By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panel discusses the shift towards optical interconnects in AI infrastructure, with bullish views on Broadcom, Marvell, and niche players like POET. However, they also highlight risks such as competition from cheaper electrical alternatives, the maturation of silicon photonics, and potential internalization of optical solutions by hyperscalers.
Risk: Hyperscalers internalizing optics and moving to in-package, custom stacks faster than expected, compressing third-party transceiver margins and demand.
Opportunity: The secular trend of increasing demand for faster, lower-power data movement, potentially lifting vendors like Broadcom, Lumentum, Coherent, and Marvell.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Most investors think the artificial intelligence revolution is about building smarter chips. And for good reason. Companies that provide the computing power behind AI, most notably Nvidia, have been some of the market's biggest winners in recent years.
But what if the next major AI opportunity isn't about processing information? What if it's about moving information?
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
As AI clusters grow larger and more powerful, moving data between processors is becoming almost as important as processing the data itself. And that's why a growing number of investors are paying attention to a little-known corner of the technology industry: optical networking.
Training today's most advanced AI models requires a huge number of processors working together simultaneously. While these processors perform enormous amounts of computation, they also spend significant time exchanging information. As such, every processor must constantly communicate with other processors across the network.
The problem is that as AI clusters become larger, the communication challenge grows rapidly.
For a long time, the technology industry focused primarily on building faster graphics processing units (GPUs). But many experts now believe that communication between processors is becoming one of the most important constraints on AI performance.
In other words, the bottleneck could shift from just making processors run faster to helping them communicate faster.
Traditionally, information inside data centers has been transmitted through electrical connections.
That approach has worked remarkably well for decades. However, as data volumes continue to rise, electrical systems face increasing challenges related to power consumption, heat generation, and bandwidth limitations.
That's one reason the industry is investing heavily in optical networking. Instead of transmitting information through electrical signals, optical systems use light.
The advantages can be significant. Optical technologies can move large amounts of data quickly while consuming less power and generating less heat than traditional electrical connections.
For AI companies building increasingly massive data centers, those benefits matter a lot. The result is growing interest in photonics and optical networking technologies that could become critical components of next-generation AI infrastructure.
Several companies are positioning themselves to benefit from this trend.
Large infrastructure providers such as Broadcom (NASDAQ: AVGO) and Marvell Technology (NASDAQ: MRVL) already play important roles in networking and connectivity. As AI spending rises, demand for their products could grow alongside it.
Investors are also paying attention to specialized optical companies such as Lumentum Holdings (NASDAQ: LITE) and Coherent (NYSE: COHR), which provide technologies and products used in high-speed optical communications.
Then there are smaller companies such as POET Technologies (NASDAQ: POET). Rather than building AI chips, POET is developing optical technologies to more efficiently move information through AI systems.
While these businesses take different approaches, they are all pursuing the same opportunity: helping solve AI's growing communication challenge.
The investment thesis is straightforward: If AI spending continues to grow, the infrastructure needed to connect AI systems will likely grow as well.
Every new AI cluster requires processors, networking equipment, switches, optical modules, lasers, cables, and other supporting technologies. The companies that provide these components could benefit from continued investment over the years.
At the same time, investors should remember that identifying a trend is often easier than identifying the eventual winners. Not every company participating in the optical networking market will succeed. Some technologies will gain widespread adoption, while others may never achieve meaningful commercial scale.
This makes optical networking an exciting opportunity but also a highly uncertain one.
The first phase of AI investing rewarded companies that helped machines think. The next phase may reward companies that help those machines communicate.
As AI systems become larger and more complex, the ability to move information quickly and efficiently is becoming increasingly important. For investors searching for the next chapter of AI infrastructure spending, this may be one trend worth watching closely.
Before you buy stock in Broadcom, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Broadcom wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you’d have $398,052! Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $1,181,688!
Now, it’s worth noting Stock Advisor’s total average return is 892% — a market-crushing outperformance compared to 205% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
**Stock Advisor returns as of June 28, 2026. *
Lawrence Nga has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Broadcom, Coherent, Lumentum, Marvell Technology, and Nvidia. The Motley Fool has a disclosure policy.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Four leading AI models discuss this article
"Optical interconnects will become a structural, not episodic, driver of AI infrastructure spending, potentially re-rating optics and networking names as data movement becomes the new bottleneck."
The piece spotlights a plausible pivot in AI capex from chips to connective tissue—data-center interconnect, photonics, and optical modules—as clusters scale. If demand for faster, lower‑power data movement persists, optics could become a larger share of AI infra budgets, potentially lifting vendors like Broadcom, Lumentum, Coherent, and Marvell alongside niche players such as POET. The thesis rests on a secular trend, not a short-term spike, but it’s still a cyclical bet: hyperscale spend, supplier capacity, and commodity silicon pricing matter. Watch competition from cheaper electrical alternatives, the maturation of silicon photonics, and the ROI rhythm for cloud operators amid macro headwinds and potential AI efficiency gains that reduce traffic growth over time.
But the emphasis on optics may be overstated; AI efficiency improvements and smarter mesh networking could damp traffic growth, capping optical capex. If hyperscalers optimize layouts, reuse hardware, or push software-defined networks, the addressable market may be smaller or more cyclical than the article implies.
"Optical interconnects are the only viable path to scaling AI training clusters, shifting the value capture from pure compute to high-speed data transport infrastructure."
The shift toward optical interconnects is not just a trend; it is a physical necessity for scaling GPU clusters beyond current limits. As we move toward 'rack-scale' AI architecture, copper interconnects hit a wall due to signal attenuation and power density. Companies like Broadcom (AVGO) and Marvell (MRVL) are effectively 'picks and shovels' plays, but the real alpha lies in silicon photonics integration. If we can move the laser source closer to the compute die, we reduce latency and power consumption by orders of magnitude. However, investors must distinguish between companies with proprietary, scalable manufacturing processes and those merely selling commodity optical transceivers, where margins are destined to compress as competition intensifies.
The industry may pivot to proprietary 'in-package' optical solutions controlled entirely by hyperscalers like Nvidia or Google, effectively commoditizing third-party component suppliers and crushing their pricing power.
"Optical interconnects solve a real problem, but the article overstates how much *new* capex this unlocks versus how much is already embedded in hyperscaler infrastructure roadmaps."
The article correctly identifies that intra-cluster communication is becoming a real constraint—latency and bandwidth between GPUs matter as much as compute now. But it conflates 'a real problem' with 'a massive TAM opportunity.' Optical interconnects are already baked into modern data center designs (Nvidia's NVLink, InfiniBand, etc.). The real question: does shifting from copper to photonics in *existing* architectures drive incremental capex, or is this already priced into networking budgets? POET, LITE, and COHR are speculative bets on adoption acceleration, not guaranteed beneficiaries. The article also ignores that hyperscalers (Meta, Google, OpenAI) often design custom silicon and may internalize optical solutions rather than buy from vendors.
Optical networking adoption in AI data centers is already well underway and largely reflected in current valuations of AVGO, MRVL, LITE, and COHR. The article presents this as an emerging trend when it's already a known factor priced into consensus.
"Adoption of new optical interconnects will be slower and more concentrated among Broadcom and Cisco than the article implies, limiting upside for pure-play photonics names."
The article correctly flags interconnect bandwidth and power as scaling constraints in large AI clusters, yet it downplays how quickly electrical SerDes and co-packaged optics from Nvidia, Broadcom, and TSMC are extending copper reach. Most near-term AI builds still use pluggable 800G/1.6T modules supplied by incumbents rather than novel photonic ICs. Smaller names such as POET and Lumentum face multi-year qualification cycles and face margin pressure if hyperscalers consolidate vendors. The Motley Fool disclosure reveals positions in several names it promotes while steering readers away from Broadcom, suggesting promotional intent over balanced timing analysis.
If hyperscale capex stays elevated through 2026 and pluggable optics hit thermal walls faster than expected, even modest share gains by LITE or COHR could justify current multiples.
"Hyperscalers internalizing optics could compress external transceiver demand and margins faster than anticipated."
Specifically, Gemini’s focus on external silicon photonics as the alpha misses a bigger risk: hyperscalers could internalize optics and move to in-package, custom stacks faster than the market expects. That would compress third-party transceiver margins and demand well before the 2025–2027 capex cycles assume, even if PoET/LITE/COHR win some design wins. External suppliers might ride a shorter, choppier cycle rather than a long, secular uplift.
"The transition to co-packaged optics will likely commoditize existing transceiver vendors as hyperscalers prioritize vertical integration to solve thermal constraints."
Claude is right that this is largely priced in, but we are missing the 'thermal wall' catalyst. Grok notes pluggables are the current standard, but if thermal density forces a shift to co-packaged optics (CPO) sooner than 2026, the entire supply chain shifts from a commodity model to a proprietary one. This isn't just about bandwidth; it's about power efficiency. If CPO becomes the standard, the incumbents' current margins will be cannibalized by custom, vertically integrated hyperscaler solutions.
"CPO adoption timing, not inevitability, determines whether pluggable optics vendors get a 2–3 year margin window or face immediate margin compression."
Gemini's thermal-wall-to-CPO pivot is the real inflection, but it's being treated as inevitable. The timeline matters enormously: if hyperscalers delay CPO adoption by 18 months due to qualification costs or yield issues, pluggable suppliers (LITE, COHR) get a longer runway than the 'cannibalization by 2026' thesis implies. Nobody's quantified the actual thermal density threshold that forces the switch. That's the number to watch.
"Internalization plus slow qualification could truncate third-party CPO upside even if thermal limits hit sooner than expected."
Gemini's push for a thermal-wall CPO inflection by 2026 underplays the qualification and yield hurdles Claude flags. Even if density thresholds arrive, hyperscalers could absorb the shift internally via custom silicon rather than outsourcing to LITE or COHR, shortening the external revenue window. That compresses margins for pluggable incumbents faster than any power-efficiency catalyst expands them.
The panel discusses the shift towards optical interconnects in AI infrastructure, with bullish views on Broadcom, Marvell, and niche players like POET. However, they also highlight risks such as competition from cheaper electrical alternatives, the maturation of silicon photonics, and potential internalization of optical solutions by hyperscalers.
The secular trend of increasing demand for faster, lower-power data movement, potentially lifting vendors like Broadcom, Lumentum, Coherent, and Marvell.
Hyperscalers internalizing optics and moving to in-package, custom stacks faster than expected, compressing third-party transceiver margins and demand.