GPU Market Pulse — Week of June 15, 2026

Week of June 15, 2026 — a synthesis of 6 daily investment briefs and the underlying blended pricing data.

The week's most kinetic move was ALVEO_U30, which closed 11.0% lower at $0.22/hr blended. Across all instruments, the median GPU rental rate was unchanged 0.0% week-on-week.

Week in Briefs

Across 6 daily intelligence briefs this week, the analysis surfaced 31 investable hypotheses. Here's what built conviction.

Week at a Glance

Convictions That Built This Week

H100 — appeared in 5 briefs (Mon Jun 15, Tue Jun 16, Wed Jun 17, Thu Jun 18, Sun Jun 21)

A "cheap core" tier exists in US-Iowa ($1.24/hr), US-Oregon ($1.70/hr), and DE-Frankfurt ($1.79/hr), while a "premium edge" tier commands $3.59–$4.98/hr in Stockholm, Montreal, California, and Spain. The spread between cheapest (Iowa) and most expensive (Montreal at $4.98/hr) is 4×, and the intra-market spread in Montreal itself is 197% — meaning two providers in the same city are pricing H100 more than 3× differently.

Framing across the week: ⭐⭐⭐⭐⭐; H100 Demand Is Migrating to H200, Creating a Structured Arbitrage in SPOT Markets; H100 Accelerated Depreciation Is Diverging From Catalog Survival Expectations — Implying N

A100 — appeared in 2 briefs (Wed Jun 17, Sun Jun 21)

| Hypothesis | Conviction | Key Ticker to Watch | Time Horizon | |---|---|---|---| | H1: APAC Legacy GPU Scarcity Premium | High | A100_40GB\|SPOT\|jp-tokyo, V100_32GB\|SPOT\|jp-tokyo | 30–60d | | H2: H100 Two-Tier Market Bifurcation | High | H100\|SPOT\|us-iowa vs. H100\|SPOT\|ca-montreal | 14–30d | | H3: Inferentia Regional ASIC Repricing | Medium | INFERENTIA\|ASIC\|SPOT\|de-frankfurt 90d history | 7–14d | | H4: Power Moat Compounding (Infrastructure) | High | Intel feed: Missouri/Ohio permits; PJM queue | 60–180d | | H5: H100 Spot Floor Rising (Demand Floor) | Medium |

Framing across the week: APAC Legacy GPU Scarcity Is a Real, Durable Premium — Not Noise; Summary Signal Map; _Japan's Capacity Crunch Is the Preview of a Global Sovereign AI Infrastructure Shortage — _

Inferentia — appeared in 2 briefs (Wed Jun 17, Thu Jun 18)

It is a compressed version of the same dynamic playing out in the broader inference silicon market: AWS is flushing excess Inferentia v1 capacity in its overbuilt us-east regions, but is simultaneously supply-constrained in Europe. This creates a real arbitrage signal — EU inference workloads running on Inferentia are facing effective compute inflation of 164%, making it economical to migrate to GPU alternatives (A10, L4) or to bid up Inferentia2 reserved contracts.

Framing across the week: The AWS Inferentia Regional Strategy Signals a Broader ASIC Inference Repricing Cycle; Inferentia's Frankfurt Surge Is An Early Warning of EU ASIC Scarcity — A Precursor to Infe

Blackwell — appeared in 2 briefs (Mon Jun 15, Sat Jun 20)

This means Blackwell's arrival has barely dented H100/H200 pricing, either because supply is so thin it doesn't create pricing pressure, or because the market is treating Blackwell and Hopper as non-substitutable for different workload types. This is a structural divergence from the historical norm of ~17–54% annual decay per generation step.

Framing across the week: ⭐⭐⭐⭐; Legacy Inference GPU "Sticky Survival" is Anomalous and Will Correct — The T4/A100 Depreci

Open Questions Into Next Week

• Amazon's Trainium External Sales Strategy Creates a Structural Overhang on Nvidia's Mid-Market — But the 2026 Window Is Still Nvidia's (raised Jun 20)
• CoreWeave's Debt Structure Is a Leveraged Bet on Inference Pricing — The August 12 Print Is the Binary Event (raised Jun 20)
• H100 On-Demand Is In a Stealth Price War That Will Compress Neocloud Margins by Q3 2026 (raised Jun 21)

Market Snapshot

Blended GPU compute costs remained stable 0.0% this week across major cloud providers, led by ALVEO_U30's 11.0% decline.

Top Movers

Blended pricing = average of spot, on-demand, and 1-year reserved rates across major cloud providers.

Training vs Inference

Training-class GPU pricing held steady this week (avg $4.23/hr, +0.2% WoW), while inference-class pricing rose (avg $0.52/hr, +0.6% WoW).

The training-to-inference price ratio stands at 8.1x — stable compared to last week. The elevated spread suggests strong demand for training compute relative to inference, consistent with ongoing large model training activity.

Regional Spotlight: Middle East

Middle East trades at a 22% premium to global averages this week, with 13 GPU types available across the region. The most expensive GPUs in the region are GB200 ($21.54/hr), H200 ($12.36/hr), H100 ($7.31/hr). The 22% regional premium reflects emerging infrastructure and limited provider competition.

For detailed pricing data across all Middle East sub-regions, see the full regional profile.

Implications

For cloud buyers: Asia Pacific continues to offer the lowest average GPU pricing ($2.12/hr blended average). For workloads with regional flexibility, the gap between Asia Pacific and Middle East is $1.95/hr — a 92% premium. Compare regional pricing →

For semiconductor analysts: GPU pricing trends remain broadly stable this week. H100 (-0.2% WoW) and MI300X (+0.0% WoW) are tracking within normal ranges. Blackwell (B200) blended pricing at $7.40/hr (-1.6% WoW) provides an early read on next-generation adoption curves. View all GPU profiles →

For GPU investors: Stable pricing supports predictable returns for existing deployments. Model scenarios with the GPU ROI Calculator →