Ayar Labs
Fabless chiplet + light-source IP/product vendor: sells the TeraPHY optical I/O chiplet and SuperNova external laser as a co-packaged offering to CPU/GPU/AI-accelerator makers and system builders; manufactures via foundry partners (GlobalFoundries), monetizes through product sales, NRE/engineering engagements, and ecosystem integration (NVLink Fusion).
Both points are priced primary rounds; valuation ~275% higher over ~15 months on AI-infra demand, not disclosed revenue.
Earnings, margins, COGS & capex
Ayar Labs is a venture-funded, pre-scale-revenue company. It does not publish financial statements. The visible financial facts are its capital stack: ~$870M raised cumulatively across Series A-E, most recently a $500M Series E (Mar 2026) at a ~$3.75B valuation, up from ~$370M cumulative and a ~$1B mark at the $155M Series D (Dec 2024). Products (TeraPHY, SuperNova) were slated to reach high-volume production readiness around mid-2026, so meaningful product revenue is nascent. Any revenue figure not sourced to the company should be treated as unknown.
Revenue trend
Margins
n/a
burn-funded by equity
COGS structure
Not disclosed. Structurally: foundry wafer cost (GlobalFoundries GF Fotonix silicon-photonics process), external DFB laser array (SuperNova) sourced/co-developed with laser partners (MACOM, Lumentum, Sivers), and advanced co-packaging/assembly & test. Cost curve and yield on the laser + fiber-attach steps are the key gross-margin swing factors for CPO.
Capex
Fabless model, so no fab capex. Series E proceeds are earmarked to scale high-volume production and test capacity (equipment, packaging/test lines, reliability qualification) and expand global operations (incl. a new Hsinchu, Taiwan office) rather than to build a wafer fab.
Latest earnings
not applicable
Company messaging: TeraPHY + SuperNova targeted for high-volume production readiness ~mid-2026; CEO Mark Wade framed the on-chip optical-I/O market as maturing 2026-2028. Series E explicitly positioned as a path toward a potential IPO in the next ~1-2 years.
- Total funding raised
- ~$870M cumulative
- Last round
- $500M Series E, Mar 2026
- Last valuation
- ~$3.75B
- SuperNova wavelengths
- up to 16 (CW-WDM MSA DFB laser array; powers up to 16 TeraPHY ports)
Growth drivers
- AI compute scale-out hitting the copper-interconnect power/bandwidth/reach wall inside racks and between accelerators
- Shift from pluggable optics to co-packaged optics (CPO) and in-package optical I/O for GPU/accelerator scale-up fabrics
- Design-in with NVIDIA (NVLink Fusion ecosystem), AMD, and system builders (Wiwynn) for rack-scale AI
- Foundry productization with GlobalFoundries plus roadmap work with Intel and a demonstrated TSMC COUPE-based solution (via Alchip)
- Adjacent markets — aerospace/defense (Lockheed Martin) and telecom/radar (Ericsson) exploring optical I/O for weight/bandwidth
Bull & bear
If in-package optical I/O becomes the default interconnect for AI scale-up, Ayar Labs is the best-positioned independent merchant supplier, endorsed by the entire compute ecosystem and funded to reach volume.
- The AI interconnect wall is real: copper cannot deliver the bandwidth-density, reach, and energy-per-bit that next-gen GPU fabrics need, forcing an optical transition Ayar is purpose-built for
- Unique three-way validation - NVIDIA, AMD, and Intel all invested - is a signal no competitor can fully replicate
- NVLink Fusion membership plus GlobalFoundries productization and Wiwynn rack integration show real design-in, not just demos
- $500M Series E from blue-chip crossover investors (Neuberger Berman, ARK, Insight Partners, Sequoia Global Equities, QIA, 1789 Capital) both funds the ramp and pre-seasons the story for an IPO
- As a neutral merchant vendor, Ayar can sell to every accelerator maker, whereas an in-house solution locks to one silicon roadmap
Ayar is a pre-revenue hardware company selling into a market its own strategic investors are building in-house, at a $3.75B mark that prices in a commercial ramp that has not yet happened.
- CPO at volume is unproven: laser reliability, fiber-attach yield, thermal, and field-serviceability remain the reasons hyperscalers have been slow to adopt
- The strategic investors (NVIDIA, Intel, Marvell-adjacent players) are the same ones most able to internalize optics and cut a merchant vendor out
- Marvell now owns Celestial AI and Lightmatter is heavily funded - Ayar is not the only credible independent, and both have public/large backing
- Valuation up ~275% (from ~$1B Series D in Dec 2024 to ~$3.75B Series E in Mar 2026, ~15 months) on AI-infra enthusiasm, not on disclosed revenue - multiple compression risk if the ramp slips
- No disclosed revenue, margins, or backlog means the price rests on a narrative that requires flawless execution through 2026-2028
What it is worth
Last-priced private round (Series E, Mar 2026) cross-checked against comparable silicon-photonics M&A (Marvell/Celestial AI ~$3.25B upfront).
Down-round or muted exit if the CPO ramp slips, strategic customers internalize optics, or AI-infra sentiment cools - a pre-revenue mark run up ~275% in ~15 months has meaningful compression risk.
~$3.75B round mark holds into an IPO window over ~1-2 years, with re-rating gated on demonstrated volume revenue and margins.
IPO or acquisition well above $3.75B if CPO adoption inflects in 2026-2028 and Ayar captures merchant share of the AI optical-interconnect TAM; strategic acquirers (a hyperscaler or large semi) could pay a control premium.
~$3.75B post-money on ~$500M Series E, up from ~$1B at the Dec-2024 Series D. No public price and no disclosed revenue, so the mark is a venture/strategic-value figure, not an earnings- or cash-flow-based valuation. The Celestial AI deal (~$3.25B upfront, up to ~$5.5B with earnout, for a pre-scale peer) is the nearest anchor and broadly supports the order of magnitude.
SWOT
Strengths
- Strategic investors span the whole compute stack — NVIDIA, AMD Ventures, and Intel Capital all on the cap table plus foundry partner GlobalFoundries
- First-mover with a proven, demonstrated in-package optical I/O chiplet (TeraPHY) + external light source (SuperNova) architecture
- Deep, well-capitalized war chest (~$870M raised) to fund a long hardware productization cycle
- Ecosystem design-in momentum: NVIDIA NVLink Fusion membership and Wiwynn rack-scale collaboration
Weaknesses
- Pre-volume revenue; commercial ramp and unit economics of CPO still unproven at scale
- External-laser + fiber-attach reliability, yield, and serviceability are hard, unsolved-at-volume CPO problems
- Dependent on foundry (GlobalFoundries) and laser suppliers for the critical path
- Customer concentration risk — the same giants that invest (NVIDIA, AMD, Intel) are also building or could build competing in-house optics
Opportunities
- Structural transition of AI scale-up interconnect from copper to optics over 2026-2030
- Becoming the merchant CPO standard if hyperscalers prefer a neutral third party over vertically-integrated silicon
- Expansion into defense/aerospace (Lockheed Martin) and telecom (Ericsson) optical-I/O use cases
- A credible IPO window if AI-infra capital markets stay open and the production ramp lands
Threats
- In-house verticalization — NVIDIA (silicon photonics / Quantum-X, Spectrum-X photonics), Broadcom, and Marvell (via Celestial AI) building CPO internally
- Marvell's ~$3.25B upfront acquisition of Celestial AI (completed Feb 2026; up to ~$5.5B with revenue earnout) hands a deep-pocketed public rival a competing Photonic Fabric
- Well-funded private rival Lightmatter (Passage platform) targeting in-package optical interconnect
- Timing risk: if copper (and near-packaged optics) stretches another generation, CPO adoption slips and burn continues
Moats, dependencies & bottlenecks
Moats
AMD, Intel Capital, GlobalFoundries) Powerful for design-in and credibility, but the same backers are potential competitors, so it is a double-edged moat.
Technology / IP lead in in-package optical I/O (TeraPHY chiplet + SuperNova external laser) First-mover demonstrated product, but Lightmatter, Marvell/Celestial, and Broadcom are converging fast; lead measured in a year or two, not a decade.
Multi-year GlobalFoundries (GF Fotonix) silicon-photonics process co-development and packaging/test qualification are hard to replicate quickly.
~$870M raised buys runway, but capital alone is not defensible when rivals are $50B+ public companies.
Dependencies
Foundry / manufacturing Primary manufacturer of Ayar's photonics chiplets on the GF Fotonix platform; single-foundry concentration until Intel/TSMC paths mature.
MACOM (MTSI), Lumentum (LITE), Sivers Semiconductors External DFB laser (SuperNova) is the reliability-critical, yield-sensitive part of CPO.
Customer / platform Revenue depends on being designed into GPU scale-up fabrics (e.g. NVLink Fusion); these customers can also build optics in-house.
Pre-revenue burn requires continued access to private capital or an IPO window; a funding-climate reversal pressures the ramp.
Go-to-market / operations Rack-scale deployment depends on ODM partners and scaling of assembly/test lines funded by Series E.
Advantages
- Only independent CPO vendor backed simultaneously by NVIDIA, AMD, and Intel
- Modular chiplet + external-laser architecture that decouples the light source from the compute die (serviceability/reliability argument)
- Merchant, silicon-agnostic model can serve every accelerator maker
- Well-capitalized to survive the long hardware productization cycle
- Early ecosystem design-in (NVLink Fusion, GlobalFoundries, Wiwynn)
Weaknesses
- No disclosed revenue, margins, or backlog to underwrite the $3.75B mark
- Unproven CPO unit economics and volume yield
- Strategic-investor customers are also would-be competitors
- Dependence on third-party foundry and laser suppliers for the critical path
- Richly valued on AI narrative; vulnerable to sentiment/multiple compression
Bottlenecks
- Volume manufacturing yield and reliability of the co-packaged laser + fiber-attach at data-center scale
- Getting designed into locked GPU/accelerator roadmaps against in-house optics teams
- Advanced-packaging and test capacity ramp (the stated use of Series E proceeds)
- Converting eval/NRE engagements into repeatable, high-margin volume product revenue
- Standardization uncertainty - CPO interfaces and sockets are not yet settled industry standards
Top signals & trends
Top signals
Strong investor conviction and IPO-prep, but valuation ran ahead of disclosed fundamentals.
Concrete platform design-in with the dominant AI-compute vendor.
Validates the space but arms a large public competitor with a rival Photonic Fabric.
Typical late-stage roster that pre-positions for an IPO.
Commercial proof is still ahead, not behind; execution risk remains.
Trends
The core secular tailwind and reason the round happened.
Largest structural threat to a merchant vendor.
Validates value but concentrates competition in deep-pocketed acquirers.
Optics' energy-efficiency advantage strengthens the adoption case.
Funds the ramp today; a reversal would pressure pre-revenue names hardest.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Primary foundry manufacturing Ayar's silicon-photonics chiplets on the GF Fotonix platform.
Laser / optical component partner for the SuperNova light source.
DFB laser supplier partner; also a broader optics competitor.
Sweden-listed (Nasdaq Stockholm) DFB-laser-array partner for SuperNova.
Investor (Intel Capital) and process-integration partner; demonstrated a 2Tbps photonic chip-to-chip link with Ayar (2024).
Demonstrated a TSMC COUPE-based CPO solution with Ayar via Alchip; a future foundry/process path, not yet confirmed as volume production.
Investor and design-in target via NVLink Fusion for rack-scale AI.
Investor (AMD Ventures) and prospective accelerator customer.
Taiwan-listed ODM collaborating to bring CPO to rack-scale AI systems (Mar 2026).
Investor (Lockheed Martin Ventures) exploring optical I/O for radar/aerospace weight and data-rate gains.
Investigating optical I/O for telecom/radar systems.
Private, heavily funded rival; Passage platform (M1000 / L200) targets in-package optical interconnect across the full die - a more integrated approach vs Ayar's modular chiplet + external laser.
Photonic Fabric interconnect (chiplet/switch/packaging); acquired by Marvell for ~$3.25B upfront (up to ~$5.5B with earnout), completed Feb 2026, giving a large public player a direct CPO offering.
CPO switch/optics programs (e.g. Bailly) and dominant merchant networking silicon; can bundle optics with switching at scale.
Investor and prospective customer, but also developing its own silicon-photonics/CPO (Quantum-X, Spectrum-X photonics) - the classic build-vs-buy threat.
Vertically integrated optics/laser leader; strong in datacom transceivers and a natural CPO/optical-engine competitor and supplier.
Private; low-power linear/near-package optics for AI interconnect - overlapping problem space.
Private photonic-compute/interconnect startup pursuing an adjacent optical-networking approach.