梳状激光器如何重塑 AI 数据中心光链路

Optical frequency combs generate many wavelengths from a single laser instead of using separate lasers per channel, cutting power, cost, and failure points as AI data centers scale to 16+ wavelengths per fiber.

September 24, 2026

Data center operators building AI clusters need more bandwidth without expanding power and space budgets. One emerging answer is better handling of the light that carries data.

Wavelength-division multiplexing (WDM) increases fiber capacity by sending multiple colors of light down the same strand. The catch is economic and operational: operators need enough stable wavelengths, at reasonable cost and power, with high reliability.

The conventional method – using a separate laser for every wavelength – is straining as channel counts climb. Each added color brings another laser, driver, and control loop, increasing power draw, cost, and potential failure points. Optical frequency comb generators take a different approach, producing many evenly spaced wavelengths from a single source rather than an array of individual lasers.

Comb-based light sources aren’t new. They have circulated in telecom research for years. But the co-packaged optics that many designs envision remain closer to pilot than volume production.

“Potentially, comb lasers allow you to pack many more wavelengths, much tighter together without fear of interference,” said Frank Smyth, founder and CTO of Pilot Photonics, in an interview with Data Center Knowledge.

On a spectrum analyzer, a comb source appears as evenly spaced lines resembling the teeth of the comb. In a laser array, wavelength count and component count are directly linked. Add a wavelength, and you add another laser, driver, and locking loop.

“A comb breaks that link. One pump laser and one silicon nitride die produce all the wavelengths at once, on a grid set by the resonator itself, so the hardware does not grow as you add channels,” Marcello Girardi, CEO and co-founder of Solinide, told Data Center Knowledge.

Arrays handle a handful of channels well, but they hit limits as the channel count rises. “Four wavelengths is a solved problem, and arrays solve it well,” Girardi said. “The difficulty starts at eight, and by 16, the component count limits you.”

Because a comb’s wavelengths come from one pump laser and one passive resonator, spacing is set lithographically rather than tuned individually.

Steven Estrella, director of product management at Quintessent, said combs reduce the number of lasers required for the same channel count compared to conventional continuous-wave distributed feedback (CW DFB) arrays. “This translates to fewer points of failure, thereby increasing reliability, while also reducing wavelength control complexity over a changing environment,” he said. “With the current InP [indium phosphide] CW DFB laser supply shortage, reducing the number of required lasers is more important than ever.”

Data centers used coarse WDM for years – often four widely spaced wavelengths per fiber. Denser, telecom-style plans now pack eight to 32 wavelengths closer together. “Now you have a much more challenging wavelength drift and laser array manufacturing yield issue,” Smyth said. “And this is where comb lasers can offer benefits.”

Where Combs May Land First

Opinions differ on the first deployment beachheads. Quintessent targets link distances between 30 and 100 meters. “Our comb lasers are designed to power DWDM links within the rack and between racks,” Estrella said. “This is another benefit of comb laser technology, since the power and bandwidth problem statement is most acute within the rack.”

Vivek Raghunathan, CEO and co-founder of Xscape Photonics, agreed, pointing to scale-up links between GPUs and switches as uniform enough to standardize a design.

Girardi sees a different starting point: “The first socket is scale-out, between rows, because that is where co-packaged optics is actually being deployed.” He agreed that scale-up is the better long-term fit, but said it awaits the next generation of optical scale-up networking.

Related:Data Centers vs. Telcos: Different Roles, Shared Dependence

Smyth noted limits today in per-wavelength output power for scale-out, especially as per-lane rates climb, though he said his company is working to close that gap.

Raghunathan said total link-level power could drop by 2x to 3x by amortizing one pump laser and control loop across many wavelengths rather than duplicating the chain per channel. Lower link power cuts cooling needs. And when the light source sits in a swappable front-panel module rather than on the package, those watts come out of the budget that GPU silicon competes for. That placement also lets operators swap the light source – the least reliable optical component – without touching the rest of the system.

What’s Real Today and What’s Next

Comb lasers have struggled to gain adoption before. An independent analyst report, Cignal AI’s Optical Component Startup Tracker (July 2026), cites concerns about single-point-of-failure risk, tunability, and stability.

Maturity varies by company. “Various players seem to sit somewhere between lab demo and evaluation kit,” Smyth said.

Solinide says it already ships a product. “We have a product, and we sell it,” Girardi said. “Today, we ship rack-mounted units rather than modules, so the price sits in a different category, and the reason is form factor and volume, not the technology.”

A key enabler still missing across the industry is a standardized external laser source module for co-packaged optics. No qualified version exists yet. Girardi expects prototypes through 2027 and volume production in 2028, when he also expects costs to fall enough to enable a crossover against mature pluggable optics.

Estrella said Quintessent is sampling its comb lasers with customers.

Standards and Lock-In

Standards will determine whether comb deployments become a multi-vendor market or remain tied to a handful of closed systems. ELSFP (External Laser Source Small Form-factor Pluggable) covers the physical light-source module, and the OCI MSA (Optical Compute Interconnect Multi-Source Agreement) spans the broader interconnect. Together, these efforts would allow operators to qualify a socket rather than a specific vendor stack, helping to avoid single-supplier lock-in.

“When platforms move from eight channels to 16 and beyond, that’s the point where WDM has scaled past what incumbent laser arrays can yield, and it marks the transition from combs as one option among several to combs as the architecture,” Raghunathan said.

Smyth summed up what still stands between the technology and broader adoption: “The biggest things standing between here and wider adoption are power per wavelength and high-yield manufacturing.”

Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He consults to industry and media organizations on technology issues.

Want more Data Center Knowledge stories in your Google search results?

Editor's Choice

Aug 21, 2026

Want more Data Center Knowledge stories in your Google search results?

Join the entire data center ecosystem September 21 – 23, 2026, at the Gaylord Texan Resort & Convention Center in Dallas, TX.