AI pricing pagesWholesale & landlords
Tracking 742 prices · Checked 2 hours ago
What AI costs
Wholesale and landlord prices
What GPU owners are paid by hyperscalers, labs and enterprises on multi-year contracts, what landlords charge for the buildings, and who sells capacity to whom. From SEC filings, company disclosures and SAVRN's contract research.
5disclosed contract rates per GPU-hour
$155.60median shell lease per kW-month, 19 leases
18 of 47data center contracts with no disclosed rate
53GPU clouds and landlords, with their customers
From owning to listing
The same GPU-hour, priced at every step
Each bar is one price for one GPU for one hour. Green is the owner's cost, copper is what a wholesale buyer pays on a multi-year contract, and the dark bars are what providers list. The multiple is against the owner's cost.
NVIDIA GB200 NVL72All GB200 NVL72 prices →
Owner's costSemiAnalysis, owner's cost, July 23, 2026
Wholesale contractHIVE (BUZZ) to Cohere, through Bell AI Fabric
2.0x
Median list price7 list prices
5.8x
Lowest committed priceMicrosoft Azure
6.5x
Highest list priceMicrosoft Azure
16.9x
NVIDIA GB300 NVL72All GB300 NVL72 prices →
Owner's costSemiAnalysis, owner's cost, July 23, 2026
Lowest committed priceAstra Intellectus
1.6x
Wholesale contractHIVE (BUZZ) to Investment-grade enterprise, not named; Nebius to Microsoft
1.7x
Median list price4 list prices
6.1x
Highest list priceFireworks AI
8.5x
NVIDIA Vera Rubin NVL72All Vera Rubin NVL72 prices →
Owner's costSemiAnalysis, owner's cost, July 23, 2026
Wholesale contractNscale to Anthropic, SAVRN estimate
1.6x
What sellers get
Contract rates per GPU-hour
Each disclosed contract's value divided by its GPUs and the hours in its term: what the seller receives before its own costs.
| Seller | Buyer | GPU | Per GPU-hour | Term | Terms as disclosed | Source |
|---|---|---|---|---|---|---|
| Nscale | Anthropic | Vera Rubin NVL72 | N/AGPU count not disclosed | 6 years | About $45B over about 6 years for about 460 MW of Vera Rubin capacity, as reported: $1,358.70 per kW-month | Aker, Aug 26, 2026 ↗ Reuters ↗Aug 26, 2026 |
| Nebius | Microsoft | GB300 NVL72 | $3.97 | 5 years | $17.4B for 100,000 GB300 GPUs over 5 years | Nebius 6-K ↗Sep 8, 2025 |
| HIVE (BUZZ) | Investment-grade enterprise, not named | GB300 NVL72 | $3.96 | 5 years | $350M for 2,016 GPUs over 5 years | HIVE Digital TechnologiesAug 17, 2026 |
| HIVE (BUZZ) | Cohere, through Bell AI Fabric | GB200 NVL72 | $3.63 | 3 years | $220M for 2,304 GPUs over 3 years | HIVE Digital TechnologiesJun 18, 2026 |
| HIVE (BUZZ) | Dell contracts | Blackwell, Dell-supplied | $3.40 | 2 years | $30M for 504 GPUs over 2 years | HIVE Digital Technologies |
What the tokens bill
One Vera Rubin GPU-hour, sold as tokens
Anthropic pays Nscale about $5.83 for an hour of one Vera Rubin GPU. At Claude Fable 5.1 list prices, the tokens that hour produces bill $111.03: 19.0x what Anthropic pays.
$111.03per GPU-hour, billed as tokens $70.0Ma rack-year, every hour sold See the full build-up, on the Vera Rubin GPU page →What landlords charge
The rent that gets quoted, and what the tenant pays
Landlords quote base rent per kilowatt-month. The tenant also pays for its power, the building's taxes, insurance and maintenance, its improvements, and rent that rises every year. Across the 47 leases SAVRN read, $128.00 quoted becomes $261.46 over the term: 2.04 times the quote.
Quotedbase rent, Year 1
$128.00per kW-month
$128.00
Paidall-in, term average
$261.462.04x the quote
$128.00
+$30.71
+$63.92
+$22.40
- Base rent, Year 1$128.00the number everyone quotes
- Rent increases+$30.71the 3% annual escalator in nearly every contract, averaged over the term
- Power+$63.92passed through to the tenant: 730.5 hours x PUE 1.25 x $0.07 per kWh
- Taxes, insurance, maintenance+$22.40triple-net opex, 17.5% of rent from a disclosed NOI margin
- Improvements+$16.43$1.5M per MW over 15 years, market practice rather than a filing
- Network, fiber, cross-connectNot disclosedabsent from all 47 contracts
The bare powered building is $3.40 to $10.60 of the $128.00: WhiteFiber files that rent for a Montreal shell. The rest of the quote is fit-out, cooling, redundancy, capital recovery and margin.
Per critical kilowatt-month. From Twenty Cents to Twenty-Five Dollars, SAVRN's reading of 47 disclosed data center leases. Every lease is below.
| Landlord | Site | Tenant | MW | Term | Contract value | Per kW-month | Basis | Filing |
|---|---|---|---|---|---|---|---|---|
| IRENIREN | Childress, TX / Sweetwater expansionUnder construction | Microsoft | 200 | 5 yr | $9.7B | $808.30 | Full-stack | |
| BitdeerBTDR | A102, MalaysiaService start Q1 2027; target 350 MW by Q1 2028 | Unnamed AI cloud offtaker | 9.5 | 5 yr | ~$400M | $701.80 | Full-stack | |
| Galaxy Digital (Helios)GLXY | Helios Phase I, Dickens TXDelivered | CoreWeave | 133 | 15 yr | ~$1.7B project cost / disclosed rate not specified | $200.00 | Shell | |
| Riot PlatformsRIOT | Rockdale, TXAnnounced Aug 2026 | Anthropic (per Bloomberg / KBW) | 191 | 20 yr | $9.1B | $198.50 | Shell | |
| TeraWulfWULF | Hawesville, KY (Justified Data campus)DC Lease commences 9/30/2026 | Anthropic | 401 | 20 yr | $19B | $197.40 | Shell | sec.gov ↗ |
| WhiteFiberWYFI | MTL-3, Montreal QCPurchase option exercised, WhiteFiber now owns MTL-3 | Cerebras Systems | 5 | 5 yr | CAD 84M (5 yr × CAD 1.4M/mo × 12) | $196.00 | Shell | |
| CoreWeaveCRWV | Various build-to-suit sitesUnder construction | Not disclosed (implied hyperscaler) | 393 | 16 yr | $14.7B (max) | $194.79 | Shell | sec.gov ↗ |
| TeraWulfWULF | Abernathy, TX (JV)Announced | Fluidstack (Anthropic downstream) | 168 | 25 (w/ 20/15 shortening options) | $9.5B | $188.50 | Shell | sec.gov ↗ |
| Core ScientificCORZ | Denton TX / Muskogee OK / Auburn ALActive, being restructured w/ AMD deal | CoreWeave | 382 | 12 yr | ~$10.2B | $185.00 | Shell | |
| WhiteFiberWYFI | NC-1 campus, NCAnnounced | Nscale | 40 | 10 yr | $865M | $180.20 | Shell | |
| Hut 8HUT | River Bend, LouisianaAnnounced Dec 17, 2025 | Fluidstack (Anthropic downstream) | 245 | 15 yr | $7.0B | $158.70 | Shell | sec.gov ↗ |
| CleanSparkCLSK | Sandersville, GeorgiaDeliveries begin Q4 2027 | High-IG global tech company (unnamed) | 175 | 20 yr | $6.6B base | $157.10 | Shell | |
| TeraWulfWULF | Lake Mariner CB-5 (expansion)Announced | Fluidstack (Anthropic downstream) | 160 | 10 (25 with options) | $3.0B | $156.30 | Shell | |
| Hut 8HUT | Portfolio totalTotal contracted | Fluidstack + IG | 949 | 15 yr | $26.6B contracted | $155.70 | Shell | |
| Applied DigitalAPLD | Ellendale, ND (Polaris Forge 1)PF1-01/02 delivered; PF1-03 nearing | CoreWeave | 250 | 15 yr | ~$7.0B (portfolio pro-rata) | $155.60 | Shell | |
| Hut 8HUT | Beacon Point campus (2 tenants)Announced | IG hyperscaler (×2 tenants) | 704 | 15 yr | $19.6B ($9.8B each) | $154.70 | Shell | |
| TeraWulfWULF | Lake Mariner, NY (CB-1/2/3/4)Executed Aug 13, 2025 | Fluidstack (Anthropic downstream) | 200 | 10 yr | $3.7B | $154.20 | Shell | sec.gov ↗ |
| Core ScientificCORZ | Pecos TX / Muskogee OK / Hunt County TXAnnounced Jul 28, 2026 | AMD | 377 | 15 yr | ~$11B base (analyst est.) | $150.00 | Shell | sec.gov ↗ |
| Cipher MiningCIFR | Barber Lake, Colorado City TX (587 acres)Announced | Fluidstack (Anthropic downstream) | 168 | 10 yr | $3.0B (base) / $7B w/ options | $148.80 | Shell | sec.gov ↗ |
| Cipher MiningCIFR | Black Pearl (site 2)Rent commencement staggered Oct 2026 - Mar 2027 | Amazon (AWS) | 216 | 15 yr | $5.5B | $141.50 | Shell | sec.gov ↗ |
| Core ScientificCORZ | Auburn AL / Dalton3 GAAnnounced Jul 28, 2026 | Undisclosed neocloud | 152 | 15 yr | ~$3B (implied, blended into $14B analyst total) | $140.00 | Shell | |
| Applied DigitalAPLD | Polaris Forge 2 (200 MW site)Announced | Investment-grade hyperscaler | 200 | 15 yr | $5.0B | $138.90 | Shell | |
| Applied DigitalAPLD | Polaris Forge 3 (300 MW site)Announced May 6, 2026 | Investment-grade hyperscaler | 300 | 15 yr | $7.5B | $138.90 | Shell | |
| Applied DigitalAPLD | Delta Forge 2 (210 MW)Announced May 6, 2026 | IG hyperscaler #2 | 210 | 15 yr | $5.2B base / $12.7B w/ renewals | $137.60 | Shell | sec.gov ↗ |
| Cipher MiningCIFR | Stingray, Andrews TXAnnounced | Amazon (AWS) | 70 | 15+ yr | $2.0B base / $2.0B–$5.7B w/ extensions | $133.30 | Shell | sec.gov ↗ |
| Galaxy Digital (Helios)GLXY | Helios Phase II, Dickens TXUnder construction | CoreWeave | 260 | 15 (implied) | n.a. ($10.4B was NOT in filing) | $130.00 | Shell | |
| Riot PlatformsRIOT | Rockdale, TXQ2 2026: 25 MW + 25 MW expansion committed (Nov 2026 + May 2027) | AMD | 25 | 10 yr | $311M base | $103.70 | Shell | sec.gov ↗ |
| Applied DigitalAPLD | Delta Forge 1 (430 MW)Announced Apr 23, 2026 | IG hyperscaler #2 | 430 | 15 yr | $7.5B | $96.90 | Shell | |
| WhiteFiberWYFI | MTL-1 shell rent (landlord side)Active | n.a. | 4 | 15 yr | $13,467-42,394 / month | $3.40 | Powered shell (floor) | |
| Cipher MiningCIFR | Barbers Ferry (site 3)In development | Undisclosed | n.a. | n.a. | n.a. | Not disclosed | ||
| CoreWeaveCRWV | Multi-siteActive | IBM | n.a. | multi-yr | n.a. | Not disclosed | ||
| CoreWeaveCRWV | Multi-siteActive | Meta Platforms | n.a. | 6 yr | up to $14.2B | Not disclosed | ||
| CoreWeaveCRWV | Multi-siteActive | Microsoft | n.a. | multi-yr | n.a. (largest customer) | Not disclosed | ||
| CoreWeaveCRWV | Multi-siteActive | OpenAI | n.a. | 5+ yr | $22.4B (uplift from initial $11.9B) | Not disclosed | ||
| Galaxy Digital (Helios)GLXY | Helios Phase III, Dickens TXOption exercised, brings campus to full 800 MW | CoreWeave | 133 | n.a. | n.a. | Not disclosed | ||
| HIVEHIVE | Merritt, BC (BUZZ)Announced Aug 17, 2026 | Unnamed IG enterprise | n.a. | 5 yr | $350M (~$70M ARR) | Not disclosed | ||
| HIVEHIVE | Merritt, BC (BUZZ)Closed Jun 18, 2026 | Bell AI Fabric / Cohere (sovereign) | n.a. | 3 yr | $220M (~$70M ARR) | Not disclosed | ||
| HIVEHIVE | Manitoba Tier IIIAnnounced | Multiple | n.a. | 2 yr | $30M | Not disclosed | ||
| HIVEHIVE | Greater Toronto AreaPlanned H2 2027 | Sovereign AI pipeline | 320 | n.a. | CAD 3.5B capex | Not disclosed | ||
| IRENIREN | West TexasAnnounced | NVIDIA (compute agreement) | n.a. | 5 yr | $3.4B | Not disclosed | ||
| MARA HoldingsMARA | Multi-site pipelineCapital partnership; no tenant contract | n.a. (no tenant signed) | n.a. | n.a. | n.a. | Not disclosed | ||
| Nebius (US)NBIS | Multi-siteAnnounced | Meta Platforms | n.a. | multi-yr | $3B | Not disclosed | ||
| Nebius (US)NBIS | Multi-site (US + EU)Announced Sep 8, 2025 | Microsoft | n.a. | 5 yr | $17.4B base | Not disclosed | sec.gov ↗ | |
| Oracle Cloud InfrastructureORCL | Oracle CloudDeployment starting Q3 2026 | n.a. (customer offering) | n.a. | n.a. | n.a. | Not disclosed | ||
| Oracle Cloud InfrastructureORCL | Multi-site (Stargate)Starting 2027 | OpenAI | n.a. | 5 yr | ~$300B (WSJ-reported) | Not disclosed | ||
| Riot PlatformsRIOT | Corsicana, TX (756 MW)LOI | Anthropic (KBW attribution) | 756 | n.a. | >$1B/yr rent implied | Not disclosed | ||
| WhiteFiberWYFI | n.a.Announced | Hyperbolic (Modal Labs end customer) | n.a. | 2 yr | ~$17M | Not disclosed |
Who sells to whom
GPU clouds and their customers
U.S. GPU clouds, neoclouds and AI hosts with the capacity they have contracted and the customers named in their disclosures, largest disclosed contracted power first.
| Company | Segment | Fleet | Contracted MW | Disclosed contracts | Customers | Source |
|---|---|---|---|---|---|---|
| Nscale (US)Private | Full-stack AI factory developer | Vera Rubin NVL72 / GB300 / GB200 | 1,350 | Anthropic Monarch $45B/460 MW/6-yr; Microsoft up to $24B; $96B+ contracted | Anthropic, Microsoft | nscale.com ↗ |
| CrusoePrivate | Vertically integrated neocloud | GB200 NVL72; AMD MI300X/MI355X | 1,200 | $15B JV; $7.1B JPMorgan construction loan | Oracle/OpenAI, Microsoft | crusoe.ai ↗ |
| FluidstackPrivate | Capacity aggregator / builder | H100/H200/B200/GB200 (>100k GPUs) | 980 | ~$27B base-term hosting value; Anthropic $50B US build | Anthropic (downstream), Google backstop $1.8B→$3.2B | fluidstack.io ↗ |
| Cerebras SystemsPublic (NASDAQ: CBRS) | Wafer-scale inference cloud | Cerebras WSE + NVIDIA integration | 600 | $20B OpenAI agreement; Digi Power X $1.1B→$2.5B | OpenAI, enterprises | sec.gov ↗ |
| Together AIPrivate | Inference cloud / GPU clusters | H100/H200/B200/B300 | 500 | $1.15B bookings; IBM $240M (B300); Rumble/Quake 180 MW Atlanta | IBM, Rumble/Quake (supplier), enterprises | sacra.com ↗ |
| Rumble / Quake AIPublic (NASDAQ: RUM) | GPU cloud (Together AI supplier) | B300 (via Together AI) | 180 | $270M initial; >$3B run-rate ambition | Together AI | rumble.com ↗ |
| ZyphraPrivate | AMD inference cloud | AMD MI355X (15 MW) | 15 | 15 MW MI355X | Enterprises | zyphra.com ↗ |
| CoreWeavePublic (NASDAQ: CRWV) | AI-native neocloud | H100/H200/GB200/GB300 | n.a. | OpenAI $22.4B; Meta up to $14.2B (6-yr); $21B Meta 2032 ext.; Microsoft largest customer | OpenAI, Microsoft, Meta, IBM, Anthropic | sec.gov ↗ |
| DapplePrivate | Dedicated AI infrastructure (early stage) | n.d. | n.a. | $100M+ contracts (company-cited; unverified) | 2 disclosed customers | dapple.co ↗ |
| DigitalOcean (Paperspace)Public (NYSE: DOCN) | General cloud + GPUaaS | H100/H200/B300; AMD MI300X/MI325X/MI350X | n.a. | $281M Q2'26 revenue; $234M AI customer ARR | 680K+ SMB/dev customers | investors.digitalocean.com ↗ |
| Groq (GroqCloud)Private | Inference cloud (NVIDIA hardware post-Dec 2025) | LPU legacy + NVIDIA systems | n.a. | NVIDIA ~$20B non-exclusive LPU license (Dec 2025) | Developers (6M+), enterprises | groq.com ↗ |
| HyperbolicPrivate | Open GPU marketplace / serverless inference | H200 | n.a. | ~$17M | Modal Labs | hyperbolic.xyz ↗ |
| IBM CloudPublic (NYSE: IBM) | Enterprise cloud | H100/H200; buys Together AI B300 ($240M) | n.a. | IBM–Together AI $240M | Enterprises; CoreWeave top-5 customer | ibm.com ↗ |
| LambdaPrivate | AI-native neocloud | GB300 NVL72 / Hopper / Ampere | n.a. | Microsoft multibillion (tens of thousands GB300); Anthropic $35B reported 31 Aug 2026 | Microsoft, Anthropic | reuters.com ↗ |
| Lightning AI (incl. Voltage Park)Private | Full-stack AI cloud + owned GPUs | ~36,000 Hopper/Blackwell (24k H100 SXM5 ex-Voltage Park) | n.a. | Dedicated contracts 6-12 mo | voltagepark.com ↗ | |
| Nebius (US)Public (NASDAQ: NBIS) | AI-native neocloud | B300/B200/H200/H100 | n.a. | $17.4B Microsoft 5-yr; $3B Meta | Microsoft, Meta | nebius.com ↗ |
| Oracle Cloud InfrastructurePublic (NYSE: ORCL) | Hyperscaler | GB200/GB300; AMD MI355X/MI450 | n.a. | ~$300B OpenAI (WSJ-reported, from 2027) | OpenAI | oracle.com ↗ |
| RunpodPrivate | Developer cloud / aggregator | B200/H200/H100 (aggregated) | n.a. | >$120M ARR | 1M+ developers | runpod.io ↗ |
| TensorWavePrivate | AMD neocloud | AMD MI325X (8,192) / MI355X | n.a. | >$100M ARR | AMD ecosystem, enterprises | tensorwave.com ↗ |
| Theseus InfrastructurePrivate (Anthropic + MAM + GIC JV) | Captive capacity vehicle | n.d. | n.a. | MAM/GIC majority equity (Aug 2026) | Anthropic | macquarie.com ↗ |
| Amazon Web ServicesPublic (NASDAQ: AMZN) | Hyperscaler | P6-B200/B300; GB200 UltraClusters | n.a. | Global | Anthropic (Project Rainier), enterprises | aws.amazon.com ↗ |
| Astra IntellectusPrivate | GPU cloud (long-commit pricing) | GB300 NVL72 / B300 / H200 / H100 | n.a. | n.d. | Enterprises (3-yr contracts) | astraintellectus.com ↗ |
| Cirrascale Cloud ServicesPrivate | Dedicated GPU cloud | NVIDIA/AMD/Cerebras/Qualcomm | n.a. | US-West/East/South DCs | Enterprises, AI labs | cirrascale.com ↗ |
| Denvr DataworksPrivate | GPU cloud | GH200/H100/A100; H200/GB200 reserve-only | n.a. | n.d. | Enterprises, ISVs | denvrdata.com ↗ |
| falPrivate | Generative-media inference | H100 dedicated + serverless | n.a. | n.d. | Media/creative developers | fal.ai ↗ |
| Foundry TechnologiesPrivate | GPU cloud (orchestration) | H100 | n.a. | n.d. | AI labs | datacenterdynamics.com ↗ |
| GMI CloudPrivate | AI-native GPU cloud | GB200/B200/H200/H100 | n.a. | Public cloud; MW n.d. | Developers, APAC enterprises | gmicloud.ai ↗ |
| Google CloudPublic (NASDAQ: GOOGL) | Hyperscaler | B200; TPU | n.a. | Global; backstops Fluidstack leases | Anthropic (TPU), enterprises | cloud.google.com ↗ |
| Massed ComputePrivate | Bare-metal GPU cloud | B300/B200/H200/H100 | n.a. | n.d. | Developers, enterprises | massedcompute.com ↗ |
| Microsoft AzurePublic (NASDAQ: MSFT) | Hyperscaler | ND GB300 v6; H100/H200 | n.a. | Global; buys neocloud capacity (Nebius $17.4B, Nscale ~$24B, IREN $9.7B, Lambda) | OpenAI, enterprises | azure.microsoft.com ↗ |
| NVIDIA DGX CloudPublic (NASDAQ: NVDA) | Vendor cloud (hosted on partners) | DGX GB200/B200 | n.a. | Runs on partner capacity (CRWV, ORCL, MSFT) | Enterprises via NVIDIA | nvidia.com ↗ |
| Prime IntellectPrivate | Compute aggregator + training platform | Aggregated H100/H200/B200 | n.a. | Aggregated supply | AI researchers, agent developers | primeintellect.ai ↗ |
| SaladCloudPrivate | Distributed consumer GPU cloud | Consumer GPUs (60K+ daily active) | n.a. | 450K+ nodes | Inference developers | salad.com ↗ |
| SF ComputePrivate | Compute marketplace | Marketplace (>$100M hardware under management) | n.a. | Marketplace | Labs, startups; capacity sellers | sfcompute.com ↗ |
| ShadeformPrivate | Multi-cloud GPU marketplace | 30+ clouds, 100+ regions | n.a. | Aggregator | Developers, enterprises | shadeform.com ↗ |
| TensorDockPrivate | GPU marketplace | Marketplace | n.a. | Acquired by Voltage Park (per Non-Filers) | Developers | tensordock.com ↗ |
| Vast.aiPrivate | GPU marketplace | 17,000+ GPUs across 350+ hosts | n.a. | 40+ data centers (host-supplied) | Developers; hosts | vast.ai ↗ |
| VultrPrivate | Independent cloud / GPU cloud | NVIDIA B200/H100; AMD MI300X/MI325X/MI355X | n.a. | 32 global locations | SMB/enterprise | vultr.com ↗ |
Landlords and their tenants
Developers and former bitcoin miners that lease powered buildings to GPU clouds and labs.
| Company | Segment | Sites | Contracted MW | Disclosed contracts | Tenants | Source |
|---|---|---|---|---|---|---|
| Applied DigitalPublic (NASDAQ: APLD) | AI-hosting developer (shell lease) | Polaris Forge 1-3 (750) + Delta Forge 1-2 (640) | 1,390 | ~$7.0B CRWV; $5.0B + $7.5B IG hyperscaler; $7.5B + $5.2B IG hyperscaler #2 | CoreWeave, 2 IG hyperscalers | ir.applieddigital.com ↗ |
| Hut 8Public (NASDAQ: HUT) | Miner-to-AI host (shell lease) | River Bend LA 245 (Fluidstack) + Beacon Point 704 (2 IG tenants) | 949 | $7.0B + $19.6B = $26.6B contracted 15-yr | Fluidstack/Anthropic (Google backstop), 2 IG hyperscalers | hut8.com ↗ |
| TeraWulfPublic (NASDAQ: WULF) | Miner-to-AI host (shell lease) | Lake Mariner 360 (Fluidstack) + Abernathy 168 JV + Hawesville 401 (Anthropic direct) | 929 | $3.7B + $3.0B (Fluidstack); $9.5B Abernathy JV; $19B Anthropic 20-yr | Fluidstack, Anthropic, Google backstop $1.3B | investors.terawulf.com ↗ |
| Core ScientificPublic (NASDAQ: CORZ) | Miner-to-AI host (shell lease) | AMD 377 + undisclosed neocloud 152 + CoreWeave ~382 | 911 | ~$11B AMD 15-yr; ~$3B neocloud; ~$10.2B CoreWeave 12-yr | AMD, CoreWeave, undisclosed neocloud | sec.gov ↗ |
| Galaxy Digital (Helios)Public (NASDAQ: GLXY) | AI-hosting developer (shell lease) | Helios Phase I 133 delivered; II 260; III 133 option; 800 MW campus | 800 | Phase I ~$1.7B project; rates redacted | CoreWeave | galaxy.com ↗ |
| Cipher MiningPublic (NASDAQ: CIFR) | Miner-to-AI host (shell lease) | Barber Lake 168 (Fluidstack) + Black Pearl 216 (AWS) + Stingray 70 (AWS) | 454 | $3.0B base/$7B options (Fluidstack); $5.5B AWS 15-yr; $2.0B Stingray | Fluidstack/Anthropic, AWS | theoutpost.ai ↗ |
| Riot PlatformsPublic (NASDAQ: RIOT) | Miner-to-AI host (shell lease) | Rockdale AMD 25→50 + Anthropic 191; Corsicana 756 MW LOI | 241 | $311M AMD 10-yr; $9.1B Anthropic 20-yr; >$1B/yr Corsicana implied | AMD, Anthropic (Bloomberg/KBW attribution) | riotplatforms.com ↗ |
| Ionic DigitalPrivate | Private landlord | 240 MW Texas leased by Nscale for Microsoft (~104,000 GB300) | 240 | Nscale/Microsoft | nscale.com ↗ | |
| IRENPublic (NASDAQ: IREN) | Miner-to-AI host (full-stack) | Childress + Sweetwater; 200 MW MSFT | 200 | $9.7B Microsoft 5-yr; $3.4B NVIDIA compute | Microsoft, NVIDIA | theoutpost.ai ↗ |
| CleanSparkPublic (NASDAQ: CLSK) | Miner-to-AI host (shell lease) | Sandersville GA 175 MW; Texas portfolio exclusivity | 175 | $6.6B base 20-yr (deliveries Q4 2027) | Unnamed high-IG global tech company | cleanspark.com ↗ |
| WhiteFiberPublic (NASDAQ: WYFI) | HPC data center + GPU cloud | MTL-3 5 MW (Cerebras) + NC-1 40 MW (Nscale) + MTL-1 shell | 49 | CAD 84M (Cerebras 5-yr); $865M (Nscale 10-yr); ~$17M Hyperbolic | Cerebras, Nscale, Hyperbolic/Modal | whitefiber.com ↗ |
| Digi Power XPublic (NASDAQ: DGXX) | Small-cap AI host | Columbiana AL 40 MW (Cerebras); RFS Dec 15 2026 | 40 | $1.1B → up to $2.5B | Cerebras | digipowerx.com ↗ |
| Hyperscale DataPublic (NYSE American: GPUS) | Small-cap AI host | Michigan 20 MW → 52 MW option; 10 yr + 2×5 | 20 | >$1.2B (>$3.0B expanded) | Unnamed CA neocloud | hyperscaledata.com ↗ |
| BitdeerPublic (NASDAQ: BTDR) | Miner-to-AI host (full-stack) | A102 9.5 MW; target 350 MW by Q1 2028 | 10 | ~$400M 5-yr | Unnamed AI cloud offtaker | bitdeer.com ↗ |
| MARA HoldingsPublic (NASDAQ: MARA) | Miner pivot (no tenant) | ~1 GW near-term / >2.5 GW ambition; Starwood JV | n.a. | No tenant contract | None signed | mara.com ↗ |
The filings behind the numbers
What the contracts say
Terms as filed with the SEC. Rates are SAVRN derivations from those terms unless marked disclosed, and each links to the document it came from.
| Company | Counterparty | Terms | Rate | Basis | Filing |
|---|---|---|---|---|---|
| Nebius | Microsoft | $17.4B for 100,000 GB300 accelerators over 5 years | $3.97per GPU-hour | Implied | 6-K, Sep 8, 2025 ↗ |
| SpaceX | Anthropic | $1.25B a month through May 2029 for about 325,000 GPUs | about $5.27per GPU-hour | Derived | S-1, May 20, 2026 ↗ S-1/A, Jun 3, 2026 ↗ |
| CoreWeave | Landlord not named | $14.7B maximum, 393 MW, 16-year build-to-suit lease | $194.79per kW-month | Derived | 10-Q, Q2 2026 ↗ |
| TeraWulf | Fluidstack joint venture | $9.5B, 168 MW, 25 years, Abernathy | $188.5per kW-month | Derived | 8-K, Ex. 99.1 ↗ |
| Hut 8 | n.a. | $7B, 245 MW, River Bend | $158.7per kW-month | Derived | 8-K, Dec 17, 2025 ↗ |
| TeraWulf | Fluidstack | $3.7B, 200 MW, Lake Mariner CB-3 and CB-4 | $154.2per kW-month | Derived | 8-K, Ex. 10.1, executed Aug 13, 2025 ↗ |
| Cipher | Fluidstack | $3B, 168 MW, Barber Lake | $148.8per kW-month | Derived | 8-K, Ex. 99.1 ↗ |
| Cipher | AWS | $5.5B, 216 MW, 15 years at 3.0%, Black Pearl | $141.5per kW-month | Derived | 8-K, Ex. 99.1 ↗ |
| Applied Digital | n.a. | $5.2B, 210 MW, Delta Forge 2 | $137.6per kW-month | Derived | 8-K, Ex. 99.1, May 6, 2026 ↗ |
| Cipher | AWS | Full rent schedule filed, Stingray | $133.33per kW-month | Disclosed, Year 1 | 8-K, Ex. 99.1 ↗ |
| Riot | AMD | $311M, 25 MW, 10 years | $103.7per kW-month | Derived | 8-K, Ex. 99.1, Jan 16, 2026 ↗ |
The SAVRN reference set
Seven references, one system
Capital, delay, policy, water, grid, permits and scarcity are one system seen from seven sides. Each reference is primary-sourced, refreshed on a schedule, and links into the others, so a project can be followed from the money committed to the docket that stopped it.
Capital Atlas
Who is funding AI data centers: capital providers, named sites, financing transactions, and the blocked-and-delayed record, state by state and company by company.
Open the atlas →
Delay Watchlist
Project-level slippage: named projects, jurisdictional actions, utility and federal dockets, SEC disclosures, and the interconnection queue.
Open the watchlist →
Moratorium Tracker
The policy layer: every state measure and local moratorium mapped and cited, with one page per state.
Open the tracker →
Water Tracker
Every federal, state, local and interstate water measure that touches a data center, plus a fifty-state water ledger.
Open the water tracker →
Grid Operator Watchlist
ERCOT, PJM, MISO, CAISO, SPP, NYISO, ISO-NE and FERC: the large-load rules, filings and queue positions that decide whether a site can energize.
Open the watchlist →
Permits and Power Development
Data center and substation permits, large loads in the interconnection queues, and the power plants planned to serve them, refreshed hourly from public records.
Open the permits tracker →
SAVRN Scarcity Tracker
Eleven primary-sourced indicators of AI infrastructure scarcity, one composite index, read against the token index, SAVRN's own cost basis and the frontier ceiling.
Open the scarcity tracker →
Have a specific address? Site Intelligence reads all seven references against one location, county, utility and grid operator included, and cites every finding.
Check an address →