How SF Compute Company, Foundry, Fluidstack, Nscale, Together, RunPod, Modal, Vast.ai, Prime Intellect, Bittensor, Compute Exchange, Silo AI.
AI-native GPU-capacity-markets, spot-GPU-marketplaces, spot-inference-pricing, compute-derivatives, compute-futures, GPU-hour-forwards, reserved-instance-secondaries, neocloud-aggregators, decentralized-compute is one of the highest-signal fintech-meets-infrastructure categories for 2026 — validated by chronic H100, H200, B200, GB200, GB300, MI300X, MI325X, MI350, TPU-v5p, TPU-v6-Trillium, Trainium2 shortages, hyperscaler multi-year reservation gluts sitting idle overnight, and the emergence of San Francisco Compute Company, Foundry, Fluidstack, Nscale, Together, RunPod, Modal, Vast.ai, Lepton (Nvidia-acquired), Prime Intellect, Bittensor, Compute Exchange, Silo AI (AMD-acquired), Sesterce, Yotta, Crusoe as the credible-scale exchanges, aggregators, spot-marketplaces, decentralized-compute-networks, inference-serving-platforms rewriting how AI teams actually source GPU-hours. Founders here underwrite on GW-under-contract, GPU-hour throughput, price-discovery liquidity, time-to-first-token, tokens-per-dollar, utilization uplift over reserved baselines, and the enforceable-contract, settlement, custody, escrow stack — not vanity gross-capacity claims.
Four unlocks: (1) H100, H200, B200, GB200, GB300, MI300X, MI325X, MI350, TPU-v5p, TPU-v6-Trillium, Trainium2 remain persistently supply-constrained while multi-year hyperscaler reservations sit under-utilized overnight and on weekends — the exact structural gap SF Compute Company batch-auction, Foundry per-second billing, Compute Exchange forward-book, Fluidstack, Nscale, RunPod, Vast.ai, Prime Intellect were built for. (2) Nvidia acquired Lepton AI Jul-2025 to formalize the DGX-Cloud-Lepton marketplace, AMD acquired Silo AI Aug-2024 for €665M, Nvidia acquired OctoAI Oct-2024, and CoreWeave IPO'd Mar-2025 $NASDAQ:CRWV $23B+, Nebius $NASDAQ:NBIS $10B+, Cerebras $NASDAQ:CBSN filed-Sep-2024 — the credible-scale investable, hyperscaler-strategic, government-underwritten public-comp bar is finally set. (3) GPU-backed project finance, ABS is now real capital-stack: Blackstone-CoreWeave $7.6B, Blue-Owl-Crusoe $3.4B, Macquarie-Lambda $500M, Applied-Digital-NVIDIA-preferred, Nebius $700M-Sep-2024, plus new GPU-backed ABS structures rated by KBRA, Fitch making forward-capacity, reserved-secondary, spot-clearing financeable and derivatives-ready. (4) The BIS Jan-15-2025 AI-diffusion rule (rescinded May-2025), Jan-13-2025 export-control update, Dec-2-2024 SME, Oct-2023, Oct-2022 controls, CHIPS-Act Section-9902/9903/48D/45X, FEOC created hard compliance moats that reward operators with real settlement, custody, know-your-buyer, end-use-attestation infrastructure — a compliance stack most peer-to-peer and decentralized alternatives cannot meet for enterprise workloads.
Seed $3-15M for matching-engine, supplier-onboarding, settlement, first-hyperscaler-reserved-secondary, first-neocloud-listing, first-enterprise-anchor (SF Compute Company Sequoia, Lightspeed, Foundry a16z, Sequoia, Lightspeed $80M+ Series A-Jul-2024, Modal Redpoint, a16z $16M-Series-A, $80M-Series-B, Fluidstack strategic, Prime Intellect Founders Fund, Menlo, CoinFund $15M-Series-A-Feb-2025). Series A/B $15-500M for GW-under-contract, hyperscaler-strategic, Nvidia/AMD-preferred, GPU-backed-ABS, reserved-instance-clearing (Foundry a16z-Sequoia-Lightspeed, Together AI $305M-Series-B Salesforce-Coatue-Kleiner-Nvidia $3.3B post-Feb-2025, Fireworks AI Sequoia, Benchmark $52M-Series-B, $552M-secondary-2025 $5B+ valuation, Nscale $155M-Series-A-Sep-2024, $700M-strategic-Aug-2025, multi-GW UK-Norway pipeline, Sesterce EU-multi-GW, Crusoe Founders Fund-Blackstone-Blue Owl $686M-Series-D-Dec-2024, Applied Digital $NASDAQ:APLD NVIDIA-preferred, Macquarie-CIM $5B-Sep-2024, Lambda Macquarie $500M, $480M-Series-C, Voltage Park private-Andy-Jassy-adj). Growth, IPO, strategic $200M-$50B+ (CoreWeave $NASDAQ:CRWV $23B+ post-Mar-2025-IPO, Blackstone $7.6B project-finance, $1.5B-Series-C-May-2024, Nebius $NASDAQ:NBIS $10B+ post-Yandex-spinoff, $700M-Sep-2024-Nvidia-Accel-Orbis, Cerebras $NASDAQ:CBSN filed-Sep-2024 $8B+ TBD, G42 $335M-Aug-2024, Groq $640M-Series-D-Aug-2024 $2.8B, BlackRock-Cisco-Neuberger, SambaNova $5B+, Lepton AI Nvidia-acquired-Jul-2025, OctoAI Nvidia-acquired-Oct-2024 ~$250M, Silo AI AMD-acquired-Aug-2024 €665M, MosaicML Databricks-acquired-Jul-2023 $1.3B). Non-dilutive material — CHIPS-Act Section-9902/9903 $52.7B, Section-48D 25%-ITC, Section-45X advanced-manufacturing, DOE Loan Programs Office LPO Title-17, $6.6B-TSMC-Arizona, $6.4B-Samsung-Taylor, $8.5B-Intel, IRA Section-45X, EU Chips Act €43B, UK AI Growth Zones, Norway Freyr-adj, BIS Validated-End-User VEU, DOE OCED, ARPA-E, NSF NAIRR, DoD Trusted-Capital, DIU, In-Q-Tel, SBIR-STTR non-dilutive-government, $10M-$8B+ ceiling.
Overclaiming GW-under-contract, GPU-hour throughput, utilization uplift, tokens-per-dollar, settlement integrity, hyperscaler-partnership without third-party-audited, MLPerf-Inference, LMSYS-Arena-serving, artificial-analysis.ai, KBRA/Fitch-rated-ABS, Nvidia-DGX-Lepton, AMD-Instinct-preferred, hyperscaler-signed, enterprise-anchor, published-uptime, on-chain-attested track-record — post-CoreWeave-Mar-2025-IPO, Nebius-2024-Yandex-spinoff, Cerebras-Sep-2024-filing, Lepton-Jul-2025-Nvidia, Silo-AI-Aug-2024-AMD, OctoAI-Oct-2024-Nvidia credible-scale bar, operators without signed hyperscaler, Nvidia, AMD, top-5-enterprise-anchor, GPU-backed-ABS-rated-capital-stack, real-settlement-and-custody get flagged by Founders Fund, a16z, Sequoia, Lightspeed, Coatue, Insight, Benchmark, Accel, Index, GV, NEA, Bessemer, Nvidia NVentures, AMD Ventures, Blackstone, Blue Owl, Apollo, Macquarie, DigitalBridge, Stonepeak, EQT diligence within 12-18 months. Underestimating BIS Oct-2022, Oct-2023, Dec-2-2024, Jan-15-2025-AI-diffusion, Jan-13-2025-export-control, ECCN-3A090, 4A090, entity-list-Huawei-SMIC, VEU-Section-742.6, CHIPS-Act-Section-9902-9903, Section-48D, Section-45X, FEOC-Section-30D, IRA, ITAR, EAR, Wassenaar, CFIUS, Team-Telecom, EO-14179, OMB-M-25-21-22, Nvidia-EULA, AMD-EULA, hyperscaler-TOS-resale-legality, right-of-use vs. transfer-of-title, know-your-buyer, end-use-attestation regulatory-and-compliance. Ignoring power, interconnection-queue, PPA, behind-the-meter, geothermal, SMR, gas-peaker, battery-ESS, MISO-ERCOT-PJM-CAISO-SPP-queue, FERC-Order-2023, FERC-Order-2222, IRA, IIJA, hyperscaler-PPA-Meta-Google-Microsoft-Amazon, Fervo, Oklo, X-energy, Kairos, Last-Energy, Base-Power, VEIR, Heron, Astrania power-and-interconnection reality — GPU capacity without firm power, interconnection is unbookable.
Investor directory · Fundraising library · Articles A–Z · Company funding database