Voltage Park

Crypto-billionaire-funded GPU cloud with ~24k H100s and a non-profit positioning — built by Jed McCaleb's Navigation Fund as a public-benefit alternative to commercial neoclouds.

1. Core Product / Service

GPU cloud serving H100 capacity to AI startups and research labs; structurally distinguished by being funded and majority-owned through Jed McCaleb's Navigation Fund (a 501(c)(3) public-benefit foundation funded with his crypto wealth) rather than VC equity [1][2].

Stack and capabilities:

• ~24,000 NVIDIA H100 SXM5 GPUs across multiple US data centers (initial Texas + expansion sites) [3].
• InfiniBand-interconnected clusters ranging from 8-GPU dev nodes to thousands-of-GPU training clusters.
• On-demand + reserved bare-metal offerings; Slurm-based scheduling on larger clusters.
• Bare-metal + virtualized modes; Kubernetes orchestration optional.
• Smaller surface area than coreweave / nebius — pure GPU rental, no managed model APIs, no agent / inference platform.

2. Target Users & Pain Points

• AI startups that can't get H100 capacity at AWS/Azure list prices and don't want to wait in queue.
• Research labs / academic groups — Voltage Park's non-profit positioning makes it palatable for institutions that prefer not to fund a for-profit middleman.
• Open-source model trainers / fine-tuners — large-cluster H100 access for shorter-than-frontier-lab training runs.
• AI safety / alignment research — Navigation Fund's stated mission includes supporting AI safety, so safety labs are a target customer cohort.

Pain points solved: H100 access at predictable below-market pricing, simpler procurement than hyperscalers, ideological alignment for buyers who care about non-profit governance.

3. Competitive Landscape

Voltage Park's edge: non-profit governance (no shareholder pressure to mark up indefinitely), patient capital (founder-funded, not VC), single-mission focus on H100 rental. Disadvantage: scale is small relative to CoreWeave/Nebius; no Blackwell capacity disclosed; missing managed inference / Bedrock-style upsell layer.

4. Unique Observations

• Pricing — the main public differentiator. H100 8-GPU on-demand listed at $2.50/H100·hour (long-term reserved); on-demand short-term ~$2.99–$3.99/H100·hour [6]. This is among the lowest H100 list pricing in the US market — well below AWS Capacity Block list ($12/H100·hour) and even below typical CoreWeave list. Reserved 1-year ~$1.99/H100·hour. Voltage Park doesn't break out H200 / B200 pricing publicly; the available SKU as of 2026-05 is H100 only.
• Capacity source = NVIDIA H100 only. No Blackwell, no Trainium, no TPU. The ~24k H100 cluster was procured in 2023 with ~$500M of Navigation Fund money [3][7]. The fleet has not been publicly disclosed to have grown materially since launch — Voltage Park has been more focused on utilization/SaaS-ization of existing capacity than aggressive Capex like CoreWeave/Nebius.
• Non-profit positioning is genuinely structural, not marketing. Voltage Park is owned by Navigation Fund, a public-benefit 501(c)(3) [5]. Surplus revenue funds the foundation's broader mission (AI safety, scientific research, education) rather than VC distributions. This is the only at-scale GPU cloud structured as a non-profit. The implication: pricing pressure to maximize margin is genuinely lower than at any VC-backed competitor.
• Crypto-wealth → AI-compute pattern. Jed McCaleb's wealth comes from co-founding Mt. Gox, Ripple, and Stellar (XLM); this is a recurring 2023–2025 archetype where crypto exits fund AI infrastructure (similar dynamics with smaller crypto founders backing Lambda, Together, etc.). Worth noting Jimmy's hermes-agent / personal-AI work also intersects with this same crypto-exit-to-AI archetype.
• AI revenue share / customer concentration: Voltage Park is private and discloses neither revenue nor customer mix. SemiAnalysis ClusterMAX 2.0 mentioned them as a credible regional cluster [4]; not in the top tier (CoreWeave Platinum, Nebius Gold) but rated as solid mid-market. Customer concentration is plausibly very different from CoreWeave's — no big hyperscaler anchor disclosed, more likely a long-tail of mid-size AI startups + research labs. The non-profit governance both enables (no need to lock in giant single-customer deals) and limits (no scaling-via-debt-against-Microsoft-contract playbook) the model.
• Strategic question: at sub-$3/H100·hour list and a fixed cluster, the unit economics depend entirely on utilization. If the H100 generation is rapidly displaced by Blackwell/Rubin, Voltage Park's lack of new-gen capex creates a glide path down rather than up. The non-profit structure protects from short-term shareholder pressure but doesn't change the fundamental refresh-cycle math.

5. Financials / Funding

• Founder funding: ~$500M from Jed McCaleb / Navigation Fund (2023) [2][3].
• Revenue: not disclosed (private, non-profit-owned).
• Cluster cost basis: ~24,000 H100s ≈ $25k–$35k/GPU at 2023 procurement prices ⇒ ~$600M–$840M GPU CapEx alone, plus DC/colo/network — implying initial Navigation Fund support + likely some debt or vendor financing on the margin.
• No subsequent priced equity rounds publicly disclosed; the company is structured to operate on internal cashflow + foundation top-ups.
• Compare: coreweave raised ~$28B equity + debt in 12 months pre-IPO; Voltage Park's ~$500M is < 2% of that. The scale gap is the central thing to understand about this entity.

6. People & Relationships

• Founder / Funder: Jed McCaleb — Mt. Gox (1.0), Ripple co-founder, Stellar co-founder; Navigation Fund principal donor.
• Navigation Fund: 501(c)(3) public-benefit foundation; majority owner of Voltage Park [5].
• Leadership: Voltage Park CEO (Mark Adams as of public reporting); operating team distinct from Navigation Fund staff.
• NVIDIA: GPU supplier (H100 only as of 2026-05).
• Customers: not publicly disclosed in detail; mix of AI startups + open-source projects + research labs.
• Competitors: coreweave, lambda-labs, runpod, nebius, vast-ai.

Sources

[1] https://voltagepark.com/ (2026-05-10) [2] https://www.theinformation.com/articles/billionaire-jed-mccaleb-funded-500-million-gpu-cloud-voltage-park (2026-05-10) [3] https://techcrunch.com/2023/10/30/voltage-park-launches-with-500m-and-24000-h100-gpus/ (2026-05-10) [4] https://www.semianalysis.com/p/clustermax-20-the-industry-standard (2026-05-10) [5] https://navigation-fund.org/ (2026-05-10) [6] https://voltagepark.com/pricing (2026-05-10) [7] https://www.datacenterdynamics.com/en/news/voltage-park-h100-cluster-2024 (2026-05-10)