Gelsinger-Backed PicoJool Raises $27.5 Million to Tackle the Optical Bottleneck Inside AI Data Centres

The race to build ever-larger artificial intelligence systems is usually told as a story about chips: who makes the fastest processors and who can buy enough of them. But inside the world's biggest AI data centres, a less visible constraint is becoming just as important — how quickly and efficiently data can move between those chips.

PicoJool, a Palo Alto-based start-up focused on that problem, announced on Thursday, September 24, that it has raised $27.5 million in a Series A funding round led by Socratic Partners, with participation from Hudson River Trading, according to SiliconANGLE and a company announcement.

The round follows a $12 million seed round in November 2025 led by Playground Global, the venture firm co-founded by former Intel chief executive Pat Gelsinger. PicoJool has now raised $39.5 million in total.

The connectivity bottleneck

Modern AI models are trained and run on clusters of thousands of processors that must constantly exchange data. A single GPU is only as useful as its ability to send and receive information from its neighbours, whether they sit on the same board, in the same rack or across a data centre.

For short distances, those connections have traditionally used copper wires. But as data rates rise, copper struggles: signals degrade over distance, and pushing more data through copper consumes more power. Optical connections, which use light to carry data through fibre, can move far more data over longer distances with lower losses, but they have historically been more expensive and power-hungry at the component level.

As AI clusters grow, the energy consumed by moving data is becoming a significant share of total power use. That makes efficient optical interconnects part of the AI infrastructure race, alongside memory, advanced packaging, networking and power delivery.

Gelsinger, whose Playground Global backed PicoJool's seed round, summed up the argument. "The network is the AI, and connectivity is becoming one of the defining constraints on performance," he said, according to SiliconANGLE.

What PicoJool builds

PicoJool develops optical chips and modules based on vertical-cavity surface-emitting lasers, or VCSELs. VCSELs are small semiconductor lasers that emit light perpendicular to the surface of the chip. They are widely used in data communications and consumer electronics — including facial recognition sensors in smartphones — because they are relatively inexpensive to manufacture and energy-efficient.

PicoJool's products include 100G and 200G VCSEL interconnects, as well as microVCSEL configurations offering 50G NRZ and 64G NRZ/PAM4 options. NRZ and PAM4 refer to different methods of encoding data into light signals. The company says its technology can support connections with bandwidth of up to 3.2 terabits per second.

The company's founder and chief executive, Al Yuen, said reaching 200G per lane had been a key technical milestone. "Reaching 200G per lane proved that the performance is there. Our focus is now execution," he said, according to SiliconANGLE.

Where the money will go

PicoJool plans to use the new funding to expand its team in the United States and Taiwan, qualify its products with customers, accelerate research and development, grow its operations and sales, and prepare for mass production. The company is working with WIN Semiconductors, a Taiwanese compound semiconductor foundry, on manufacturing.

Taiwan's role is significant. The island is home to much of the world's advanced semiconductor manufacturing and packaging, as well as a dense ecosystem of optical and compound semiconductor suppliers. Being close to that ecosystem can help a hardware start-up move faster from prototypes to volume production.

Why investors are interested

PicoJool's Series A is modest compared with the mega-rounds raised by AI model developers, but its strategic importance may be greater than its size suggests.

AI computing economics are increasingly a systems problem. Hyperscalers and AI companies are spending hundreds of billions of dollars on data centres, and every improvement in the efficiency of data movement can translate into more computing capacity within the same power budget. That makes optical interconnect suppliers potential beneficiaries of spending by multiple chipmakers and cloud providers, rather than dependent on the success of a single AI company.

The investor mix is also telling. Socratic Partners, which led the round, is a technology investor. Hudson River Trading, one of the world's largest quantitative trading firms, operates a business in which the speed of data movement is critical, and its participation suggests an interest in low-latency connectivity technology.

A competitive field

PicoJool faces formidable competition. Large companies such as Broadcom, Nvidia, Marvell and Coherent are major players in optical networking for data centres. Nvidia and Broadcom have also been developing co-packaged optics, which integrate optical components directly alongside processors to reduce power consumption.

Many start-ups are working on alternatives, including silicon photonics, micro-LED-based links and new laser technologies. Each approach has trade-offs in cost, power, reach and manufacturability.

PicoJool's bet is that VCSEL-based technology, already proven at scale in other applications, can deliver the right balance of performance, efficiency and cost for short-reach connections inside AI clusters.

The broader trend

PicoJool's round is part of a wider wave of investment in the physical infrastructure behind AI. Venture capital is increasingly flowing below the application layer — into chips, networking, memory, power and cooling — as investors look for companies whose value is tied to hard engineering rather than to a temporary advantage in AI models.

On the same day, other deals highlighted the same theme, from Island's $400 million round for enterprise browser security to BigHat Biosciences' $75 million raise for AI-designed antibodies. In each case, investors were paying for control of a hard-to-replicate asset — hardware, data or a security choke point — rather than for AI capability alone.

What it means for India

For India, which is building its own semiconductor ecosystem and has a large community of chip-design engineers, the growth of the optical interconnect market is significant. Photonics and advanced packaging are emerging areas where Indian engineers, both at home and in the diaspora, are increasingly involved, and where new start-ups could find opportunities as AI infrastructure spending expands.

The bottom line

As AI systems grow, the challenge is no longer only how to compute faster but how to move data more efficiently. PicoJool's $27.5 million round is a bet that light, delivered by tiny lasers, will play a growing role in solving that challenge.

If the company can move from technical milestones to mass production and customer adoption, it could become one of the many specialised suppliers quietly powering the AI era.