What Does POET, Elon Musk & AI Have in Common- Bottlenecks

The Next Bottleneck in AI Isn’t Compute. It’s Light.

As clusters scale toward hundreds of thousands of accelerators, the constraint is shifting from how many GPUs a company can buy to how fast it can move data between them — and that is the problem POET Technologies was built to solve.

For two years, the AI infrastructure conversation has revolved around a familiar set of bottlenecks: GPUs, high-bandwidth memory, CoWoS packaging, liquid cooling, power delivery, and data-center real estate. Every one of those matters. But as clusters grow from tens of thousands of accelerators toward hundreds of thousands — and the largest operators now talk openly about millions — a quieter constraint is moving to the front of the line: data movement.

The physics are unforgiving. Training and running frontier models requires thousands of processors working in concert, constantly exchanging information. As the cluster grows, the volume of data crossing the system rises faster than the compute itself. At some point, building faster chips stops helping, because the information cannot get between them quickly enough. Traditional electrical interconnects make the problem worse as speeds climb: they burn more power, throw off more heat, and become harder to scale.

That is why the industry is turning to light. Optical networking moves data as photons rather than electrons, and a growing consensus holds that photonics is the next major infrastructure upgrade for AI data centers — the same kind of step-change that high-bandwidth memory was for the memory layer.

Where POET Fits

POET Technologies $POET develops optical technology aimed squarely at this problem: accelerating data movement across AI networks and data centers. Its core product, the POET Optical Interposer, is designed to simplify the assembly and integration of optical components — making optical systems smaller, more power-efficient, and potentially cheaper to manufacture.

The analogy POET’s followers like to use is a transportation system. If NVIDIA $NVDA builds the high-performance engines that make AI run, POET is trying to build the highways that let all those engines actually move. And nearly every large AI deployment needs that connective tissue.

The manufacturability angle is the part that gets overlooked. Optical hardware has historically been held back by slow, manual, low-volume assembly. The companies that win the optical decade will be the ones that make photonics behave like semiconductors — repeatable, scalable, and producible in volume. POET’s Optical Interposer is a bet on exactly that transition.

Even the Compute King Is Going Optical

The strongest evidence that the interconnect bottleneck is real comes from the company that sells the compute. At GTC 2025, NVIDIA unveiled its Spectrum-X and Quantum-X silicon photonics switches — co-packaged optics designs that move the light engine right next to the switch chip. NVIDIA says the platforms are built to connect millions of GPUs at 1.6 terabits per second per port while delivering roughly 3.5x better power efficiency than traditional networking. When the company whose entire business is selling processors decides it must also build the optical fabric between them, the bottleneck has stopped being theoretical.

Just as telling is who NVIDIA brought in to do it. Its co-packaged optics effort was announced with a roster of named optics partners — including Coherent (NYSE: $COHR), Corning, Lumentum, TSMC and others — the established supply chain that already builds the lasers, transceivers, and photonic components the AI buildout runs on. Coherent and the incumbents attack the problem from the top down, with mature volume manufacturing and deep customer relationships. $POET attacks the same problem from a different angle: an interposer platform designed to make optical engines cheaper and more scalable to assemble. The giants are validating the category. $POET offers small-cap, high-optionality exposure to the same secular shift.

Why Now

The investment case is unusually simple to state. If you believe AI spending keeps growing — and most investors do — then demand for networking infrastructure grows alongside it, because every new cluster needs processors, storage, switching, and increasingly sophisticated optical connections. POET is trying to supply part of that stack. Crucially, it does not need to win the entire market to matter: the company generated roughly $1.1 million in revenue in 2025, which means its size today is tiny relative to the opportunity. Even modest design wins with large customers could move the revenue line by an outsized amount — a small base against an enormous addressable market.

And Now the Operators Are Buying In

It is not only the chipmaker. The operators racing to build the biggest clusters are paying up for optics, too. In late June 2026, Elon Musk received FTC clearance to acquire Mesh Optical Technologies, a startup founded by former SpaceX $SPCX engineers who built the optical links between Starlink satellites. The target was not a GPU company or a power company — it was an optical interconnect company. When the person racing hardest to build the world’s largest AI clusters decides the missing piece is optical data movement, it tells you where the bottleneck has moved. POET is one of the few listed, investable ways to get exposure to that same shift.

The Risk, Stated Plainly

None of this is a sure thing. $POET is early in commercialization. It has announced partnerships and customer engagements, but its revenue base is still small relative to the opportunity, and it must prove its technology can scale into widespread adoption. It competes against far larger, better-resourced companies — Coherent and the other incumbents NVIDIA chose among them. Investors are not buying a proven leader; they are buying the possibility that $POET becomes one. That is the trade — large potential reward against real execution and competitive risk.

The first wave of the AI trade was about compute. The next wave is about connecting it. NVIDIA is building its own optical fabric, Musk is buying one outright, and POET is betting its whole business on the same handoff — from a fraction of their size.

Sources

1. Nasdaq / The Motley Fool — “Why Everyone’s Talking About POET Technologies Stock Right Now” — https://www.nasdaq.com/articles/why-everyones-talking-about-poet-technologies-stock-right-now

2. NVIDIA Newsroom — Spectrum-X and Quantum-X co-packaged optics photonics switches (GTC 2025, names Coherent et al.) — https://nvidianews.nvidia.com/news/nvidia-spectrum-x-co-packaged-optics-networking-switches-ai-factories

3. U.S. Federal Trade Commission — Early Termination Notice, Transaction 20261601 (Elon Musk / Mesh Optical Technologies) — https://www.ftc.gov/legal-library/browse/early-termination-notices/20261601

4. TechCrunch via The Next Web — “FTC clears Musk to acquire SpaceX alumni startup Mesh Optical Technologies” — https://thenextweb.com/news/ftc-clears-musk-acquire-mesh-optical-spacex-data-centers