After reading this, the reader knows Delos Data reportedly raised $100 million to move data faster inside AI systems, and why memory traffic is becoming a chip priority.

Delos Data raises $100M for AI networking

The Intel-veteran startup is building chips and software for data movement among accelerators, memory and storage in AI data centers.

Delos Data said it raised $100 million to develop networking technology for AI data centers, according to Reuters on September 15.

In plain terms, Delos is targeting the time processors spend waiting for information. AI systems increasingly combine different accelerators, memory tiers and storage devices. Moving data among them can leave expensive compute capacity unused even when the chips themselves are fast.

Why it matters

The AI hardware race is shifting from a focus on arithmetic alone to the whole path that feeds a model. Training and inference workloads must move model weights, prompts, intermediate values and cached context. When the connection cannot supply those bytes quickly enough, adding more processors does not deliver a proportional increase in useful work.

Delos describes its goal as making networks behave like infrastructure for agentic AI. Its website says the system is intended to provide resilient movement among compute, memory and storage. Reuters reported that the company is developing both chips and software and wants to connect heterogeneous accelerators rather than assume every machine uses one vendor’s hardware.

That is a useful target for inference. Agent systems can make repeated model calls, invoke tools and retain long context, producing less predictable traffic than a single batch job. Delos co-founder and chief technology officer Dan Daly told Reuters that uncertainty around future agent architectures makes fast data movement important. The argument is plausible, but product performance has not been independently published.

A large round before public benchmarks

Reuters identified the investors as Matrix, Playground Global, Socratic Partners, Capricorn Investment Group, Matter Venture Partners, IAG Capital Partners and DYNAMIQ. Former Intel chief executive Pat Gelsinger is a Playground partner and was quoted describing idle compute as a waste of money and energy.

The company was founded by former Intel engineers. That background is relevant because building data-center silicon requires chip design, networking and systems-software expertise. It does not establish that the resulting product will outperform existing Ethernet, InfiniBand or proprietary interconnects.

The financing amount is substantial, but the available reporting does not disclose the round’s valuation, product shipment date, customer commitments or measured throughput. Delos’s public website offers a high-level description rather than specifications. Readers should therefore treat the funding as evidence of investor support, not proof of technical performance or demand.

The market problem is also crowded. Accelerator vendors and networking suppliers already optimize links within servers and across clusters. A new entrant must show that its approach reduces latency or total cost under real workloads and fits the software used to schedule models across mixed hardware.

What comes next is unusually concrete: silicon samples, bandwidth and latency measurements, power use, supported protocols, software integrations and named deployments. Until those appear, Delos is a well-funded thesis about the data path around AI chips rather than a demonstrated alternative.

Verification

  1. UNVERIFIED — Delos Data raised $100 million. Reuters reported the announcement, but no primary company or investor release confirming the amount was located during publication. Via: https://www.reuters.com/business/delos-data-chip-startup-founded-by-intel-veterans-raises-100-million-ai-networks-2026-09-15/
  2. VERIFIED — Delos publicly describes its focus as data movement among compute, memory and storage for agentic AI. Primary source: https://delosdata.com/
  3. UNVERIFIED — The company was founded by Intel veterans and is developing chips and software for heterogeneous AI systems. Reported by Reuters; Delos’s public site did not provide matching founder biographies or detailed specifications. Via: https://www.reuters.com/business/delos-data-chip-startup-founded-by-intel-veterans-raises-100-million-ai-networks-2026-09-15/
  4. UNVERIFIED — Matrix, Playground Global, Socratic, Capricorn, Matter, IAG and DYNAMIQ participated in the round. Reported by Reuters without a located primary financing announcement. Via: https://www.reuters.com/business/delos-data-chip-startup-founded-by-intel-veterans-raises-100-million-ai-networks-2026-09-15/
  5. PARTIALLY VERIFIED — Data movement can leave accelerators underused and is central to Delos’s product thesis. Delos states the thesis; no public product benchmark was located. Primary source: https://delosdata.com/
  6. UNVERIFIED — Dan Daly said uncertainty around future agent architectures increases the need for fast data movement, and Pat Gelsinger linked idle compute to wasted money and energy. These comments were reported via Reuters. Via: https://www.reuters.com/business/delos-data-chip-startup-founded-by-intel-veterans-raises-100-million-ai-networks-2026-09-15/

Glossary candidates

  • Heterogeneous accelerator
  • Interconnect
  • Memory bandwidth
  • Agentic inference
  • Data path

Cold-reader sentence: Delos Data reportedly raised $100 million to build chips and software that move data among AI accelerators, memory and storage.