AINEWS
Nvidia paired Coherent and Lumentum optics deals with 70% fiscal 2028 growth guidance in one quarter. The read: the next AI bottleneck is the fiber between chips, not the chips themselves.
Nvidia's two optics partnership announcements landed the same week as management's 70% fiscal-2028 growth guidance, a number that implies roughly $100 billion more revenue than Wall Street had penciled in. The income statement carried the headline numbers in Nvidia's 26-Aug-2026 Q2 earnings cycle. The structural read is in the two press releases that bracketed it: a strategic partnership with Coherent and a parallel deal with Lumentum, both aimed at the optical networking hardware that links thousands of AI accelerators into a single training cluster. They point to a shift in where the AI supply chain breaks next: from silicon to the fiber between silicon.
The shift follows a playbook Nvidia has already run. In 2023 and 2024, when AI training demand outran foundry capacity, the company did not wait for the merchant supply curve to catch up. It locked in long-dated capacity at TSMC, paid upfront, and absorbed the working-capital hit to keep its product road map on schedule. The optics deals look like the same move, one layer deeper in the stack. Coherent and Lumentum are not chipmakers. They build the lasers, modulators, and transceivers that move data between GPUs across a data-center hall, the kind of hardware that becomes scarce the moment a customer adds a few thousand more accelerators to a single training cluster.
The demand side of the picture is in the Q2 transcript. Management guided to roughly 70% revenue growth in fiscal 2028, the year ending in late January 2028, a number that implies closer to $680 billion in revenue versus a Wall Street consensus near $580 billion, according to a secondary read of the same call. The roughly $100 billion delta versus consensus is forward-looking guidance, not delivered revenue. The forecast says the buyers are lined up, the build plans are signed, and the supply chain has to clear a higher bar in eighteen months than it does today.
That is where optics enter. Optical interconnect is the part of an AI data center that converts electrical signals on a chip into laser pulses on a fiber, then back again. At today's scale, an AI training cluster can pull together tens of thousands of accelerators, and the bandwidth between them starts to dictate how fast a model can be trained. As per-accelerator compute has climbed, the gap between chip-to-chip bandwidth and compute throughput has widened, and the only hardware that closes it at data-center scale is photonics. The Coherent release names "next-generation data center architecture" as the target, and the Lumentum release frames its work as "state-of-the-art optics technology" for the same customer. Both companies committed to multi-year development programs with manufacturing scale-up as the explicit deliverable.
Independent analysis lines up with the primary releases. Analysts at Futurum Group frame Nvidia's combined optics commitments as a roughly $4 billion bet, a number the firm attributes to industry reads of supplier announcements rather than a single Nvidia line item. The figure is a secondary corroboration of the mechanism, not a verified company disclosure. It is the right order of magnitude, and it lands in the same week as the two press releases. The signal is in the direction of travel, not the exact dollar.
Most coverage will frame the 70% target and the consensus gap as the news. The narrower signal is in the supply stack. In 2023, the constraint on AI scaling was foundry output. Nvidia responded by buying its way up the supply chain. In 2026, the constraint is shifting to the optical layer that turns a room full of GPUs into a single computer. Nvidia is responding the same way, two quarters apart and one layer down. The pattern, not the dollar figure, is what carries forward.
The portable test for the next round of AI capex reporting: when a vertically integrated platform company signs capacity deals one layer down from its own product, treat the deal as evidence that the layer above is about to become the bottleneck. The GPU shortage of 2023 was visible months in advance in TSMC's booking reports. The optics chokepoint ahead will be visible the same way, in the order books of Coherent and Lumentum and the supplier reports that follow. Read the vertical-integration pattern, and the next supply-chain break becomes legible before it makes the headlines.