Ciena says it cannot make optical networking gear fast enough for AI. In a Wall Street Journal interview with CEO Gary Smith, a veteran of the 2001 telecom crash, the company describes demand from AI outpacing supply of its optical networking parts, and the executive describes himself, per the subhead, as "suitably paranoid" while still optimistic. The interview is paywalled, and we could read only the Techmeme listing summarizing it. So this post does not claim to report the interview in full. Instead it checks the claim against what Ciena has put on the record: its fiscal Q3 2026 earnings release filed with the SEC and earlier interviews with Smith.
Why a networking company belongs on an AI blog: every large training run and inference fleet needs bandwidth between racks, buildings and regions, and the parts that supply it are in short supply. We have followed the same pattern across Lumentum's optical parts being sold out into 2029, TSMC's record quarter and the wider AI chip supply chain.
TL;DR: claimed vs confirmed
| Claim | Status |
|---|---|
| AI demand is outpacing Ciena's supply | CEO's statement as summarized from the WSJ; the filing lists supply constraints as a risk, and the CFO said in March that Q1 revenue "would have been higher" without them |
| Record quarter | Confirmed in the press release: $1.67B revenue, up 37% year on year |
| Adjusted EPS $2.11 | Confirmed (GAAP diluted EPS $1.83) |
| Guidance raised | Confirmed: fiscal 2026 revenue to $6.42B, plus or minus $50M, about 35% growth at the midpoint |
| Backlog about $8.5B, above $10B by year end | Reported from earnings-call coverage, not in the press release |
| Cloud provider revenue up more than 80% | Reported from call coverage, not in the press release |
| Two customers are 41.7% of revenue | Confirmed in the release; customers not named |
| The AI boom resembles 2001 | Smith's own caution, per the interview subhead; a judgment, not a finding |
What did Ciena actually report?
For the quarter ended August 1, 2026, announced September 3, Ciena reported revenue of $1.67 billion against $1.22 billion a year earlier. Adjusted earnings per share were $2.11, up from $0.67. For the fourth quarter it guided to $1.75 billion, plus or minus $50 million, with adjusted gross margin of 45% and adjusted operating margin of 20%. Full-year fiscal 2026 revenue guidance rose to $6.42 billion, plus or minus $50 million.
Smith's line in the release is that AI "continues to drive compounding waves of network investment." CFO Marc Graff pointed to "expanding supply capacity" as a factor behind future earnings acceleration. That phrase is the filed version of the supply story: capacity is being expanded because it is currently the limiter.
Linked supply chain illustration for the optical components behind AI network capacity
The backlog numbers come from call transcripts and summaries, not the press release. Coverage of the September 3 call reports a backlog of about $8.5 billion, up roughly $800 million in the quarter, with management expecting to exit fiscal 2026 above $10 billion, and says Ciena is signing multi-year supplier agreements, some with firm purchase commitments. Those details are consistent with a supply-limited business, but check the transcript or the 10-Q before relying on any single number. One more reported detail: shares fell about 9% in premarket trading on the results day despite the beat, which says investors were looking at expectations and not at the result.
What has Smith said before?
A solid wave line continuing as a dotted wave, illustrating Ciena's forecast that AI demand will keep doubling the optical market
In a March interview with Light Reading, after a quarter in which revenue rose 33.1% to $1.43 billion, Smith said he expects the optical market to double in the next few years, from roughly $12 billion to $13 billion "since the last sort of decade." Three drivers stood out:
- Opticalization of the data center. Smith argued that "direct connect" optical pathways inside data centers are coming because of physics constraints at the speed and scale of GPU clusters: "It's really a matter of not if, it's when." He said Ciena had not yet booked revenue there.
- Linked training clusters. Because of space and energy limits, GPUs are being clustered across sites. "That's just beginning," he said, with hundreds of millions of dollars of orders rolling out at that point.
- Service providers catching up. Ciena sees about five years of underinvestment in optical networks, plus demand for managed optical fiber networks for hyperscalers outside the US.
He also put numbers on the buyers: all four hyperscalers planning a step-function increase in capital spending to a combined $600 billion this year, driven by AI training and inference. In the same interview he said "network is the new power", a compact way of saying bandwidth is now as scarce as electricity. The CFO's remark that first-quarter revenue would have been higher but for supply constraints is the clearest earlier evidence that the shortage is real and not just a talking point.
Where optical gear sits in an AI cluster
It helps to separate three distances. Inside a rack, GPUs talk over short copper links and, increasingly, over optics. Between racks and buildings in one campus, data moves over optical transceivers and switches. Between sites, whether a few kilometers or a few hundred, it moves over long-haul and metro optical systems, which is where Ciena's core products sit. Smith's two growth arguments map onto the outer two: "opticalization" pushes optics deeper into the data center, and linked training clusters push more traffic across sites.
That is why a shortage matters even if you never buy a network switch. A training run that spans two regional sites needs that link to be fast and reliable, or the GPUs at one end idle while they wait. Operators adding capacity for power or space reasons, who then have to stitch sites together, create new orders for exactly the equipment that is already backlogged. The same logic applies to the laser and transceiver makers upstream, which our Lumentum coverage describes as sold out for years.
A short timeline of the signals
- March 2026. Ciena reports Q1 revenue of $1.43 billion, up 33.1%. Smith says the optical market could double, and the CFO says supply constraints held revenue back.
- September 3, 2026. Ciena reports $1.67 billion, up 37%, raises fiscal 2026 guidance to $6.42 billion and talks of expanding supply capacity.
- October 2026. Lumentum says AI optical parts are sold out into 2029, and TSMC reports record results driven by AI demand.
- October 24, 2026. The WSJ interview puts the CEO's own caution on the record.
Read in order, the pattern is consistent: demand statements, then capacity expansion, then supplier commitments. What no source here establishes is when supply catches up. The reported expectation from call coverage is a balance late in the decade, which we could not confirm in a primary document.
Why the 2001 comparison matters
Smith's career includes the telecom crash, when carriers overbuilt fiber and optical vendors saw orders vanish. His public stance, as summarized, is to stay paranoid while the order book is full. That is a useful signal for anyone reading AI infrastructure news, and it is worth separating from the numbers. Three facts in the filings keep the bubble question open:
- Customer concentration. Two customers together accounted for 41.7% of revenue, unnamed in the release.
- Cloud dependence. Call coverage reports cloud-provider revenue up more than 80%, which ties Ciena's growth to hyperscaler budgets.
- Supply commitments. Firm purchase commitments to suppliers protect capacity but create obligations if demand turns.
None of this says demand is fading. It says the business is large, concentrated and priced for growth, which is why the CEO's caution is a story in itself.
What it means for builders
Compute supply is a stack of bottlenecks, and each one sets the pace for the ones above it. Chips and memory get most attention, for example in our coverage of the memory shortage Michael Dell expects to last years and Micron's HBM exposure. Optical links, transceivers and the lasers behind them are another. For readers who build or buy AI capacity:
- Expect lead times, not just prices. If network gear is backlogged into 2028, new regions and clusters arrive on the vendors' schedule.
- Multi-site training is a real design choice. Linking regional GPU sites is already producing orders, so architecture decisions that tolerate cross-site bandwidth are not hypothetical.
- Watch spending signals. Hyperscaler capex plans, the $600 billion combined figure Smith cited, and supplier backlogs move together; a change in one leads the others.
- Read vendor claims against filings. Backlog and "demand exceeds supply" are claims; revenue, guidance and risk factors in the SEC documents are the verifiable layer.
What we could not verify
We did not read the full WSJ interview, so we do not attribute any specific quote from it beyond the subhead on the Techmeme listing. We have not independently confirmed the backlog or cloud-growth figures; they come from third-party coverage of the call. And we have no statement on when supply and demand will balance. If you need those, read the transcript and the 10-Q directly.
Figures are from Ciena's September 3, 2026 release, Light Reading's March 6, 2026 interview and call coverage, as of October 24, 2026.