Gary Smith, who has led optical networking vendor Ciena since May 2001, told The Wall Street Journal that the current AI infrastructure build-out should be judged on its own terms. Ciena went public in the late 1990s at a $3.4bn valuation, later passed $40bn, then lost more than 90% of its value in the bust and is now worth about $45.75bn. Smith said losing bets will appear in the AI era too, yet the differences from the fibre boom outweigh the similarities for businesses planning capacity.
Why Smith sees a different infrastructure cycle
Smith, 66, described his stance as shaped by direct experience of the earlier collapse, telling the Journal he remains suitably paranoid about investment cycles. Ciena survived that period under his leadership, while much of the industry did not recover its peak valuations for years. The Journal interview was conducted by Steven Rosenbush and focused on optical networking demand linked to AI workloads. For Smith, that history gives weight to his distinction between speculative fibre laying and present orders.
Ciena has built 3-nanometre versions of its WaveLogic chips that move 1.6 terabits per second over a single strand of fibre, with a 2-nanometre version in development. Smith described AI demand as a sequence of waves, with data-centre interconnect still ramping up and inference just starting. AI agents come next, followed by robotics, in his account. He argued that people have grossly underestimated the infrastructure required across those stages. The product detail matters because throughput per fibre determines how much traffic existing routes can absorb.
The contrast with the late-1990s boom rests on utilization and buyers, according to the accounts cited by the Journal. About 94% of the fibre laid during the telecom boom went unused, Citi analyst Heath Terry told the Journal in an earlier interview. Smith pointed to a different customer base today that includes highly profitable cloud giants. He said Ciena has not been able to keep up with demand in the present cycle. That shift from speculative construction to constrained supply frames the debate over whether spending is excessive.
What AI capacity orders mean for business
Ciena expects compound annual growth of 30% from fiscal 2026 to 2029 and expects to end fiscal 2026 with an order backlog of about $10bn, according to the Journal. For companies deploying AI, a backlog of that size points to longer lead times for high-capacity links between data centres. Large cloud buyers can secure slots earlier, while smaller firms may face queues or need to plan interconnect purchases further ahead. Procurement timelines thus become part of AI project planning, not an afterthought.
Other assessments cited alongside the interview urge caution before treating all AI spending as productive. Apollo economist Torsten Slok said on Monday that US companies outside tech now treat AI spending as a cost. Gartner has warned of AI infrastructure write-downs, while Nvidia has put $170bn behind neoclouds. The practical lesson is to separate contracted, utilized capacity from speculative bets on future workloads. Buyers should ask vendors about utilization rates, delivery dates, upgrade paths for WaveLogic-class optics, and cancellation terms.
The marker to watch is whether Ciena closes fiscal 2026 with an order backlog near $10bn while sustaining progress toward its 30% growth path through 2029. Delivery against those figures would support Smith's wave account, from interconnect through inference to agents. A shortfall, or rising write-downs of the kind Gartner discusses, would point toward overbuilding. The next fiscal reports will show which account fits the facts.
