Vector Core Compute

Vector Core Compute (VC2) is an enterprise inference cloud from Vista Equity Partners and Cambium Capital. It splits AI inference across Intel Xeon CPUs, SambaNova SN40 RDUs and NVIDIA Blackwell GPUs instead of running everything on GPUs. It targets AI providers and enterprises whose agentic workloads need cheaper, faster inference at high volume.

Vector Core Compute (VC2) is a purpose-built enterprise inference cloud formed by Vista Equity Partners and Cambium Capital and launched in early June 2026 at Computex. It is billed as the first commercially available cloud built on fully disaggregated inference. Rather than running a GPU-only stack, VC2 sends each stage of an inference request to a different kind of processor. Intel Xeon 6 CPUs handle orchestration, workflow routing and tool execution for agentic workloads. NVIDIA Blackwell GPUs handle the compute-heavy prefill stage and prompt caching. SambaNova SN40 Reconfigurable Dataflow Units (RDUs) handle decode, which is token generation at scale. The platform is backed by a $3.5 billion compute commitment to SambaNova and is supported by Intel. SambaNova's CEO called it the largest commercial deployment of SambaNova technology in the company's history. According to IT Brief, independent measurement by Artificial Analysis found the architecture was at least two to three times faster than a GPU-only stack. Intel says it delivered the fastest enterprise inference recorded on the MiniMax 2.5 model. The first site is live in Los Angeles, with Chicago, Seattle and Phoenix in development. The plan is to cover more than 50 US metros by converting existing data centres into inference-focused sites. Together.ai, which serves about 400 trillion tokens a month, is the first commercial customer. Vista has secured early access for its 90+ portfolio companies. VC2 competes with specialist inference clouds and GPU neoclouds.

Category

Infrastructure & Cloud

Pricing

Contact for pricing

Target Market

CTOs, CIOs, AI Platform Teams, Enterprise Developers

Deployment

Cloud-only

Founded

2026

• ✓
  Disaggregated inference architecture

  Splits each inference request across CPUs, RDUs and GPUs so every stage runs on the silicon best suited to it, instead of a one-size-fits-all GPU stack.

• ✓
  SambaNova SN40 RDU decode

  Runs token generation (decode) on SambaNova Reconfigurable Dataflow Units, backed by a $3.5 billion compute commitment, to raise output speed at scale.

• ✓
  NVIDIA Blackwell prefill and prompt caching

  Uses Blackwell GPUs for the compute-heavy prefill stage and prompt caching, where GPUs remain strongest at processing long incoming prompts.

• ✓
  Intel Xeon 6 orchestration

  Xeon 6 CPUs handle orchestration, workflow routing and tool execution, which agentic workloads lean on between model calls.

• ✓
  Metro-distributed inference sites

  Converts existing data centres into inference-focused sites, live in Los Angeles with Chicago, Seattle and Phoenix in development, to place capacity near US metros.

• •
  Inference capacity for AI platform providers

  An inference provider like Together.ai, serving about 400 trillion tokens a month, adds VC2 capacity to run open models for its agentic customers.

• •
  Agentic code generation and customer service

  Enterprises run multi-step coding and customer-service agents on VC2. Faster decode and CPU-side tool execution cut end-to-end response time.

• •
  Claims handling and high-volume data processing

  Insurers and data-heavy businesses push large document and claims volumes through agent pipelines and pay less per task than on GPU-only stacks.

• •
  Always-on monitoring agents

  Continuously running monitoring agents need steady low-cost inference. VC2 targets this with disaggregated hardware tuned for sustained token generation.

• •
  Portfolio-wide AI rollout for Vista companies

  Vista's 90+ portfolio software companies use early access to VC2 to add agentic features without building their own inference infrastructure.

No list pricing was published at launch, and neither the vendor site nor the launch coverage gives a metering model (per token, per node or reserved capacity). Access appears to be sold as negotiated enterprise capacity, and Vista portfolio companies get early access.

This page was written from 6 sources, 5 on domains other than vectorcorecompute.com.

1. 1.vectorcorecompute.com — vectorcorecompute.comvendor
2. 2.vistaequitypartners.com — vista equity partners and cambium launch vector core compute
3. 3.itbrief.co.uk — vista launches vector core compute for ai inference
4. 4.siliconsemiconductor.net — Vista Equity Partners and Cambium launch Vector Core Compute
5. 5.aithority.com — vista equity partners and cambium launch vector core compute
6. 6.intel.com — intel announces new ai innovations at computex