ayar labs optical interconnect technology moving ai data through light

Ayar Labs Bets Big On Light To Break AI Data Jams

Training a frontier AI model means getting thousands of GPUs to behave like one computer, and the wiring between those chips is increasingly the weak link. Ayar Labs, a San Jose startup, wants to replace the copper in that wiring with light. In March 2026, it raised $500 million at a $3.75 billion valuation to make that happen at volume.

This guide explains the problem in plain terms, what Ayar’s two products do, who is building with the company, and what is still unproven.

The Problem Ayar Labs Is Trying To Solve

Inside an AI data center, chips talk to each other constantly. Copper traces and pluggable optical modules carry that traffic, but both lose ground as speeds and distances grow. They use more power, add delay, and take up space.

Ayar’s answer is co-packaged optics, where the optical engine sits right next to the processor inside the same package. A joint announcement with chip designer Alchip describes the benefit as shorter electrical paths and avoiding the power and latency penalties of pluggable optics. The underlying field is silicon photonics, which builds optical components on silicon using chip manufacturing methods.

How Co-Packaged Optics Differs From Pluggable Modules

Pluggable optical modules sit at the edge of a server or switch, so electrical signals travel across a board before they turn into light. Co-packaged optics moves that conversion into the chip package, which shortens the electrical path and trims power and delay. The idea is not new, but real products arrived recently. Broadcom added a third-generation co-packaged design to its 102.4 terabit Tomahawk 6 switch chip, and Nvidia announced two families of switches that use the technology. Nvidia’s Spectrum-X Ethernet Photonics platform was reported to reach full production in the second half of 2026.

Those are switch products. Ayar’s focus is scale-up: the links that tie GPUs together inside and between racks. That is a different and harder target, because the connection has to match the speed and latency GPUs expect from each other.

Where TeraPHY Is Made And Who Designs Around It

Ayar’s third-generation TeraPHY delivers up to 8 terabits per second and is built on GlobalFoundries’ 45-nanometer silicon photonics process, a platform Ayar helped shape, according to coverage of its OFC 2025 and Hot Chips 2025 presentations. The remote SuperNova light source uses laser cells from Sivers, adapted into an array that splits and multiplexes wavelengths. Ayar has also partnered with Taiwanese ASIC designers Alchip and Global Unichip, both closely tied to TSMC, so customers can design TeraPHY into their own chips. Using established foundry and packaging processes is a quiet strength, because volume depends on factories that already exist.

TeraPHY And SuperNova Explained

Ayar’s interconnect has two parts that work as a pair.

ComponentRoleKey detail
TeraPHYOptical I/O chiplet that encodes data onto lightUp to 8 terabits per second per unit, per SiliconANGLE
SuperNovaExternal multi-wavelength light sourceOne shared laser feeds several chips

TeraPHY is manufactured by GlobalFoundries, according to Optics and Photonics News. In 2025, Ayar also introduced a version that supports UCIe, an industry standard for linking chiplets, which lets engineers integrate it directly with processors such as GPUs, as SiliconANGLE reported. The Alchip announcement calls TeraPHY protocol-agnostic, so it can work with customer-designed chiplets.

SuperNova solves a different problem. Instead of building a laser into every chip, one external source feeds many. The design logic is to keep a heat source away from the compute silicon and make the laser easier to service on its own.

Who Is Building With Ayar Labs

Ayar sells into an ecosystem rather than straight to end users, so its partner list is the best evidence of traction.

PartnerWhat they do togetherDate
AlchipTeraPHY co-packaged with Alchip ASIC solutions on a common substrateSeptember 2025
WiwynnRack-scale AI systems built around TeraPHY and SuperNovaMarch 2026
GlobalFoundriesManufactures the TeraPHY chipletOngoing
NVIDIA NVLink FusionEcosystem membership, per one industry analysisJune 2026

Ayar also ranked in the top five of the Computing category in Fast Company’s 2026 list of the world’s most innovative companies.

What Partnerships Do Not Prove Yet

Design wins are not shipments. An analysis by Parallax argues that the silicon works, but the open question is manufacturing, with a production timeline that runs through 2027. That is the right way to read the Series E. The money is meant to scale production and test capacity, and the next milestones will be volume deliveries rather than more announcements.

Funding And Valuation In Plain Numbers

DateEvent
April 2022$130 million Series C backed by NVIDIA and HPE’s Pathfinder fund
December 2024$155 million Series D, valuation above $1 billion
March 2026$500 million Series E led by Neuberger Berman at $3.75 billion, bringing total outside funding to $870 million
September 2026Additional $150 million reported, bringing 2026 primary capital to $650 million.

NVIDIA, AMD, and MediaTek took part in the Series E, and other backers include ARK Invest, Insight Partners, Qatar Investment Authority, and Sequoia. Having both NVIDIA and AMD as investors is unusual because they compete in GPUs, and it suggests the whole industry sees the interconnect as a shared problem. Trackers disagree on lifetime totals depending on how earlier rounds are counted, so quote the company’s $870 million figure after the Series E and treat larger numbers as including later money. In September 2026, Ayar extended the Series E by another $150 million, bringing the primary capital it raised in 2026 to $650 million.

ayar labs teraphy silicon photonics chip using light for data transfer

Founders And Leadership

Six founders started Ayar in 2015, drawing on photonics research at MIT and UC Berkeley. They are Mark Wade, Chen Sun, Vladimir Stojanovic, Alex Wright-Gladstein, Milos Popovic, and Rajeev Ram. Wade became CEO in December 2023 after serving as chief scientist and CTO, following Charles Wuischpard, who led the company from 2018 to 2023. Stojanovic is the CTO. In February 2026, the company hired Sankara Venkateswaran as VP of Engineering to lead the TeraPHY organization.

What The 2026 Funding Is Actually For

Ayar raised $500 million in its Series E in March 2026 and extended it with another $150 million announced on September 10, bringing primary capital raised in 2026 to $650 million. CEO Mark Wade said copper interconnect is becoming the limiting factor for AI scale-up, and that the money lets Ayar move faster on manufacturing-ready co-packaged optics. Server maker Wiwynn made a strategic investment earlier in 2026 and works with Ayar on rack-level systems built around TeraPHY and SuperNova.

The company also opened a Bengaluru Design Center to support the move from technology development to commercialization and production. Notice what the money is not for. It is not for proving the idea. It is for factories, testing, and engineers, which is the stage where hardware companies often face their hardest tests.

The Competition And The Celestial AI Deal

Optical interconnect is crowded. Lightmatter and Ranovus are among the other names, and the sector has already seen consolidation. Marvell completed its acquisition of Celestial AI in early 2026, folding that company’s Photonic Fabric technology into its data center group. Our Celestial AI profile covers the deal in detail.

For Ayar, that result cuts two ways. It validates the category, and it shows that big chipmakers will buy their way into optics. Ayar remains independent, which is why its choice between partnering and being acquired matters.

What Could Slow Ayar Down

Four risks stand out. The first is timing, since volume shipments run through 2027 and customers choose designs years ahead. The second is big competitors, since Nvidia and Broadcom are building their own co-packaged optics inside switches and could shape the standards. The third is manufacturing, because combining optics, lasers, and chips in one package is hard to do at high yield and under heavy heat. The fourth is customer concentration, since a few hyperscalers and chip makers decide most of the market

New entrants add pressure. Volantis announced an $88 million Series A on October 1, 2026 for a VCSEL-based optical interconnect. The category is attracting money quickly, which validates the problem but shrinks any single company’s lead. Our AI bubble guide covers the capital spending behind this race, and our bottleneck calculator guide explains the same weakest link idea at PC scale.

Why This Matters For AI Power Bills

Moving data efficiently is one lever on data center energy use, which connects Ayar to the resource questions raised in our piece on why AI uses so much water. Interconnect is only one part of total demand, but it is a part where a better physical layer can help.

Ayar Labs Stock And IPO Outlook

Ayar Labs is private, has no ticker, and has not filed for an IPO. Its size, institutional investors, and production ramp are typical of a company that could list eventually, but nothing is scheduled. Pre-IPO marketplaces exist for accredited investors, and they carry real risks in pricing and availability.

Careers And Locations

Ayar is headquartered in San Jose, with Santa Clara, Hsinchu in Taiwan, and a newer Bengaluru design center. Glassdoor reviews rate career development around 4.7 out of 5 and compensation around 3.9, with some reviewers noting little equity such as RSUs for a private company.

Questions People Ask About Ayar Labs

What does Ayar Labs make?

Optical interconnect parts, mainly the TeraPHY chiplet and the SuperNova light source.

How much is Ayar Labs worth?

$3.75 billion as of its March 2026 Series E.

Who is the CEO of Ayar Labs?

Mark Wade, who took over in December 2023.

Can I buy Ayar Labs stock?

Not through a normal brokerage account. It is private.

How much does Lyzr AI cost?

Lyzr has a free Community plan, a Pro plan at $99 a month or $79 a month billed annually, and custom Enterprise pricing. Confirm current prices on Lyzr’s pricing page.

Who are Lyzr AI’s customers?

Reported customers include Willis Towers Watson, Verifone, and In-Q-Tel, with roughly 32 to 50 customers in total by reported counts.

Where is Lyzr AI based?

Lyzr is headquartered in Jersey City, New Jersey, with most of its engineering in Bangalore.

Is Celestial AI the same company?

No. Celestial AI was acquired by Marvell, while Ayar remains independent.

What To Watch In 2027

Look for volume shipments of TeraPHY, named customers moving from design wins to production, and any sign that Ayar prefers an IPO or a strategic partner. Those moves will show whether co-packaged optics becomes a standard part of AI systems. Watch Nvidia and Broadcom as well, because their own co-packaged switches set the standard Ayar has to fit.

Related Posts