AI Chip Cloud Spot Price
Newest print Apr 2026Reviewed monthlyCSV — newest observations; full series with a Pro keyAPI
H100 (CoreWeave): $6.16/hr (Q2 2026) — coreweave.com/pricing, Apr 2026 (confirmed). Basis: list price. unchanged vs Q1 2026.
AI Chip Cloud Spot Price — latest readings
15 observations across 5 series, Q2 2023–Q2 2026.
| Series | Latest ($/hr) | Source |
|---|---|---|
| H100 (Spot Market)Q2 2026 | 2.12 | Our model estimateSilicon Analysts modelApr 2026 |
| H100 (CoreWeave)Q2 2026 | 6.16 | coreweave.com/pricingApr 2026 |
| B200 (CoreWeave)Q1 2026 | 8.6 | coreweave.com/pricing (archive.org 2026-01-06, 2026-03-13)Mar 2026 |
| B200 (Lambda)Q2 2026 | 5.5 | lambdalabs.com/service/gpu-cloud/pricingApr 2026 |
| H100 (Lambda)Q2 2026 | 3.99 | lambdalabs.com/service/gpu-cloud/pricingApr 2026 |
Hourly rental cost for AI GPUs on major cloud platforms — CoreWeave, AWS, and spot markets (Vast.ai/Lambda). Tracks the H100→B200 generational transition and supply-driven price volatility.
B200 (Lambda): $5.5/hr (Q2 2026), lambdalabs.com/service/gpu-cloud/pricing, Apr 2026
2023–2024: 2 earlier prints — full ledger is Pro
- 1Q4 2023 — Peak H100 Scarcity: H100 spot premium at peak. On-demand prices 3-4× above eventual steady-state.
- 2Q1 2025 — B200 Cloud Launch: Blackwell reaches the rental market: CoreWeave lists GB200 NVL72 at $10.50/GPU-hr; B200 analyst estimates ~$6.50/hr. H100 spot rates keep falling while CoreWeave holds list pricing.
Spot
2023–2024: 3 earlier prints — full ledger is Pro
Source: coreweave.com/pricing, Silicon Analysts model, lambdalabs.com/service/gpu-cloud/pricing, coreweave.com/gpu-cloud-pricing, Vast.ai tracker, Analyst estimates · Basis: list price; spot · Anonymous view: the newest 1–3 observations per series (Nov 2024–May 2026); 5 earlier prints back to 2023 — full ledger is Pro.
Prints ledger
| Note and copy | |||||||
|---|---|---|---|---|---|---|---|
| Q2 2026 | H100 (Spot Market)Our model estimate | 2.121.99–$2.25/hr | |||||
Source: Silicon Analysts (derived), Apr 2026 · Basis: spot · Estimate Median of spot observations in the Apr 2026 snapshot: GCP a3-highgpu-8g $18.00/8 = $2.25/GPU-hr; RunPod H100 spot $1.99/GPU-hr. | |||||||
| Q2 2026 | H100 (CoreWeave) | 6.16 | |||||
Source: coreweave.com/pricing, Apr 2026 · Basis: list price · Confirmed HGX H100 8-GPU instance $49.24/hr = $6.16/GPU-hr, bundle-inclusive (Apr 2026 snapshot, cloudPricing.ts; unchanged on the live page as of 2026-07-10). Corrected 2026-07-10: prior value $4.76 was the retired GPU-component rate (still listed only on legacy /pricing/classic). | |||||||
| Q2 2026 | B200 (Lambda) | 5.5 | |||||
Source: lambdalabs.com/service/gpu-cloud/pricing, Apr 2026 · Basis: list price · Confirmed 8x B200 SXM on-demand $44.00/8 = $5.50/GPU-hr public list price (Apr 2026 snapshot, cloudPricing.ts). 1.38x Lambda H100 for 2.27x the FP8 sparse throughput. | |||||||
| Q2 2026 | H100 (Lambda) | 3.99 | |||||
Source: lambdalabs.com/service/gpu-cloud/pricing, Apr 2026 · Basis: list price · Confirmed 1x H100 SXM on-demand public list price (Apr 2026 snapshot, cloudPricing.ts). | |||||||
| Q1 2026 | H100 (Spot Market)Our model estimate | 1.20.9–$1.6/hr | |||||
Source: Silicon Analysts estimate, Mar 2026 · Basis: spot · Estimate H100 commoditizing; B200 commands 3-4× premium | |||||||
| Q1 2026 | H100 (CoreWeave) | 6.16 | |||||
Source: coreweave.com/pricing (archive.org 2026-01-06, 2026-03-13), Jan–Mar 2026 · Basis: list price · Confirmed HGX H100 8-GPU instance $49.24/hr = $6.16/GPU-hr. Basis note: since Jan 2025 CoreWeave quotes bundled instances (vCPU/RAM/local storage included), so this is not directly comparable to the 2023–24 GPU-component points ($4.76 + adders ≈ $6.2 all-in — effectively flat). Corrected 2026-07-10: prior value $1.60 was an SA estimate of market effective rates, not a CoreWeave price. | |||||||
| Q1 2026 | B200 (CoreWeave) | 8.6 | |||||
Source: coreweave.com/pricing (archive.org 2026-01-06, 2026-03-13), Jan–Mar 2026 · Basis: list price · Confirmed HGX B200 8-GPU instance $68.80/hr = $8.60/GPU-hr (bundle-inclusive). Corrected 2026-07-10: prior value $4.50 was an SA estimate not supported by CoreWeave’s published list. | |||||||
| Q1 2025 | B200 (CoreWeave) | 6.55.5–$8/hr | |||||
Source: Analyst estimates, Mar 2025 · Basis: list price · Estimate Early Blackwell rental estimate. Clarified 2026-07-10: CoreWeave published no HGX B200 list price in Q1 2025 — its first Blackwell listing was GB200 NVL72 at $42/hr for 4 GPUs ($10.50/GPU-hr, archive.org 2025-03-05). Treat as market estimate, not a CoreWeave list price. | |||||||
| Q4 2024 | H100 (Spot Market) | 1.8 | |||||
Source: Vast.ai tracker, Dec 2024 · Basis: spot · Estimate H100 spot approaching pre-scarcity levels | |||||||
| Q4 2024 | H100 (CoreWeave) | 4.76 | |||||
Source: coreweave.com/gpu-cloud-pricing (archive.org 2024-12-06), Dec 2024 · Basis: list price · Confirmed HGX H100 list still $4.76/GPU-hr — CoreWeave held list price while spot rates fell. Corrected 2026-07-10: prior value $1.89 was a spot-market tracker figure (see the H100 Spot Market series). | |||||||
| H100 (Spot Market): 3 earlier prints, Q2 2023–Q2 2024 — full ledger is Pro | |||||||
| H100 (CoreWeave): 2 earlier prints, Q4 2023–Q2 2024 — full ledger is Pro | |||||||
| Showing the newest 1–3 observations per series (Q4 2024–Q2 2026); 5 earlier observations (Q2 2023–Q2 2024) are in Pro. | |||||||
Provenance
Cloud provider public pricing pages, spot market trackers (vast.ai, cloud-gpus.com), and industry procurement data. On-demand prices are list prices without committed use discounts. Enterprise reserved pricing is 30–50% lower.
Basis: spot · list price
Published on-demand rate card, no committed-use discount. Enterprise reserved pricing runs 30-50% lower, so on-demand and reserved rates must not be charted as one series.
Rule: cross-series comparisons must be same-basis.
Sources
- Vast.ai market data (Jun 2023)
- coreweave.com/gpu-cloud-pricing (archive.org 2023-12-13) (Dec 2023)
- Vast.ai / Lambda (Dec 2023)
- cloud-gpus.com aggregator (Jun 2024)
- coreweave.com/gpu-cloud-pricing (archive.org 2024-06-18) (Jun 2024)
- coreweave.com/gpu-cloud-pricing (archive.org 2024-12-06) (Dec 2024)
- Vast.ai tracker (Dec 2024)
- Analyst estimates (Mar 2025)
- coreweave.com/pricing (archive.org 2026-01-06, 2026-03-13) (Jan–Mar 2026)
- lambdalabs.com/service/gpu-cloud/pricing (Apr 2026)
- coreweave.com/pricing (Apr 2026)
Our derivations
- Silicon Analysts estimate (Mar 2026)
- Silicon Analysts (derived) (Apr 2026)
Suggested citation
Silicon Analysts, "AI Chip Cloud Spot Price", as of Apr 2026, https://siliconanalysts.com/market-data/ai-chip-cloud-price, retrieved September 28, 2026.
Showing the newest 1–3 observations per series (Q4 2024–Q2 2026); 5 earlier observations (Q2 2023–Q2 2024) are in Pro.
Related Tools
Related Analysis
Market intelligence on this data
All market intelligence →Samsung HBM4 Yield Climbs to ~80%, Wafer Starts Up 40% as HBM4 Pricing Projected +65%
Yield: <60% → ~80% (+20pp); Capacity: Samsung HBM wafer starts +40% (180k→250k wpm), HBM4-family mix 40%→80%; Price: HBM4 ASP +65% YoY; Share shift: Samsung 21%→33%, SK hynix 58%→50% (Counterpoint, Q1→Q2)
HBM3E Contract Prices Surge ~20% for 2026; Server DRAM Spikes 90-95% QoQ as Supply Crunch Intensifies
HBM3E 2026 contract price: ~+20% full-year vs. prior contracts (Samsung, SK Hynix per TrendForce); Server DRAM contract price QoQ: +90–95% Q1 2026, +58–63% Q2 2026, +13–18% Q3 2026 (TrendForce); DDR4 commodity spot: +249% Nov 2025 to Aug 2026 ($12.18 → $42.50). Est. BOM impact on H200/B200-class GPU: HBM cost line up ~+20% YoY on contracted volumes.
SK Hynix Breaks Ground on Indiana HBM Fab: $4B US Domestic Capacity Confirmed
CapEx commitment: ~$4B USD; production ramp targeted 2029; near-term (2026–2028) HBM supply capacity unchanged.
Nvidia Flags 'Extreme' Memory Costs Compressing Margins; Supply Meets Only 70% of Demand — HBM Shortage Structural, Not Transient
Gross margin pressure: Nvidia guiding below current ~75% GM due to memory cost overage; supply gap: ~30% of current customer demand unmet; FY2028 revenue upside capped by HBM supply constraint (analyst Stacy Rasgon, Bernstein, quantifies constrained vs. unconstrained delta as implying >$200B in suppressed revenue potential).