Pat Gelsinger-backed PicoJool raises $27.5 million to scale VCSELs for AI links
The Series A, led by Socratic Partners, brings the Palo Alto startup's total funding to $39.5 million as it works toward qualifying its optical modules with customers.
PicoJool, a Palo Alto startup building laser-based optical links for AI data centers, has raised $27.5 million in a Series A led by Socratic Partners, the venture firm run by former NXP Semiconductor chief executive Rick Clemmer; Hudson River Trading also backed the round, according to Data Center Dynamics.
Add the earlier seed round, led by Playground Global, the firm of former Intel and VMware chief executive Pat Gelsinger, and total funding comes to $39.5 million, implying a $12 million seed. Both checks rest on one proposition: that the vertical-cavity surface-emitting laser becomes a volume component inside the AI rack. PicoJool's first product line, aimed at scale-up links and unveiled over the summer, supports more than 37 gigahertz of bandwidth, and the company is developing microVCSEL parts that are smaller and draw far less power than conventional wire-bonded counterparts, the report says.
The proceeds go to facilities in the United States and Taiwan and to hiring across R&D, sales, and marketing, with a view to products supporting both hyperscalers and data center operators. In Yuen's telling, the next phase is qualification rather than invention. "Reaching 200G per lane proved that the performance is there. Our focus now is execution," the chief executive said, describing a round that funds teams in both countries, product qualification with customers, and a portfolio moving from chip-level VCSELs and microVCSELs into active optical cable and near-packaged optics modules.
A founder who ran VCSEL R&D at Lumentum
Yuen's path runs through the companies he now competes with. He was general manager of Coherent's semiconductor business in the mid-2000s, then spent more than a decade running R&D development of VCSELs for data centers and AI systems at Lumentum, which the report describes as Nvidia-backed. Both Coherent and Lumentum are working on VCSELs for AI interconnectivity, part of a wider set of light-based technologies the report describes as pushing AI infrastructure performance higher. PicoJool's case to customers therefore cannot be the device category, which its larger rivals already work in, but the microVCSEL line and what it does for power and density, sold by a founder who learned the technology inside both of them.
Gelsinger, ousted from Intel in late 2024, said the team combines VCSEL performance, systems expertise and a manufacturing base "capable of supporting tremendous scale." That is a claim about capacity rather than a contract, and it lines up with the two-site expansion the round is meant to finance. The report names no hyperscaler customer, no supply agreement and no revenue figure, which leaves the round carrying technology risk and qualification risk at the same time, plus the cost of a sales and marketing buildout that suggests the customer conversations sit ahead of the company rather than behind it.
Both venture firms on the cap table are run by men who have led large chip companies, Clemmer at NXP and Gelsinger at Intel, and both put money into a parts maker. What that likely buys PicoJool is a hearing at the qualification stage with people who specify components for a living, though goodwill is not a purchase order.
What equity buys before a purchase order
The AI buildout now divides into a contracted core of powered shells leased to hyperscalers on long terms and the component layer that sells into them, financed on different terms: the core draws infrastructure capital against leases, while a laser developer raising a Series A sits at the far end of that spectrum, its equity repaid through volume in somebody else's platform rather than cash flow from a lease.
Qualification is the gate, and it runs on the buyer's calendar. Optical parts enter a platform when the hyperscaler's roadmap reaches them, so a design win converts into volume only after the buyer's systems ship, which is why the round is sized for two facilities and a hiring plan rather than for vertical integration that would let a startup set its own schedule. Demand is not the constraint—OpenAI's $750 billion compute budget, as this publication reported in August, sizes the buildout these parts would serve. What the market has not been shown is a name attached to PicoJool's pipeline, and against Coherent and Lumentum a place on a qualification list is the measure of the company.
Qualification is the gate, and it runs on the buyer's calendar.
Save this analysis and keep the funds you follow together in My Desk.
Sign in to save articles or follow funds.