· Venture Capital Tracker Editorial · investment-strategies
Delos Data Raises $100M+: Inference Networking After the GPU
Intel-veteran-founded Delos Data raised more than $100 million to build Nonstop AI servers, clusters, and a new Data Interface chip it claims delivers 10× lower latency for agentic inference fabrics. Matrix, Playground, and others backed; no valuation disclosed.
Delos Data raised more than $100 million, announced September 15–16, 2026, to scale its Nonstop AI inference hardware and a new Data Interface interconnect. Named backers include Matrix Partners, Playground, Socratic Partners, Capricorn’s Technology Impact Fund, Matter Venture Partners, and IAG, plus industry angels. No valuation and no series letter appeared in FinSMEs or SiliconANGLE coverage. Co-founders Ed Doe (CEO) and Dan Daly (CTO) are Intel veterans (FinSMEs, SiliconANGLE).
Agentic AI spans multiple accelerators. When that happens, performance is decided by how data moves — Delos is raising hardware capital for the interconnect layer, not another GPU.
Key facts
| Field | Detail |
|---|---|
| Company | Delos Data — Palo Alto |
| Round | >$100M; stage letter n/d |
| Product stack | Nonstop AI Server (up to 4 external GPUs via OAM), Clusters (OSFP fabric), Data Interface chip |
| Claim | 10× lower latency and 10× higher efficiency vs rivals (company) |
| Timing | First product debuted ~3 months earlier; Data Interface launched with the raise |
Who uses the product — and for what job
Users: hyperscalers, AI cloud builders, and enterprises composing heterogeneous inference clusters (mixed GPUs/accelerators/CPUs).
Job: move tokens and tensors between devices fast enough for multi-device agent workloads, and recover when a GPU or link fails — without locking the fabric to a single vendor’s closed interconnect.
SiliconANGLE notes each Nonstop AI Server chip has nine OSFP ports at 1.6 Tbps — positioned as network capacity comparable to Nvidia Rubin appliances (press comparison, not a third-party lab).
Why now
- Training spend still dominates headlines; inference volume and agent fan-out are the next bottleneck.
- Open interconnect bets (OAM, OSFP) appeal to buyers tired of single-vendor fabrics.
- Cornelis / optical peers show AI networking is a real check-writing category in 2026.
Why this syndicate — portfolio fit
Matrix and Playground-class capital historically finance infrastructure companies before they look like SaaS. IAG appearing again on AI networking names matches the “systems + silicon angels” pattern FinSMEs described.
Strategic rationale: raise from investors who underwrite multi-year hardware cycles and can introduce hyperscale design wins — not a consumer AI brand fund. We do not invent /fund/ links for Matrix, Playground, or Capricorn.
| Dimension | Fit |
|---|---|
| Stage | Large unnamed equity after first product |
| Thesis | Inference data movement as the agent bottleneck |
| Risk | 10× claims unverified; Nvidia ecosystem lock-in |
What remains undisclosed
- Exact amount above $100M; valuation; series letter
- Design wins / revenue; lead investor designation
- Independent benchmarks vs Nvidia fabrics
The unresolved question: whether Delos becomes a merchant interconnect standard — or a niche server SKU that hyperscalers absorb into custom ASICs.
Follow Venture Capital Tracker in Google
Add VCT as a preferred source to make our venture-capital coverage easier to find in Google Search.
Editorial note: AI tools assisted with research, structure, or drafting. Venture Capital Tracker retains human editorial responsibility for factual accuracy, relevance, and source quality before publication.
