Tuesday, November 17, 2026
AI 인프라 · 뉴스 & 분석
홈 › 자본시장 › 리포트
자본시장 · 리포트

Volantis가 20 trillion 파라미터를 초과하는 모델용 포토닉 추론 아키텍처 개발을 위해 Series A 펀딩으로 $88 million을 조성했다.

벤처 자금의 포토닉 추론이 조 단위 파라미터 규모에서 토큰당 전력을 감소시키며, 대체 추론 엔진이 GPU 중심 컴퓨팅 공급을 다양화한다.
업계 전문지Slicast · 2026년 10월 1일 19:31 UTC · 미국 · 출처: Pulse 2.0
중요도 70

Volantis has raised $88 million in Series A funding to develop and commercialize a photonic architecture designed to address memory capacity and bandwidth constraints in AI inference systems. The round was co-led by Lachy Groom and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic, and Susa Ventures. Angel investors Dwarkesh Patel, Naveen Rao, and Sholto Douglas also participated.

The founding team comprises semiconductor and photonics veterans from NVIDIA, AMD, Broadcom, and Ayar Labs, whose prior work includes the first CoWoS product, high-volume tunable VCSELs, and early silicon-photonics co-packaged optical systems.

Volantis is addressing what it calls the AI memory wall: running increasingly large models requires both sufficient memory capacity to hold model parameters and adequate bandwidth to continuously feed those parameters into compute engines. On-chip SRAM provides high bandwidth but limited capacity, while HBM-based GPU systems offer greater capacity but become constrained by bandwidth as model size increases. Volantis has designed A-1, its first system, to increase memory capacity and bandwidth by nearly two orders of magnitude simultaneously.

A-1 is being designed to run AI models exceeding 20 trillion parameters at speeds of up to 10,000 tokens per second per user while reducing inference cost per token. The system employs a photonic interconnect specifically for connections between compute chips and memory. Its optical fabric connects large numbers of memory chips into a unified pool and aggregates bandwidth as additional memory is added. Instead of external lasers, the architecture uses custom micro-VCSELs based on the existing gallium arsenide VCSEL supply chain, enabling end-to-end links that consume less than one picojoule per bit.

Volantis plans to deliver its first integrated inference engines to customers in 2027. Series A proceeds will support development and commercialization of A-1 and its photonic memory architecture, expansion of the engineering organization, and preparations for customer deployments.

원문 보기
Volantis가 20 trillion 파라미터를 초과하는 모델용 포토닉 추론… · Slicast