Volantis Raises $88M for Micro-VCSEL Chip-to-Memory Optical Interconnect

FIBER & OPTICAL COMMUNICATIONS · AI INTERCONNECT · OCTOBER 2026

San Francisco startup Volantis Semiconductor has raised $88 million in a Series A round for an optical interconnect architecture built on micro-VCSELs and integrated waveguides, aimed not at chip-to-chip links but at the harder problem of connecting compute chips to memory. The company claims its first “A-1” system lifts memory capacity and bandwidth by nearly two orders of magnitude at the same time.

optics.org reported the funding on 6 October 2026. Volantis was founded in 2022 by chief executive Tapa Ghoush, and its thesis is narrow and specific: modern AI inference is throttled by a trade-off between how much memory a system can hold and how fast that memory can be fed to compute. On-chip static RAM delivers bandwidth but little capacity; GPUs deliver capacity but less bandwidth. The result is either high cost and energy consumption, or slow compute.

Why chip-to-memory is the harder interconnect

Photonics already moves data inside data centres, but the established use case is chip-to-chip connectivity over metres of fibre. Volantis argues that chip-to-memory links are a different engineering problem, because they carry more than 100 times as much data over much shorter distances. That changes the energy and cost budget, and it rules out simply reusing transceiver-style modules with external lasers.

Instead, the company’s “A-1” system describes a new category of photonic interconnect: optical wires built from tiny micro-VCSELs and integrated waveguides, connecting large numbers of memory chips into a single unified pool so that bandwidth aggregates as memory is added. Volantis states that end-to-end links consume less than one picojoule per bit, and that the micro-VCSELs are small, temperature-stable and low power.

The design also sidesteps a supply constraint. External sources for optical links are typically indium phosphide (InP) edge-emitting lasers, where supply remains tight. Micro-VCSELs are fabricated on gallium arsenide (GaAs) epiwafers, a mature compound-semiconductor supply chain that has already shipped in very high volumes for short-reach optical and sensing applications.

Who is backing it

The round was co-led by Lachy Groom, co-founder of Physical Intelligence, and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures alongside a group of angel investors. Volantis says it plans to deliver its first integrated inference engines to customers next year.

The funding lands in the same fortnight as a second VCSEL story. PicoJool, the Palo Alto VCSEL developer, closed an additional $27.5 million Series A on 29 September 2026 to scale 200 Gb/s devices for AI data-centre connectivity, with a production ramp expected in early 2027. PicoJool states that its 200G products exceed 37 GHz of bandwidth, and that its portfolio includes quad 100G, quad 200G and 32x50G NRZ microVCSEL configurations intended to give customers multiple routes to 1.6T connectivity. It manufactures through the compound-semiconductor foundry WIN Semiconductor and is backed by Playground Global, the venture firm founded by former Intel chief executive Pat Gelsinger.

Read together, the two rounds point at the same conclusion: the optical layer is being pushed closer to the silicon in AI systems, and the VCSEL — long associated with short-reach and sensing applications rather than long-haul telecom — is being repositioned as a data-centre interconnect source.

Optical Transceivers & Modules Sourcing →

What this means for optical sourcing

• Ask which laser platform a module really uses. A specification that quotes lane rate and reach but not the emitter material is incomplete in 2026. GaAs VCSEL and InP edge-emitter routes carry different lead times, temperature behaviour and second-source options.
• Separate chip-to-memory from chip-to-chip in your RFQ. The two links have different loss budgets, different connector expectations and different cost targets. Quoting one against the other produces the wrong number.
• Test at operating temperature, not on the bench. The selling point of micro-VCSELs is temperature stability. That claim is only worth anything if the incoming inspection plan reproduces it, so specify the temperature range and the measurement method.
• Wafer-level test capability identifies a real supplier. Array-based emitters are sorted before assembly. Suppliers who cannot show wafer-level probing and binning data are assembling rather than manufacturing.
• Expect the coupling optics to be the schedule risk. Whether the link uses waveguides, fibre arrays or free-space coupling, the alignment and assembly step is where yield is won or lost — and where the qualification evidence should be demanded first.

Sources and disclosure

Sources: optics.org, “Volantis raises $88M with AI memory breakthrough claim”, 6 October 2026 — Series A of $88 million, company founded in 2022 by CEO Tapa Ghoush, architecture based on micro-VCSELs and integrated waveguides connecting compute chips to memory, the “A-1” system targeting a simultaneous increase in memory capacity and bandwidth of nearly two orders of magnitude, chip-to-memory links carrying more than 100 times as much data over much shorter distances than chip-to-chip links, end-to-end links consuming less than one picojoule per bit, use of the gallium arsenide VCSEL supply chain against constrained indium phosphide laser supply, funding co-led by Lachy Groom and Abstract Ventures with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures, and first integrated inference engines planned for delivery next year. optics.org, “PicoJool to scale VCSEL production with $27.5M round”, 29 September 2026 — additional $27.5 million Series A, 200 Gb/s devices for AI data-centre connectivity, production ramp expected in early 2027, more than 37 GHz bandwidth, quad 100G, quad 200G and 32x50G NRZ microVCSEL configurations, the use of GaAs epiwafers, manufacturing partnership with WIN Semiconductor, and backing from Playground Global. Procurement and qualification observations are TUOLAN TRADE analysis and are not attributed to any cited source. Edited and analysed by TUOLAN TRADE.