Most of the public conversation about artificial intelligence is a conversation about compute. Bigger models, faster chips, larger clusters. Yet inside the data centers where these systems actually run, a quieter constraint has moved to the center of the story. The harder problem is no longer how fast a processor can think. It is how fast data can move between processors. Linkstar, a Singapore-based deep tech company, has built its business around that overlooked problem.
The scale of the gap is striking. Over roughly the past two decades, raw computing performance has improved by about 60,000 times, while the bandwidth of the interconnects that carry data between chips has improved by only about 30 times. That leaves a gap of nearly 2,000 times between how quickly modern systems can process information and how quickly they can move it. As a result, data movement, rather than computation, has become the primary bottleneck for large-scale artificial intelligence. The performance of training and inference workloads now depends as much on network infrastructure as on the processors themselves.
This is the constraint Linkstar set out to address. The company designs photonic integrated circuit engines, the optical hardware at the heart of the transceivers that shuttle data across and between data centers. Its current focus is a 1.6 terabit engine, alongside an 800 gigabit generation, built specifically for the demands of artificial intelligence and cloud networks. The premise is simple to state and difficult to engineer. If bandwidth has become a scarce resource, then the most valuable thing a company can build is a way to move far more data with far less energy.
The reason this matters now is that the older methods of moving data are running out of room. Every AI query, cloud workload, and streamed application depends on optical interconnects to carry information between machines. As AI clusters grow from thousands of accelerators to hundreds of thousands, the demand for bandwidth keeps climbing, pushing the industry through successive transitions from 400 gigabit to 800 gigabit, then to 1.6 terabit, and beyond. At the same time, traditional electrical wiring and pluggable optical modules are hitting fundamental limits in power, density, and performance. That collision is forcing a shift toward more tightly integrated optical designs, including on-board and co-packaged optics, and it is turning optical interconnect into one of the most durable growth areas in the entire semiconductor ecosystem.
The market figures reflect that momentum. According to LightCounting, the optical transceiver market stood at about 23.8 billion dollars in 2025 and is projected to approach 60 billion dollars by 2031. Analysts at Yole Group have tracked the silicon photonic chip market climbing from roughly 95 million dollars in 2023 toward 863 million dollars by 2029, while Mordor Intelligence has projected the co-packaged optics segment growing from about 121 million dollars in 2025 to 764 million dollars by 2031. LightCounting also notes that sales of 1.6 terabit transceivers are passing two billion dollars for the first time. These numbers describe an industry at an inflection point rather than a mature one, which is part of why a focused company entering now can matter.
Linkstar’s answer is to treat the engine as a platform rather than a single product. The same underlying architecture is designed to scale from 100 gigabit all the way to 1.6 terabit, which means each generation builds on the last instead of starting over. That continuity is a strategic asset. A lead earned in one generation can compound into the next, rather than resetting every time the industry moves to a faster standard.
There is also a national dimension to the effort. Linkstar anchors its design, fabrication, and testing within a single Singapore-based ecosystem, tightening the feedback loop between the teams that imagine the hardware and the teams that build and validate it. In an industry where supply chains are often fragmented across many countries and vendors, keeping those functions close together can speed up learning and reduce the friction that slows larger, more distributed organizations.
The broader point Linkstar is making is that the foundations of the AI era are not only the famous processors that get the headlines. They are also the largely invisible plumbing that carries data from one place to another. As the company’s chief executive, Dr. Ben Yuan, frames the mission, moving data as light lets AI and cloud data centers carry far more information with far less energy, turning optical interconnect from a bottleneck into a competitive edge. For an industry straining against the limits of how fast it can move information, that is not a small ambition. It is an attempt to remove one of the most fundamental constraints standing in the way of what comes next.
