General Compute
Category: AI Infrastructure
General Compute is an AI inference neocloud that runs on purpose-built ASIC chips (SambaNova) to deliver faster and more cost-efficient AI inference than GPU-based clouds. General Compute was founded in 2025. The company is led by Finn Puklowski. Based in Covina (Los Angeles area), California, United States. Team size: 11-50. Total funding raised: $415M. Latest round: Debt. Key investors include FUSE VC, Upper90 Capital Management, Village Global Ventures, Carya Venture Partners, NZVC, Matterscale Ventures, Mana Ventures, Evercrest Capital Partners, Bullock Capital.
- Founded
- 2025
- Headquarters
- Covina (Los Angeles area), California, United States
- Team size
- 11-50
- Total funding
- $415M
Value proposition
World's fastest inference cloud running on ASICs instead of GPUs, delivering up to 16x faster inference, 8.5x higher output throughput, 7x faster time-to-first-token, and 6x better power efficiency than standard GPU clouds. Air-cooled chips deploy in weeks (not years) in existing data centers. OpenAI-compatible API for zero-code migration.
Products and solutions
1) API - OpenAI-compatible REST endpoints for AI inference, 2) Dedicated Deployments - Custom capacity with SLAs and guaranteed throughput, 3) BYO (Bring Your Own Model) - Ship custom weights on General Compute's optimized ASIC stack. All running on SambaNova SN40/SN50 inference chips.
Unique value
First ASIC-native neocloud built on SambaNova's specialized inference chips (SN50), delivering 600-700 tokens/second vs ~250 for GPUs. Air-cooled, 20kW per rack (vs 120kW for GPUs), no water cooling needed. Deploys in existing colocation facilities in weeks rather than multi-year GPU data center builds.
Target customer
AI developers, AI agent builders, enterprises running production inference workloads, coding agent platforms, voice AI applications, and any organization needing low-latency, high-throughput AI model serving.
Industries served
AI Agents, Coding/Developer Tools, Voice AI, Enterprise AI, Autonomous Agents, Real-time AI applications
Technology advantage
Exclusive neocloud partnership with SambaNova for SN50 chips ($300M+ chip order). Air-cooled ASIC architecture (20kW/rack vs 120kW for GPUs) enables rapid colocation deployment. No GPU allocation to protect — fully independent from NVIDIA ecosystem. First to deploy SambaNova SN50 chips at scale. Up to 1,000 tokens/second throughput.
How they differentiate
Unlike GPU neoclouds (CoreWeave, Lambda, etc.) that are tied to NVIDIA's roadmap and require water-cooled infrastructure, General Compute is built entirely on SambaNova's inference-optimized ASICs. This gives them 16x faster inference, air-cooled deployment in weeks, 6x better power efficiency, and independence from NVIDIA's supply chain and pricing.
Main competitors
CoreWeave (GPU neocloud), Groq (now NVIDIA-owned, inference ASICs), Cerebras (inference chips, went public $57B IPO), TensorWave (AMD-based inference cloud), Fireworks AI (inference platform), OpenRouter (multi-model inference gateway)
Key partnerships
SambaNova Systems (exclusive chip supply partnership, $300M+ SN50 chip order), Upper90 Capital Management ($400M debt facility and equity investor), Colocation partnerships with data center providers and crypto miners repurposing infrastructure
Major milestones
May 2026: Raised $15M seed round at $60M valuation led by FUSE VC, Secured $300M+ of SambaNova SN50 chips on order, Launched cloud inference platform, Processed 108M tokens on Day 1, 590M on Day 2, 3B tokens/day within first week. July 2026: Secured up to $400M debt facility from Upper90 (first deal using inference chips as collateral), Announced 16x faster inference than GPU clouds.
Growth metrics
Processed 108M tokens on launch day, 590M tokens on Day 2, 3B tokens/day within first week of operation (May 2026).
Market positioning
Early-stage inference neocloud positioning itself as the fastest and most cost-efficient alternative to GPU-based clouds for AI inference. Targeting the rapidly growing inference market as AI shifts from training to inference/agent workloads. Positioned as the "CoreWeave for inference" with SambaNova instead of NVIDIA.
Geographic focus
United States (headquarters in Covina, CA; operations in San Francisco, CA). Also has team presence in Brazil (Curitiba), Paraguay (Asuncion), New Zealand (Auckland), and Los Angeles.
About Finn Puklowski
Chairman of Fluency Academy (EdTech, backed by $50M from General Atlantic); Founder of Zeno Property Limited (real estate development, New Zealand); Former consultant at Clarke Group (apartment development). Education: Harvard Business School (Endeavor Scaling Entrepreneurial Ventures Program, Strategy Execution); MIT Sloan + CSAIL (AI: Implications for Business Strategy); Stanford (Endeavor Innovation & Growth Program).
Latest news about General Compute
More AI Infrastructure companies
Official website: https://www.generalcompute.com