Panning for Gold in the Light: Where AI Goes Next, Through the Eyes of a Tech Investor
Lately, "light" has become the most crowded trade in the public markets.
Lately, "light" has become the most crowded word in the secondary market.
Optical modules, CPO (Co-Packaged Optics), thin-film lithium niobate, and all-optical switches have taken turns trending, with previously obscure companies being repeatedly bid up by capital mid-session. A joke making the rounds: you want to stand in the light, not just watch it stand there.
This isn't a narrative unique to A-shares. In April 2026, Lightelligence listed on the Hong Kong Stock Exchange, becoming the "world's first AI optical computing stock." And at this year's COMPUTEX, Jensen Huang shared the stage with Marvell's CEO, clearly pointing to "connectivity" — the domain where light operates — as the next decisive battleground for AI infrastructure. He called Marvell "the next trillion-dollar company" on stage, a single sentence that ignited the entire optical communications sector.
As an early-stage investment firm that has been positioning in the optical space for years, our view is this: light is not a "universal cure" for AI, but as large models push compute, power consumption, and real-time performance to their limits simultaneously, light has become very difficult to bypass.
In this research report, we try to address several questions:
- In optical communications, what is real demand versus mere concept?
- How far away is optical computing, really?
- And in this wave of opportunity, where do Chinese companies stand?
Interactive perk:
When do you think optical computing technology will be realized? Share your thoughts in the comments. By 17:00 on July 16, 2026, the 2 most thoughtful commenters will receive a book recommended by Feng Li.
We continue to follow technological developments in the optical space. If you are a relevant entrepreneur or practitioner, feel free to contact the author of this article, Yongcheng Yang (yangyongcheng@freesvc.com).
01 Putting the story of light on one map: from materials and chips to communications and computing
The optical chain is long, with many technical branches. Materials, chips, devices, modules, communications, interconnects, and computing nest within one another, and several concepts are often conflated. Before formally entering this research report, we need to place them back onto a single map.
"Light" chain roadmap.
If the entire chain were compressed into one sentence, it could roughly be understood as: first create the light, then write information onto it, let it transmit between different devices, and finally receive and restore it; one step further, use light directly to complete computation.
Following this chain downward, the furthest upstream is materials and light sources. Lasers are responsible for generating light; materials such as silicon, indium phosphide, and thin-film lithium niobate respectively undertake light guiding, light emission, or modulation functions, and influence device speed, loss, and integration approach. Below that are optical chips and optical devices, including modulators, detectors, wavelength-division multiplexers, optical phased arrays, etc. They respectively handle loading signals onto light, receiving light, merging or splitting different wavelengths, and controlling light direction.
Optical devices such as lasers, modulators, and detectors, packaged together with electrical chips like drivers and DSPs and interfaces, constitute optical modules. They connect host equipment to optical fiber, responsible for completing the conversion between electrical and optical signals. In other words, the optical module is not a separate technical route, but the core "interface" in an optical communications system.
Optical communications refers to a larger system: using light to transmit information. In the past, light has been widely used for transoceanic, intercity, base station, and data center communications; in the AI era, it has further entered servers, GPUs, and even chips on the same board. These shorter-distance, higher-density applications are typically called optical interconnects. Optical modules, CPO, and LPO address how light enters devices and approaches chips; all-optical switches are responsible for directly switching optical paths in the network.
Optical computing takes one more step forward. Optical communications solves how to transmit data faster; optical computing asks whether light itself can do the computing. The former still serves information transmission, while the latter uses physical properties of light such as propagation, interference, and diffraction to complete operations. The two share some underlying materials, chips, and manufacturing capabilities, but face different problems and are at different industrial stages.
The various "light" concepts in today's market can basically all be understood by placing them back on this chain.
02 AI isn't running out of compute — the roads are jammed
Most people assume AI's bottleneck is compute, that GPUs (Graphics Processing Units) aren't enough. But what hit the ceiling first wasn't "compute" — it was "connect."
I. The memory wall and the physical limits of PCBs (Printed Circuit Boards)
Think of each GPU as a factory. Training a large model means tens of thousands of factories continuously moving and coordinating. What actually stalls the assembly line is rarely any single factory's capacity, but the road between them. The first wall is the memory wall.
To alleviate this, the industry created HBM (High Bandwidth Memory), stacking layers of memory like a sandwich — effectively spreading a line's thickness into a plane, with capacity leaping accordingly. But circuit boards themselves have limits: today's AI server boards are already stacked to roughly several dozen layers, while phones have only six to eight, approaching the ceiling. Even the "electronic cloth" used in these laminated boards has been bid up as a hot theme in the secondary market.
Core hardware composition of AI training and inference servers. GPUs, HBM, CPUs, and DDR handle computation and data access; PCIe, NVLink, NVSwitch, etc. handle high-speed interconnects; NICs, storage, power supply, cooling, and management modules together support the entire machine's operation.
II. Three layers of interconnect under simultaneous pressure
More棘手的是，这不是一段路在堵，而是三层路同时拥塞。
Server-to-server interconnect, what the industry calls Scale-out;
Inside one server, interconnect among multiple GPUs, called Scale-up;
Further inward, chip-to-chip interconnect on the same board, called Scale-in.
All three layers are increasingly beyond electricity's capabilities.
A telling comparison: copper cable can typically only stably transmit about three centimeters on a board, and only several meters inside a chassis, while optical fiber remains composed at one hundred meters or even several kilometers; on power consumption, one copper interconnect takes roughly fifteen watts, while switching to optical interconnect takes roughly five.
So rather than saying "light" suddenly became hot, it's more accurate to say that electrical paths first approached their own physical ceiling.
III. Why NVIDIA got "anxious"
NVIDIA's recent moves are quite telling: it has successively invested in Marvell, Lumentum, Coherent and other industry chain companies, and reached cooperation with Corning. NVIDIA and other AI compute giants have proposed technical routes including CPO, MicroLED (Micro Light-Emitting Diode), microring modulators, and OCS (Optical Circuit Switch). A giant that originally only did compute is now personally directing optical communications for a simple reason: its demand is there, and existing suppliers' pace can't keep up.
When the chain master gets anxious, the entire industry is forcibly ignited. The demand is real, but technical routes have not yet converged, and the mud-sand-mixed reality is equally real — not lacking those who came to "ride the light wave."
Some chose the opposite direction. Rather than painstakingly connecting thousands of chips, why not make an entire wafer into one enormous chip, letting most computation happen internally to bypass the interconnect hurdle? The representative doing this is America's Cerebras. But this is a purely electrical chip route, with non-trivial costs in cooling and yield, ultimately unable to bypass certain other physical limits.
Yet one thing almost no one doubts: everyone has identified the same exit — light.
03 Why light, specifically?
Why light can take this baton starts with a piece of basic physics.
Photons have no mass, so they are inherently superior to electrons in speed, power consumption, and signal-to-noise ratio. The simplest proof: undersea cables can send signals across thousands of kilometers to the other shore — something electricity cannot do.
This is also the underlying judgment behind our decade-long bet on light, never wavering: in many links of communications and computing, light's physical endowment is simply better than electricity's.
In fact, optical communications has long been everywhere — we just don't normally feel it. Transoceanic undersea cables, metropolitan area networks between cities, backhaul between generations of base stations, and even interconnection of individual servers in data centers — all fundamentally run on optical fiber. But AI has introduced new demands in this round: inside servers, between chips — these short-distance scenarios where electricity previously sufficed — now need to switch to light.
So how is the signal loaded onto light? At bottom, two paths: one is having the light source carry the signal itself; the other is having the light source shine steadily, then adding a "modulation" behind it to alter it. The latter path follows silicon photonics, the current mainstay, but single-wave rate basically tops out at 200G. To go higher requires a new material called thin-film lithium niobate, an almost unavoidable option for hitting 1.6T, 3.2T high rates. On this line, Chinese teams are relatively ahead, and we invested in one company.
However, there's another layer of uncertainty about light's incremental role in AI infrastructure, and it comes from AI itself. Right now the biggest demand side for compute is training, with companies everywhere stacking GPU clusters; but the next step, inference — how large and how to build data centers specifically for it — no one can say for certain right now. If intelligent agents and on-device assistants really take off, with ever-rising requirements for low latency, that would push optical demand up another level; but if inference mainly reuses existing training clusters, the increment would be much smaller. This uncertainty is an important reason why optical communications is "hot yet chaotic."
There's also an unavoidable hard bone in optical modules called DSP (Digital Signal Processor), equivalent to the "CPU" of optical communications, specifically responsible for straightening out signals that have been distorted or blurred — without it, high-speed signals are simply unrecognizable. The problem is this chip is both expensive and power-hungry, accounting for roughly one-third of optical module cost and nearly half its power consumption; delivery lead times for high-end DSPs have stretched to nearly a year, and they're basically held by two American companies — Marvell and Broadcom.
800G optical module internal structure schematic. The transmit and receive ends respectively complete electrical-to-optical and optical-to-electrical conversion, with main components including DSP, driver, laser, AWG wavelength-division multiplexer, photodetector, and TIA amplifier, etc.
To speak frankly: we aren't being choked by anyone's administrative order, but our own technology genuinely hasn't reached that level yet. It's too complex — a chip with very high technical complexity and difficulty — and we still need some time to catch up on high-end optical DSPs.
The broad direction of light is doubted by almost no one; but which specific path to take, who succeeds first, is far from consensus.
So in this still-unclear landscape, where exactly does China stand?
04 On this chessboard, China stands on the "creation" side for the first time
I was previously asked a question: Is optical chip yet another "domestic substitution" direction?
My answer: No. It's different from the story we're familiar with. Back when we looked at GPUs, the essence was domestic substitution — NVIDIA had already emerged overseas, and we were chasing from behind; but today when we look at the new directions in the optical domain, such as CPO, LPO (Linear-drive Pluggable Optics), thin-film lithium niobate, OPA (Optical Phased Array), etc., China and overseas started at the same time from the same starting line. From the current position, we are not at a disadvantage this time, and in some links we're even ahead — thin-film lithium niobate chips, for example. Of course, in the technical areas where we lead, there is often no clear "anchor," meaning no overseas benchmark to reference. Many peers might see this as uncertainty and risk, but in my view, this is precisely a good thing, because it means you are the first.
If you are the first to eat the crab, you also have the chance to grab the biggest, best crab.
Currently in the AI optical interconnect track, there are two types of players. One is the several optical module giants that have risen most sharply in the secondary market. What they're mainly doing is already-mass-produced traditional optical modules, with demand primarily coming from NVIDIA's data center builds — essentially continuing to grow within an existing large market. In this traditional pluggable optical module market, the major-player landscape is already relatively stable, with little opportunity for new entrants like startups; but in new directions, startups and giants basically stand at the same starting line, and because they're more focused and faster in decision-making, they actually have a window period.
The optical chessboard is large. Over the years we've been laying out position by position along each technical route, investing in relevant entrepreneurial teams along every key path.
FreeS Fund's investment layout in the optical domain.
First, the most closely watched: optical computing. We first invested in Lightelligence in 2017; it listed on the Hong Kong Stock Exchange this April, taking a photonic-electronic hybrid route — using light to accelerate the matrix multiplication that weighs heaviest in AI, with remaining operations still handled by electrical chips. Another company we invested in 2022, LightStandard, takes a different path, using silicon photonics plus phase-change material to merge storage and computation into a single unit, doing "compute-in-memory" — the benefit being smaller unit size and lower power consumption. For the same thing, we bet on two different routes.
The second line is thin-film lithium niobate. As mentioned earlier, pushing rates toward 1.6T, 3.2T, this material is almost unavoidable. Whether upstream wafers or chip design, Chinese teams have been ahead these past few years; Anpai Xinyan, which we invested in in 2021, is one of them.
The third line is OPA, optical phased array. Its principle is somewhat like the flat-panel antennas in 5G, using phase to control light direction without relying on mechanical rotation. This has two major applications: one is all-optical switches (OCS), the other is solid-state LiDAR — the latter beginning to ship in vehicles this year. There aren't many teams globally that can pull this off; Lice Technology, which we invested in in 2019, is one of them.
Beyond these three, in silicon photonics we invested in Luowei Technology, which follows the FMCW (Frequency-Modulated Continuous Wave) route; in optical modules we invested in AfaLight; and in metalenses we invested in Xuanxiang Technology. Connecting these together, from optical interconnect to optical computing, from materials to devices, we have positions along almost every key path in light.
Of course, we're not at the forefront on every line. Like that hard bone DSP — as mentioned earlier, our technology hasn't reached that level yet. In terms of category completeness, we're probably the most complete; but to say every category is at the top — certainly not yet, that takes time. But precisely because the direction is right and the layout is complete, I'm confident in the final outcome.
And among this entire layout map, what excites us most is still the diamond in the crown. What exactly is new about optical computing, and why did we bet on two routes to gamble on its future?
Betting on a farther future
Optical communications uses light to "transmit" data faster; optical computing goes one step further — it simply lets "compute" itself be handed to light.
Today's AI computation relies on thousands of electrical chips multiplying and adding around the clock, consuming power and generating heat; while light runs fast and consumes almost no energy. If light can shoulder the most repetitive, most burdensome calculations in AI, it would theoretically be both faster and more efficient. So optical computing is one step beyond optical communications — its ceiling is higher, but also farther and harder.
Most so-called optical computing today has actually only taken a small step: using light to replace the multiply-accumulate operator, swapping out the most core step for light, while the overall computational topology and architecture remain identical to electrical chips. The true next step is to use light's physical properties themselves to complete an entire computation. This sounds mystical, but actually has a simple example.
Any waveform can be decomposed into a stack of sine waves of different frequencies — the Fourier transform, which algorithmically is a matrix operation; while a beam of white light passing through a prism and separating into various colors, this utterly ordinary phenomenon, is essentially the prism doing that calculation for you — you just read the result at the exit. Prisms, gratings, metalenses, and other devices are "physical computers" that consume no electricity.
After white light passes through a prism or grating, it is decomposed into light of different frequencies. Optical devices use their own physical structure to directly complete frequency decomposition; input the optical signal, and read the result at the exit.
But this path is still early, with clear shortcomings. Implementing computation directly through light's physical structure means model parameters are固化 in the device — unchangeable after training, like a photo that is fixed once developed. So it only excels at fixed, unchanging tasks, such as a certain type of image recognition; it cannot switch models on demand like a general-purpose GPU, making it more suitable for edge-side, fixed-scenario small tasks. Expecting optical computing to now replace the electrical compute market dominated by NVIDIA GPUs is still premature.
So why did we dare to bet early? Because for this kind of hard tech, the direction is almost never wrong — the suspense is only who succeeds, and when. Like silicon carbide: its first-principles validity was established very early, but when it was first made, downstream demand hadn't emerged, no one used it; early companies had to survive on other products until new energy vehicles' high-voltage platforms emerged, demand arrived, and then they took off.
Light is somewhat luckier than silicon carbide, because the AI application scenario is visibly right there.
Companies built through grinding away at technology have a very distinctive valuation curve shape. For many years, it stays pinned at very low levels, often with valuations under two billion; but once first principles are validated by the market and single-product sales exceed one hundred million, there are only very few rounds of window in between before valuations shoot straight past ten billion. After that, the question is no longer whether it's expensive, but whether you can get any allocation at all.
Connecting the entire map above, what we want to do is actually quite clear. Grounded in the present, establishing positions in optical communications, optical interconnects, optical chips — domains already happening; while also looking to the future, pre-positioning for farther directions like optical computing.
The era of light has just begun. We want to stand in the light, together with more entrepreneurs.
Illustrations in this article were AI-assisted, using tool: ChatGPT