Optical networking vs nuclear SMR: two AI-infrastructure plays

AI data centers consume two physical inputs: bandwidth and electricity. Optical networking stocks sell the lasers and transceivers that move AI data; nuclear and SMR stocks supply the firm, carbon-free power that runs it. This page compares the two routes side by side: demand drivers, bottlenecks, time horizons, and risks.

How do optical networking and nuclear SMRs differ as AI infrastructure?

DimensionOptical networking & photonicsAI electricity & nuclear SMRs
What you ownLasers, optical transceivers, fiber, and switch silicon that move AI dataReactors, generators, grid equipment, and nuclear fuel that power AI
Representative tickersCIEN, COHR, LITE, MRVL, GLW, FNSMR (NuScale), OKLO, CEG, VST, GEV, LEU, BWXT
Demand driverCluster bandwidth: the 800G-to-1.6T upgrade cycle and co-packaged opticsData-center electricity, set to roughly double to about 945 TWh by 2030 (IEA)
Revenue todayAlready at scale: Ciena booked $4.77B in FY2025 (+19%), Coherent a record $5.81B (+23%)Generators like Constellation and Vistra earn now; pure SMR developers are largely pre-revenue until about 2030
Valuation anchorCurrent earnings and order books: Coherent earned $3.53 in non-GAAP EPS in FY2025Generators priced on contracted power; SMR developers on licensing progress and offtake options
Where the bottleneck sitsTransceiver supply: 800G output projected 40-60% short of demand through 2027 (McKinsey)Grid hardware: U.S. transformer lead times stretch to about four years
Regulatory gateLittle beyond trade rules; product cadence is set by hyperscaler and Nvidia roadmapsNRC licensing: NuScale's 77 MWe design, approved May 2025, is only the second SMR design cleared
Time horizonProduct-cycle driven; revenue lands with each spec transitionDecade-scale infrastructure; first dedicated SMRs target around 2030
Main riskSpec transitions reshuffle vendor share; hyperscaler customer concentrationLicensing and construction timelines; commercial SMRs are still years away

When does optical networking make sense?

Optical networking suits investors who want revenue that lands now. The AI optical-transceiver market is projected to grow from about $16.5 billion in 2025 to $26 billion in 2026, roughly 57% in a single year (TrendForce), and supply cannot keep up: McKinsey projects 800G transceiver output falling 40-60% short of demand through 2027 (McKinsey). The trade-off is cycle risk, since each spec transition (800G to 1.6T, then co-packaged optics) can reshuffle vendor share.

That demand is already printed in company accounts, not just forecasts. Ciena reported fiscal-year 2025 revenue of $4.77 billion, up 19% year-over-year, with orders exceeding revenue and a growing backlog tied to its expanding role in AI (Ciena Investor Relations). Coherent reported record fiscal-2025 revenue of $5.81 billion, up 23%, and non-GAAP earnings of $3.53 per share (Coherent Corp.). These are businesses an investor can value on current earnings and order books, which is the clearest single difference from the SMR side of this comparison.

The next leg looks structural rather than just cyclical. Silicon photonics, which builds optical functions directly onto chips to cut power per bit, is forecast to grow from about $3.69 billion in 2025 to $33.8 billion by 2034 (Global Market Insights). Whoever leads the 1.6T and co-packaged-optics transitions captures that wave, which is why the spec cycle is the opportunity and the risk at once.

AI creates a new wave in demand for optical transceivers and accelerates LPO/CPO adoption.

— LightCounting

When does nuclear and SMR power make sense?

Nuclear and SMR electricity suits investors with a longer horizon. The IEA projects global data-center electricity use roughly doubling to about 945 TWh by 2030, with U.S. consumption up roughly 130% over 2024 and data-center demand growing about 15% a year, more than four times faster than total electricity demand (IEA). Hyperscalers are contracting for that power directly: Microsoft’s 20-year agreement takes the full roughly 835 MW output of the restarted Three Mile Island reactor, now the Crane Clean Energy Center (Constellation Energy), Google is buying up to roughly 500 MW from Kairos Power reactors with the first unit expected around 2030 (DatacenterDynamics), and Amazon has backed SMR developer X-energy toward more than 5 GW of new capacity by 2039 (Data Center Frontier).

The constraint here is hardware, licensing, and time, not demand. A small modular reactor is a unit of roughly 300 MWe or less, against 1,000 MWe-plus for conventional plants that take a decade to build (World Nuclear Association). The regulatory gate is opening slowly: in May 2025 the NRC approved NuScale’s uprated 77 MWe design, only the second SMR design ever cleared in the U.S., allowing a six-module, 462 MWe plant configuration to be referenced in license applications (U.S. Department of Energy). The fuel chain is being rebuilt in parallel, with Centrus producing the first HALEU from a U.S.-licensed facility in more than 70 years (U.S. Department of Energy). Meanwhile the grid itself is a bottleneck: U.S. transformer lead times have stretched to about four years (pv magazine USA).

Powering industries critical to our nation’s global economic and technological competitiveness, including data centers, requires an abundance of energy that is carbon-free and reliable every hour of every day, and nuclear plants are the only energy sources that can consistently deliver on that promise.

— Joe Dominguez, CEO, Constellation Energy

How do the risks of Optical networking & photonics and AI electricity & nuclear SMRs differ?

The optical side carries product-cycle and customer risk. A handful of hyperscalers drive the order flow, and each generational change in interconnect (800G to 1.6T, linear-drive optics, co-packaged optics) can hand share to a different vendor, so today’s supply shortage is no guarantee about who supplies the next cycle. The nuclear side carries execution and policy risk instead: pure SMR developers are largely pre-revenue, their value hinging on NRC milestones and first-of-a-kind construction, while even willing buyers wait on grid hardware with four-year transformer lead times. The demand sides fail differently too. An AI capex pause would hit optical orders within quarters, whereas contracted power deals such as the 20-year Crane agreement keep paying through a downturn but cap the upside to what was signed.

Which investors are better suited to Optical networking & photonics versus AI electricity & nuclear SMRs?

Optical networking is the faster-cycle play: AI bandwidth demand is visible now and supply-constrained through 2027, but each spec transition reshuffles the winners. Nuclear and SMR power is the longer-duration play: hyperscaler deals lock in demand, but reactors arrive on decade timelines. They are two inputs of the same AI build-out, and many investors pair them.

Related concepts & themes

FAQ

Should I invest in optical networking or nuclear SMR stocks for AI exposure?

They are complements, not substitutes: one sells AI data centers their bandwidth, the other their electricity. Optical demand is already monetizing, with the AI optical-transceiver market projected to grow from about $16.5 billion in 2025 to $26 billion in 2026 (TrendForce). Power demand compounds more slowly but for longer: the IEA projects data-center electricity use roughly doubling to about 945 TWh by 2030 (IEA). Many investors hold both legs and size them by horizon.

How do the risks of Optical networking & photonics and AI electricity & nuclear SMRs differ?

Neither is low-risk, but the risks differ in kind. Optical vendors earn revenue today and McKinsey projects 800G transceiver supply falling 40-60% short of demand through 2027, yet every spec transition (800G to 1.6T, then co-packaged optics) can reshuffle vendor share (McKinsey). On the power side, operating generators such as Constellation already book cash flows, while pure SMR developers remain pre-commercial, with first dedicated units targeted around 2030 (DatacenterDynamics). Picking the riskier end of either side, not the side itself, is what changes the risk profile.

How can Optical networking & photonics and AI electricity & nuclear SMRs be used together in one portfolio?

A common framing is one growth-cyclical sleeve and one infrastructure-duration sleeve. Optical networking behaves like a fast product cycle tied to AI capex, while electricity assets monetize over decades, anchored by hyperscaler offtake deals such as Microsoft-Constellation, Google-Kairos, and Amazon with Talen and X-energy (Constellation Energy). An institutional mandate can treat the power leg as regulated-utility duration and the optics leg as technology beta; an individual investor gets a similar balance by holding both at smaller size.

Do optical networking companies already make money from AI?

Yes, and at scale. Ciena reported fiscal-year 2025 revenue of $4.77 billion, up 19% year-over-year, with orders exceeding revenue and a growing AI-driven backlog (Ciena Investor Relations). Coherent reported record fiscal-2025 revenue of $5.81 billion, up 23%, with non-GAAP EPS of $3.53 (Coherent Corp.). The market behind them is still expanding, with AI optical-transceiver sales projected to grow from about $16.5 billion in 2025 to $26 billion in 2026 (TrendForce). That is the core contrast with SMR developers, whose first dedicated reactors target around 2030.

Where do SMRs stand with regulators, and when is first power expected?

Licensing is moving but early. In May 2025 the NRC approved NuScale's uprated 77 MWe design, only the second SMR design ever cleared in the U.S.; a six-module plant built on it would deliver 462 MWe, and the approved design can now be referenced in combined-license and construction-permit applications (U.S. Department of Energy). An SMR is defined as roughly 300 MWe or less per unit, versus 1,000 MWe-plus for conventional reactors (World Nuclear Association). On timing, Google's deal with Kairos Power targets its first reactor around 2030 (DatacenterDynamics).

How do generators like Constellation differ from SMR developers like NuScale and Oklo?

They sit at opposite ends of the same theme's risk curve. Constellation operates an existing nuclear fleet and books cash flows today, and its Microsoft agreement contracts the full roughly 835 MW of the restarted Crane Clean Energy Center for 20 years (Constellation Energy). SMR developers are largely pre-revenue and depend on NRC licensing and first-of-a-kind construction, with demand anchored by commitments like Amazon's backing of X-energy toward more than 5 GW of new capacity by 2039 (Data Center Frontier). The fuel chain adds a third leg: Centrus produced the first U.S.-licensed HALEU in more than 70 years (U.S. Department of Energy).

Sources & references

  1. Global AI Optical Transceiver Market to Reach US$26 Billion in 2026 · TrendForce, 2026-04-20
  2. Opportunities in networking optics: Boosting supply for data centers · McKinsey & Company, 2025-09-01
  3. Energy demand from AI · Energy and AI · International Energy Agency, 2025-04-10
  4. U.S. transformer market faces severe supply constraints as lead times extend to four years · pv magazine USA, 2026-05-11
  5. Constellation to Launch Crane Clean Energy Center, Restoring Jobs and Carbon-Free Power to The Grid · Constellation Energy, 2024-09-20
  6. Google signs nuclear SMR deal with Kairos for data center power · DatacenterDynamics, 2024-10-15
  7. AI creates a new wave in demand for optical transceivers and accelerates LPO/CPO adoption · LightCounting, 2025-01-15
  8. Silicon Photonics Market Size (2025-2034 forecast) · Global Market Insights, 2025-10-01
  9. Ciena Reports Fiscal Year 2025 Financial Results · Ciena Corporation, 2025-12-11
  10. Coherent Corp. Reports Fourth Quarter and Full Year Fiscal 2025 Results · Coherent Corp., 2025-08-13
  11. NRC Approves NuScale Power's Uprated Small Modular Reactor Design · U.S. Department of Energy, 2025-05-30
  12. Small Modular Reactors (definition and status) · World Nuclear Association, 2025-09-01
  13. Google and Amazon Make Major Inroads with SMRs to Bring Nuclear Energy to Data Centers · Data Center Frontier, 2024-10-17
  14. Centrus Produces Nation's First Amounts of HALEU · U.S. Department of Energy, 2023-11-07