substack author, SemiAnalysis
The article notes tightness in TSMC's N3 logic wafer capacity as part of the 'great AI silicon shortage' that could worsen. TSMC is a direct beneficiary of sustained GPU demand and capacity constraint
The article notes tightness in TSMC's N3 logic wafer capacity as part of the 'great AI silicon shortage' that could worsen. TSMC is a direct beneficiary of sustained GPU demand and capacity constraints that underpin pricing power in the foundry market. Risk: Geopolitical risk; execution hiccups in advanced node ramps.
Daniel Nishball Substack author, SemiAnalysis
IREN is named alongside CoreWeave and Nebius as a Neocloud with share prices at low end of range but benefiting from GPU rental pricing surge. The author's analysis suggests pricing power and ROIC imp
IREN is named alongside CoreWeave and Nebius as a Neocloud with share prices at low end of range but benefiting from GPU rental pricing surge. The author's analysis suggests pricing power and ROIC improvement for providers with shorter-duration contracts and near-term capacity additions. Risk: Execution on data center builds; reliance on GPU availability.
Daniel Nishball Substack author, SemiAnalysis
Nebius (NBIS) is explicitly mentioned as another Neocloud with negative market sentiment but benefiting from the same supply/demand dynamics. The article notes all compute will be in-demand regardless
Nebius (NBIS) is explicitly mentioned as another Neocloud with negative market sentiment but benefiting from the same supply/demand dynamics. The article notes all compute will be in-demand regardless of relative performance. Risk: Competitive pressure from hyperscalers and larger Neoclouds; financing risk if capital markets remain skeptical.
More from SemiAnalysis

This newsletter, published April 02, 2026, features Daniel Nishball discussing AMZN, CRWV, MU, TSM, IREN, NBIS. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Daniel Nishball  · Tickers: AMZN, CRWV, MU, TSM, IREN, NBIS