Thesis
Author: Freeman LewinEvents
Our read
Our read
A reported 22% rent increase for older Nvidia chips makes Micron’s guide and ASML’s machine wins at TSMC and Samsung read as one capacity signal: AI compute scarcity is persisting across generations and pulling investment into memory and lithography. This take would be falsified if used-GPU rents and utilization roll over while Korean memory indicators and EUV bookings fail to convert to shipments or installed capacity. Test it against GPU rental prices and utilization by model, Korean HBM and DRAM exports and contract prices, and EUV bookings, deliveries, and fab-installation schedules.
GPU cloud rental price tracker API H100 A100 historical rates
Higher rents for older Nvidia accelerators suggest demand is not confined to the newest generation, supporting a durable capacity shortage rather than a brief product-cycle spike.
NVDAoperating-metric inflectionfinance.yahoo.com
Commitments to ASML’s new EUV machines from TSMC and Samsung are the physical-capacity response implied by sustained accelerator scarcity, extending the signal from rental economics into fab investment.
ASMLcapex-expansion eventfinance.yahoo.com
The Seoul signal provides an external check on Micron’s aggressive guide, indicating that persistent compute demand may already be transmitting into the memory supply chain.
MUguidance-language deltafinance.yahoo.com
Next step
See how Discover can help you act on the sources behind it.
[withheld] — Cloud GPU compute pricing index — historical & real-time price/index series for GPU rental (H100, H200, A100 SXM4, B200, RTX 5090, RTX PRO 6000 WS), plus a token price index, token volume index, workload cost index, LLM coding analytics, datacenter site map, and power market data.. Small but well-capitalized data vendor — 19 employees, $33M total funding, narrow but high-value B2B compute-pricing niche. Use cases: Alternative/financial data for AI-infrastructure investing, cloud cost benchmarking, compute market forecasting, and quant signals on AI infra-exposed public equi…
[withheld] — Global customs import/export trade data — shipment-level records (date, HS code, product description, origin/destination, exporter/importer, quantity, unit, value, port) across many countries and product categories, including semiconductor-specific HS codes (e.g. 85340090); delivered as Customs Data, Statistical Data, B/L Reports, and Customised Analytical Reports.. Small-to-mid B2B trade-intelligence vendor — 12 employees, New Delhi-based, market-research industry classification, with coverage spanning multiple global regions. Use cases: Supply-chain and trade-flow intelligence, …
[withheld] — Reseller/distributor access to the DRAMeXchange DRAM/NAND Flash Market Intelligence Service — tiered (Silver/Gold/Platinum) membership providing downloadable daily spot & monthly contract DRAM/NAND prices (2017-present), a Semiconductor Research Index (DXI, from 2020), historical Daily Express data back to Nov. 2003, monthly datasheets, quarterly reports, and analyst consulting access.. Established mid-size market-research distributor — 37 employees, ~$23M revenue (Kawasaki, Japan); this DRAM/NAND product is one of many syndicated research memberships…
[withheld] — Global Semiconductor Database & Market Intelligence — continuously updated project-level data on fab expansions, new plant construction, and equipment upgrades worldwide, delivered via the PECWeb Integrated Platform with geospatial mapping and analytics dashboards.. Large, established data/market-intelligence vendor — 820 employees, ~$104M revenue; semiconductor vertical is one of many industry databases on its broader Global Market Intelligence (GMI) platform. Use cases: Supply-chain diversification analysis, semiconductor capex/investment tracking, decision…
[withheld] — GPU cloud rental reference-rate index — daily on-demand (guaranteed and interruptible) USD/GPU-hour benchmark rates for H100, H200, B200, B300, and A100 across three operator segments (Hyperscaler, Neocloud, Marketplace), with regional breakouts (US & EU) and committed-tenor term series.. Very small, specialist operation — 2 employees (Ethra SAS, Paris), positioned as an independent citable GPU pricing benchmark rather than a large vendor. Use cases: Institutional benchmarking of AI-compute costs, procurement/purchasing decisions for GPU capacity, investment res…
[withheld] — DRAM & NAND Flash memory-chip price intelligence — monthly NAND Flash Wafer Contract Price reports and (per [withheld]'s broader price-trends site) DRAM Spot Price / DRAM Contract Price series; delivered as paid downloadable research reports and a membership-based market-research [withheld] built on over a decade of historical pricing data.. Established, industry-standard memory-price data provider under [withheld], a global market-intelligence and consulting firm; Apollo lists ~$10M revenue for the [withheld] entity specifically, though the parent TrendFo…
[withheld] — GPU cloud rental pricing indexes (H100, A100, B200 Spot-Contract Composite Indexes) — hourly USD/GPU-hour benchmark prices built from hyperscale cloud, neocloud, GPU marketplace, and rental-router data blended with analyst-run contract-pricing surveys; the H100 index is also distributed on Bloomberg under ticker SAH100SC.. Mid-size, established semiconductor research firm — ~90 employees; this specific index product already carries institutional-grade distribution via Bloomberg. Use cases: Institutional/quant benchmarking of AI-compute costs, procurement and purchasing analytic…