A funding story came back around this week. On November 4, Data Centre Dynamics carried a report that silicon photonics startup Xscape Photonics has closed a $44 million Series A funding round, led by IAG Capital Partners with participation from Cisco Investments and Nvidia, bringing the company’s total raised to $57 million. It is also, as a piece of news, two years old.
The round it describes closed in October 2024. The company’s own announcement dates to mid-October of that year, and Columbia Engineering — the university out of whose labs Xscape was founded — covered it on October 31, 2024. What Data Centre Dynamics republished this week is the 2024 announcement, not a new raise. Readers who took the November dateline at face value would be forgiven for thinking Xscape had just banked $44 million. It had not.
That distinction is the reason to write about the round at all. A $44 million Series A from 2024 is, on its own, stale news. But the two years since have produced a legible record of what the money bought: an evaluation kit, a foundry-built multi-colour laser, a second funding extension, a first pluggable product, and a show-floor demonstration of a comb laser generating up to 100 colours from a single silicon chip. The funding is old. The evidence about whether the thesis works is new — real hardware, real foundry partners, and still no public customer, no published benchmark, and no performance figure that would let an outsider test the company’s boldest claims.
What was actually announced
Precision first, because the republished story invites exactly the confusion it created. In October 2024, Xscape Photonics announced the close of a $44 million Series A, led by IAG Capital Partners, with investment from Altair, Cisco Investments, Fathom Fund, Kyra Ventures, LifeX Ventures, NVIDIA, and OUP. The round brought the company’s total raised to $57 million. The stated purpose was to accelerate development of ChromX, the company’s programmable, multi-colour photonics platform for AI data-centre fabrics.
Xscape was founded in 2022 in Santa Clara, California, out of research that began more than a decade earlier in Columbia Engineering’s photonics labs. The founding team spans the Columbia groups behind that research: Vivek Raghunathan, the chief executive; Alexander Gaeta, the president; Yoshi Okawachi, vice-president of research and development; and Keren Bergman and Michal Lipson, both Columbia professors and members of the company’s board of advisors.
The company’s pitch, then and now, is built around a phrase it coined for the problem: “escape bandwidth.” A GPU’s compute is only useful if data can get out of it — to neighbouring accelerators, to memory, to the rest of the cluster — and the links out of the package have become the binding constraint on AI system performance. Xscape’s claim is that its platform maximises that escape bandwidth by a factor of ten, cuts power consumption by a factor of ten against incumbent solutions, and improves inference performance for AI workloads. Those are the company’s figures, stated in its own announcements and repeated in the coverage. No published benchmark supports them, and no independent measurement has been offered. Hold that thought; it returns below.
The technical core is colour count. A conventional optical interconnect, as Columbia’s account puts it, is restricted to transmitting data over four colours — four wavelengths — on a single fibre. ChromX is designed to put hundreds of colours on that same fibre, each an additional lane of data without another laser, modulator, or fibre to lay. The mechanism is a comb laser: a single silicon photonics device generating many evenly spaced wavelengths at once, replacing a rack of discrete lasers with one programmable source. If it works at scale, bandwidth multiplies by colour count rather than by fibre count, and the power and cost of the optical layer stop scaling with the number of discrete lasers.
What the money bought
In June 2025, Xscape launched EagleX, a laser evaluation kit for the ChromX platform — the first artefact customers could actually put on a bench. In August 2025, the company and Tower Semiconductor announced the successful demonstration of functional 16-colour CombX prototypes, and jointly unveiled what they described as the industry’s first on-chip, optically pumped, multi-wavelength laser source, built on Tower’s PH18 silicon photonics platform and supporting both CWDM and DWDM wavelength grids. The device is monolithically integrated — the programmable multi-colour laser sits directly on the silicon chip, optically pumped by a single continuous-wave external laser — and it eliminates the need for hybrid III-V laser integration, the expensive and yield-sensitive step of bonding a different semiconductor material onto silicon. Validation kits were made available.
That Tower relationship matters more than the press release lets on. The August 2025 device runs on PH18, Tower’s mature, high-volume silicon photonics process — the same platform this publication covered in November when Tower and Innolight moved a near-packaged optics device into early production on it. For a startup whose core claim is manufacturability at scale, a qualified merchant foundry process rather than a bespoke flow is the difference between a laboratory result and a product line.
In March 2026, the funding story advanced on its own. Xscape announced an additional $37 million, led by Addition, a new investor, with continued support from existing investors including IAG Capital Partners and Nvidia. That extension brought total Series A funding to $81 million and total funding since founding to nearly $95 million, and was reported to have doubled the company’s valuation. The March announcement also carried the company’s first pluggable product: FalconX, described as the industry’s first fully redundant External Laser Small Form-factor Pluggable (ELSFP) device, emitting up to eight wavelengths of light and generating more than 1 watt of optical power from a single pluggable module. FalconX is designed to comply with industry MSA standards for scale-up and scale-out fabric links.
And at the OCP Global Summit in October 2026, Raghunathan demonstrated the platform on the show floor: a multi-colour laser generating up to 100 colours from a single piece of silicon, a fully automated 64-wavelength comb device aimed at memory clusters, EagleX carrying 16 wavelengths on a single fibre, and FalconX operating in a live OCI MSA-compliant link at 50G NRZ. A trade-show floor is not a data centre, and a demonstration is not a product shipment. But it is the furthest point on a visible arc — from a 2024 funding announcement, to a 2025 evaluation kit and foundry prototype, to a 2026 pluggable product running in a standards-compliant link.
The 10x claims, and what would test them
The numbers that travel with every Xscape story — ten times the escape bandwidth, ten times less power, better inference performance — deserve the same treatment this desk gives every vendor multiple: they are company claims, not measurements. No paper, datasheet, or customer disclosure in the public record attaches a measured power-per-bit, bandwidth-per-fibre, or inference throughput to ChromX or FalconX. The March 2026 announcement says FalconX generates more than 1 watt of optical power and eight colours; it does not say what the module consumes electrically, what bit error rate it achieves, or at what reach. The October OCP demonstration ran a link at 50G NRZ and published no power figure alongside it.
There is a second, quieter claim embedded in the March announcement that is arguably more important than the 10x multiples, and more testable. Xscape said AI clusters have grown more than tenfold in two years, that a single failed laser module can stall a workload, and that hyperscalers demand ten times fewer failures than the incumbent laser level. FalconX’s answer is built-in redundancy — the “fully redundant” in its description. That is a claim about failure rates, not speed, and it is the claim a hyperscaler would actually qualify first. But redundancy and reliability are proven by qualification data and field hours, neither of which Xscape has disclosed. Until a customer says it has qualified FalconX, the reliability claim sits in the same category as the power claim: plausible, unmeasured in public, and central to the investment case.
Why the investors matter more than the amount
The $44 million figure is, by the standards of AI infrastructure in 2026, modest. What is notable about the round is who wrote the cheques, and who kept writing them. Cisco Investments and Nvidia participated in the 2024 Series A; Nvidia returned for the March 2026 extension, and Addition — Lee Fixel’s firm — led it. Cisco buys optical technology at hyperscale volumes, and Nvidia makes the GPUs whose escape-bandwidth problem Xscape claims to solve. A strategic investor’s participation is not a purchase order and not a design win; it is a signal that the investor wants a front-row seat to the technology.
The market context is large and, for once, quantified by a named source rather than a vendor’s own estimate. In the Tower announcement, LightCounting estimated that sales of optical transceivers and linear pluggable optics and co-packaged optics for AI clusters would reach more than $10 billion in 2026 — double the size of sales in 2024 — and projected $20 billion by 2030. Those are LightCounting’s estimates, not Xscape’s, and they frame the commercial logic: even a small share of a $10 billion market doubling in two years justifies a $44 million bet on a different laser architecture.
What is undisclosed, and what to watch
An honest accounting of what is not on the record is part of the story, not a footnote to it. No customer has been named. No volume, ramp schedule, or revenue figure has been disclosed. No production date for FalconX has been given; sampling is by enquiry. No measured power consumption, bit error rate, reach, or lane rate for a shipped device has been published. The 16-colour prototypes are prototypes, and the 100-colour figure from the OCP demonstration is a show-floor capability, not a product specification.
Each omission has a plausible innocent explanation — hyperscaler customers do not let suppliers name them, and qualification timelines are commercially sensitive. But the pattern means the claims that matter most to a buyer — does it work at yield, at what cost, in whose system, for how long without failing — are entirely untested in public.
The signals that will move this story from funding recap to evidence are specific. First, a customer: a hyperscaler or switch vendor naming FalconX in a system, or Xscape naming a customer itself. Second, a qualification marker: any disclosure that FalconX has entered or completed hyperscaler qualification, the gate between sampling and revenue. Third, a measured disclosure: power per bit, bit error rate, or reach from a customer system rather than a company statement.
There is also a narrower lesson in the DCD republication itself: a 2024 announcement, carried again in November 2026 with a fresh dateline and no indication that the event predates the current funding cycle by two years. The $44 million was real, the investors were real, and the platform it funded is now demonstrably further along than it was. But the round is not news, and the news is what has happened since. The next chapter is not a funding announcement. It is a customer.
Sources and further reading
Data Centre Dynamics, “Silicon photonics startup Xscape Photonics closes $44m funding round,” carried November 4, 2026 — the $44 million Series A led by IAG Capital Partners, with Cisco Investments and Nvidia participating; total raised $57 million; ChromX described as maximising “escape bandwidth” by 10x and cutting power by 10x. The round closed in October 2024; the November 2026 date is the republication, not the event. Columbia Engineering, “Xscape Photonics Secures $44 Million to Transform AI Data Centers,” October 31, 2024 — founding out of Columbia research; co-founders Keren Bergman, Michal Lipson, and Alexander Gaeta, with Vivek Raghunathan and Yoshi Okawachi; ChromX handling hundreds of colours against four for typical interconnects; full investor list including Altair, Cisco Investments, Fathom Fund, Kyra Ventures, LifeX Ventures, NVIDIA, and OUP. Xscape Photonics, “Xscape Photonics Announces $37 Million in New Funding, Launches Eight-Wavelength Laser for AI Data Center Networks,” March 2026, as carried by StorageNewsletter — $37 million additional funding led by Addition, total Series A $81 million, total funding nearly $95 million; FalconX, the first fully redundant ELSFP device emitting up to eight wavelengths, over 1W optical power; ChromX roadmap targeting 16, 32, and ultimately 128 colours; 16-colour CombX prototypes demonstrated with Tower Semiconductor in August 2025; EagleX Laser Evaluation Kit launched June 2025. Tower Semiconductor and Xscape Photonics, “Xscape Photonics and Tower Semiconductor Unveil the Industry’s First Optically Pumped On-Chip Multi-Wavelength Laser Platform for AI Datacenter Fabrics,” August 25, 2025 (GlobeNewswire) — on-chip, optically pumped, multi-wavelength laser source on Tower’s PH18 platform, supporting CWDM and DWDM grids; LightCounting estimates of optical transceiver and LPO/CPO sales for AI clusters exceeding $10 billion in 2026, double 2024, and reaching $20 billion by 2030. OCP Global Summit demonstration, October 2026 — multi-colour laser generating up to 100 colours from a single silicon chip; FalconX operating in a live OCI MSA-compliant link at 50G NRZ. All performance multiples attributed to Xscape are company claims with no published benchmarks; no customer, volume, or measured power figure for ChromX or FalconX is asserted here, because none is publicly disclosed. This article is analysis, not investment advice.