AI's Next Bottleneck Isn't the GPU: PicoJool Raises $27.5M to Fix How Chips Talk to Each Other
Everyone is asking who can buy the most GPUs. Far fewer are asking how fast those GPUs can actually talk to each other.
PicoJool has just raised $27.5 million in Series A funding to go after exactly that problem. Socratic Partners led the round, Hudson River Trading joined in, and total funding now stands at $39.5 million. Earlier backing came from Playground Global, the venture firm where former Intel CEO Pat Gelsinger is a general partner.
Here's what the company is building and why it matters:
→ PicoJool is developing VCSEL (Vertical Cavity Surface Emitting Laser) technology, optical chips that move data between processors using light instead of electrical signals
→ Its lineup includes 100G and 200G VCSEL products, plus lower-power "microVCSEL" designs that reach high bandwidth by running many optical links in parallel
→ The company says the technology can scale from 800 Gbps to 1.6 Tbps, 3.2 Tbps, and beyond, while cutting the energy needed to move each bit
→ It is working with Taiwan's WIN Semiconductors and other foundries to get its 200G products ready for production
Why this deserves attention: a giant AI cluster is only as fast as the links between its chips. If data can't move quickly, expensive accelerators sit idle waiting for their next input. And every extra watt spent on moving data is a watt not spent on computing. As clusters grow from thousands to hundreds of thousands of chips, interconnects stop being a supporting detail and become the constraint that decides how efficiently the whole system runs.
This isn't an isolated bet either. In the same week, Tower Semiconductor announced a roughly $4 billion plan to make Japan its largest hub for optical chips. Money is flowing into the "plumbing" of AI, not just the processors.
My take: the first phase of the AI boom rewarded whoever could get the most compute. The next phase will reward whoever can connect that compute without burning through power and budget. That's a less glamorous story than a new model launch, but it's often where the durable businesses get built.
The open question for PicoJool is the classic one for hardware startups: getting from a strong design to reliable, high-volume manufacturing. That's where many promising chip companies stumble.
Do you think the real winners of the AI infrastructure race will be the chip makers everyone talks about, or the quieter companies solving the connection problem?
#AI #AIInfrastructure #Semiconductors #Photonics #StartupFunding #DataCenters
— 𝔖𝔞𝔫𝔡𝔢𝔢𝔭 ℜ𝔞𝔦𝔷𝔞