Every new AI chip arrives with a bigger compute number. Fewer arrive with an honest admission that those chips increasingly sit idle, waiting for data to trickle in over copper wires that were never designed for this kind of workload.
Singapore’s Staera Semiconductor is betting its future on that gap. The deeptech startup, formerly known as New Silicon Corporation (NSC), emerged from stealth today with US$4.6 million in pre-seed funding to commercialise Lightwire.
Lightwire is an optical interconnect that generates light inside the silicon chip itself rather than bolting a light source on next to it.
Also Read: Nexstrom lands US$12M to bring 2D semiconductors to 12-inch wafers
Openspace Capital led the round, with participation from SGInnovate, Black Kite Capital, 28 Holdings, In Group and Decimus Technologies. Staera is a spin-off from the Low Energy Electronic Systems (LEES) research group at the Singapore-MIT Alliance for Research and Technology (SMART), MIT’s first research centre outside the US.
Copper is running out of road
The problem Staera wants to solve is not new, but AI has made it acute. Inside an AI system, data has to travel between processors, memory stacks, boards and racks, over distances ranging from a millimetre to a metre. According to the company, copper’s electrical performance degrades roughly a thousandfold across that range. Push more data further over copper and you pay for it in power and heat.
That is a painful trade at a time when electricity is becoming the binding constraint on AI. The International Energy Agency expects global data centre electricity use to more than double, from about 415 terawatt-hours in 2024 to around 945 terawatt-hours by 2030. In Southeast Asia, where operators are already running short of power for AI servers, every watt saved on moving data is a watt that can go to compute, or not be drawn from the grid at all.
Even Nvidia, which said in 2025 it would stay with copper for as long as it could, now plans to use both copper and optics in its AI systems. Analysts tracking AI’s emerging bottleneck economy increasingly point to interconnects, packaging and memory bandwidth, rather than raw compute, as the place where the next squeeze, and the next wave of spending, will land.
Building the lamp into the wafer
Most optical links today are assembled from separate parts: a laser or LED made elsewhere, attached next to the chip and coupled to waveguides. Every join adds cost, power draw and another thing that can fail.
Staera’s pitch is that it fabricates the whole optical link on a single commercial silicon wafer using standard fab steps. That means gallium nitride (GaN) LEDs as the light source, silicon nitride waveguides to route the light, and the CMOS circuits that drive it, all on one piece of silicon. Lightwire sits inside the bridge or interposer chip that already connects processors and memory in today’s AI hardware, so, in theory, it slots into existing packaging flows.
“Copper performance plummets over the millimeter-to-meter distances that matter most in AI accelerators, and bolt-on optics add power and complexity,” said Dr Andrew Kim, CEO of Staera. “We build the light into the chip, the entire optical link inside a chip, so the connection gets smaller, cooler and faster with each generation.”
Also Read: NASA-linked, MIT-trained founders’ nSWX raises US$2M for AI chip packaging
The headline numbers are striking. Staera says Lightwire is designed to move data at under 0.2 picojoules per bit, against about 5 pJ/bit for co-packaged optics, the current state of the art. At memory scale, it says moving 24 terabytes a second would take under 50 watts with Lightwire, versus nearly a kilowatt with co-packaged optics. It also claims a reach beyond 50 mm, compared with 2 mm for the UCIe-A die-to-die links used in today’s chiplet designs.
The fine print hides in “designed to”
Note the verb. Almost every efficiency figure Staera has released is a design target, not a measured result from a shipping product. Commercial release is pencilled in for 2030, four years and, almost certainly, several funding rounds away.
There is also an engineering trade-off worth understanding. Each Lightwire link runs at a modest 4 to 8 Gbps, far slower per lane than the high-speed electrical and laser-based links found in data centres today. The bet is on density rather than speed: more than 600 links per millimetre of chip edge, all running in parallel. That approach plays to the strengths of LEDs, which are cheap and frugal with power, but cannot match lasers on per-channel bandwidth.
Staera does have manufacturing pedigree on its side. It says it has been building GaN LEDs into silicon chips in commercial foundries since 2016, with yields now at 98 to 99 per cent. That longevity cuts both ways, though. A decade of R&D culminating in a US$4.6 million pre-seed cheque is a modest sum for a company that wants to rewire AI hardware, and a reminder of how capital-hungry chip startups are.
Investors seem comfortable with the long horizon. “We backed Staera because of the strength of its technology, the depth of its team and its potential to define a new category in semiconductor connectivity,” said Hian Goh, co-founder of Openspace Capital.
The startup also won the ECTC Start-up Competition at the IEEE Electronic Components and Technology Conference in May, and placed runner-up at Lam Research’s 2026 Lam Capital Venture Competition the same month.
A crowded race to dethrone copper
Staera’s claim to be the first and only optical interconnect built fully into the chip will be tested by a well-funded field. In the US, Ayar Labs and Lightmatter have raised hundreds of millions of dollars for silicon photonics, while Marvell agreed to acquire photonic interconnect startup Celestial AI in late 2025.
California’s Avicena is the closest conceptual cousin, using microLED arrays for short-reach chip links. Broadcom and Nvidia, meanwhile, are pushing co-packaged optics into networking switches.
Closer to home, Staera joins a small but growing cohort of Southeast Asian semiconductor startups moving beyond assembly, including photonics player LightSpeed Photonics. Its real differentiator is monolithic integration: if the light source is grown on the same wafer as the circuits, it sidesteps the assembly and alignment headaches that dog bolt-on optics. Whether that holds at volume is the question customers will ask first.
Singapore wants a bigger slice of the AI hardware stack
Staera’s emergence fits neatly into Singapore’s push to climb the chip value chain. A*STAR and the Economic Development Board recently unveiled SG Semiconductor as a national identity for the sector, A*STAR has teamed up with Synopsys on AI chip packaging, and Applied Materials has expanded its Singapore footprint with a US$500 million campus. Interconnects and advanced packaging, the unglamorous plumbing between chips, are exactly where the city-state hopes to matter.
Also Read: Amplicity raises US$1M to put Singapore’s sleeping data centre batteries to work
The fresh capital will go towards advancing the technology, expanding the team into the US, and deepening work with customers and supply chain partners. The US push is telling: the buyers who will decide Lightwire’s fate, hyperscalers and AI chip designers, mostly sit in California.
For now, Staera has a compelling physics argument, a manufacturing track record and a four-year runway to prove its numbers survive contact with real hardware. If they do, the most important part of tomorrow’s AI chips may be the light nobody sees.
The post Copper is choking AI. Staera secures US$4.6M to swap it for light appeared first on e27.