AnswersPublished17 sources
How Volantis Wants to Use iPhone-Style Lasers to Expand AI Chip Memory
Volantis says laser based VCSEL links could let one GPU connect to as many as 220 memory chips, compared with about eight in the Nvidia example cited in reports. The startup plans to deliver its first integrated inference system in 2027; its proposed optical links aim to ease memory capacity and data transfer constr...
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Volantis says laser based VCSEL links could let one GPU connect to as many as 220 memory chips, compared with about eight in the Nvidia example cited in reports.
- The startup plans to deliver its first integrated inference system in 2027; its proposed optical links aim to ease memory capacity and data transfer constraints in AI hardware.
- Volantis raised $88 million in a Series A co led by Lachy Groom and Abstract Ventures.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How does San Francisco semiconductor startup Volantis plan to use laser-based connections and VCSEL technology already used in Apple devices. Article summary: Volantis plans to replace some short-range electrical links between AI processors and memory with laser-based optical links. It says this could let a GPU access far more memory—and move data fast enough to accelerate AI . Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Volantis wants to use light to move data between AI processors and memory, addressing a constraint that can limit how much memory a GPU can access and how quickly it can use it. The San Francisco startup says its approach could connect up to 220 memory chips to one GPU, but that is a design goal—not a result demonstrated in a shipping system.3
7
Why AI chips run into a memory bottleneck
AI processors need to move large amounts of data between a GPU and memory. Volantis argues that short electrical connections limit how many memory chips can be placed close enough to serve a processor. In the Nvidia comparison reported alongside the company’s announcement, current top-end designs pair about eight high-bandwidth memory chips with a GPU.1
3
That makes the connection between processor and memory an important part of system performance, not just the processor’s computing power. Volantis is betting that changing how those components communicate could make more memory available to AI workloads.1
3
How Volantis’ VCSEL links are supposed to work
Volantis proposes using vertical-cavity surface-emitting lasers, or VCSELs, to transmit data between computing and memory chips. VCSELs are also used in Apple devices for Face ID, according to reporting on the startup’s plan.8
17
Instead of relying only on short-reach electrical connections, the company aims to use optical links to carry data across a larger distance within the system. Volantis says this could let it connect as many as 220 memory chips around a GPU—far above the roughly eight-chip comparison reported for current Nvidia hardware.3
The key distinction is that this is a proposed architecture. The available reports describe what Volantis is developing and what it hopes to achieve; they do not establish that a system has already delivered the claimed memory capacity or performance.3
7
What the company says the change could enable
More accessible memory and faster data movement could help AI inference systems handle demanding workloads. Volantis has pointed to tasks such as AI coding as a potential use case, but the performance benefits remain prospective until the system is tested and delivered.2
7
The company’s first integrated inference-engine deliveries are planned for 2027. That schedule is a target, not confirmation that the system will ship on time or meet its stated goals.7
Volantis’ $88 million funding round
Volantis announced an $88 million Series A co-led by Lachy Groom and Abstract Ventures. John Doerr, VXI Capital, Triatomic and Susa Ventures also participated, as did angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas.5
The funding is intended to support development of the company’s AI inference system. The central question now is whether Volantis can turn its optical-link design—and its 220-memory-chip target—into a working system with demonstrated performance.5
7
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "How Volantis Wants to Use iPhone-Style Lasers to Expand AI Chip Memory"?
Volantis says laser based VCSEL links could let one GPU connect to as many as 220 memory chips, compared with about eight in the Nvidia example cited in reports.
What are the key points to validate first?
Volantis says laser based VCSEL links could let one GPU connect to as many as 220 memory chips, compared with about eight in the Nvidia example cited in reports. The startup plans to deliver its first integrated inference system in 2027; its proposed optical links aim to ease memory capacity and data transfer constraints in AI hardware.
What should I do next in practice?
Volantis raised $88 million in a Series A co led by Lachy Groom and Abstract Ventures.