UBS Outlook on AI Computing Chain: Strong Orders but Supply Chain Constraints Limit Revenue Realization
Taylor Wilson
A new UBS report says accelerating data-center capex is driving surging orders across the AI compute chain, but lengthening lead times for networking chips, CPUs, and indium phosphide are becoming a shared bottleneck that caps how much revenue the industry can actually book.
Demand is red-hot — so why might revenue still get stuck?
UBS analyst David Vogt notes that global data-center infrastructure spending has accelerated steadily over the past three months, lifting orders and backlogs across the board.
Yet lead times for networking chips — the core processors inside data-center switches that route traffic — CPUs, and indium phosphide (a semiconductor material used to make high-speed optical components) are stretching longer. Companies simply cannot fill all potential demand.
This means → customers have placed orders and budgets are ready, but parts aren't arriving fast enough to ship product — strong orders ≠ booked revenue.
Arista: why might its growth guide jump from 27.5% to 33%?
Vogt expects Arista to raise its FY2026 revenue-growth guidance from 27.5% to 33%, driven by steadily rising order volumes over the past 9–12 months and an acceleration in deferred-revenue recognition.
Arista continues to deepen partnerships with hyperscale cloud providers and emerging cloud platforms; both revenue and profit are expected to beat consensus.
Industry surveys show data-center network-switch demand has strengthened for several months. Despite a Broadcom networking-chip shortage, Arista's quarterly results and forward outlook remain well supported.
Celestica: has the supply-chain drag been fixed?
The supply-chain issues that weighed on server and TPU (tensor processing unit — Google's custom AI chip) revenue last quarter have been resolved.
Demand from Meta and Amazon for Ethernet switches remains strong. Vogt expects this quarter's revenue to come in 3%–4% above consensus, reaching roughly $4.5 billion.
EPS could rise from his prior estimate of $2.29 to $2.40, above the company's guided range of $2.14–$2.34. In plain terms = once the supply bottleneck clears, pent-up revenue gets released all at once.
Extreme Networks and Lumentum: how much upside is left?
Demand for Extreme Networks' wired and wireless gear is strong; revenue is expected near the top of its $335 million guidance. But Vogt flags that the current share price has already priced in a beat-and-raise.
This means → results will likely look good, but the stock may not deliver much more surprise — the good news is already "in the price."
For Lumentum, demand for 200G EMLs (electro-absorption modulated lasers — a core high-speed optical component) remains robust, pushing expected revenue past $1 billion with the September quarter up roughly 120% year-on-year. Although supply limits may cap shipments, operating leverage should spread fixed costs and lift EPS 3%–4% above Vogt's prior estimate.
In this AI capex wave, what is the single variable to watch?
On the order side, the acceleration in AI infrastructure spending is fully visible — demand is not the question for any of these four companies.
The real suspense is on the supply side: whether lead times for networking chips, indium phosphide, and other critical components can ease further before earnings season.
This reflects a deeper signal: the bottleneck in this AI cycle has shifted from "is anyone buying?" to "can you ship fast enough?" For investors, tracking supply-chain dynamics now matters more than tracking order data.
Content is for reference only, not financial advice.