Following Nvidia’s better-than-expected earnings report last Wednesday, UBS analysts were inundated with investor inquiries, primarily focused on the chipmaker’s near-term growth visibility and the durability of its long-term revenue pipeline.
Below, UBS analyst Timothy Arcuri addressed the most frequently asked questions, unpacked key commentary from the earnings call, and provided expanded visibility into a multi-year AI infrastructure boom.
Visibility to data center revenue doubling yet again?
An area where we have received a few investor inbounds, but still seems somewhat overlooked is NVDA’s commentary on its pipeline. The company noted on its FQ1:26 earnings call that it has visibility into “tens of gigawatts” of AI infrastructure projects in the “not too distant future”. Assuming a “low case” pipeline of 20GW and NVDA’s stated range of ~$40-50B per GW, this puts its total revenue opportunity for this pipeline at a minium of ~$1T. While the company did not specify a timeframe for this pipeline, based on our conversations, we believe these projects are likely to be rolled out over a 2-3 year period. Using the average of this timeframe, this suggests the company may effectively have “visibility” to ~$400B/yr in data center revenue, or about 2x our $233B data center revenue estimate for C2026. This is obviously very heady, but we did note in UBS’ deep dive on OpenAI’s Abilene AI Factory that Crusoe alone has ~20GW in project pipeline and this is just one digital infrastructure project developer. The upshot of this is that we believe investor concerns around growth sustainability should be allayed by some of this commentary from the earnings call.
The math around the rack numbers given on the call
One of the main investor questions we got coming out of the call was what NVDA was trying to imply by the GB200 rack numbers it provided on the call – which were so far above conventional wisdom that it spurred some confusion. The company said “on average major hyperscalers are each deploying nearly 1,000 NVL72 racks or 72,000 Blackwell GPUs per week and are on track to further ramp output this quarter”. Taken at face value, this implies a GPU run rate of nearly 1MM/Q for each hyperscaler – so far above most consensus estimates that it was hard to foot. Therefore, we believe the company was not trying to communicate a revenue “run-rate” but simply trying to reassure investors that GB200 rack issues are resolved and a large quantity of racks are moving from the ODMs and OEMs now to customers – consistent with our commentary into the call that investor concerns about supply chain inventory were overblown. We would not try to do anything more with these numbers as we think the company meant this to be more illustrative than quantitative.
NVLink bolsters growth in Networking.
Networking revenue grew to ~$5B in FQ1:26 (+64% Q/Q), $1B of which NVDA attributed to NVLink revenue which was up substantially Q/Q. We believe this is almost entirely tied to the ramp in shipments of GB200 NVL72 rack scale systems, each of which includes a 72-GPU NVLink domain (vs an up to 8-GPU domain for HGX systems). NVDA is recognizing NVLink revenue separately for these NVL72 systems, which is/was not the case for HGX boards where revenue has been consolidated into the Compute sub-segment of Data Center. As such, we would expect Networking revenue to track more closely to NVL72 rack shipments going forward, albeit maybe with a little bit of a lag.
Gaming growth driven by… gamers.
The sharp improvement in Gaming revenue in FQ1 (up nearly 50% Q/Q and well ahead of expectations) has prompted many investors to question whether there was some component of 50-series RTX cards being pulled into China for AI workload purposes. Though this cannot be completely discounted, we suspect any such pull-in was likely very limited due to: 1) availability of Blackwell-based RTX GPUs being still too limited in the gaming channel to enable larger-scale deployment, 2) RTX 50-series GPUs are PCIe based and do not support NVLink for scale-up, and 3) NVDA had to some degree starved the gaming channel for Blackwell out of the gate as it prioritized capacity for data center applications so the FQ1 (April) growth was driven by back-filling the channel following these severe supply shortages.
Gross margin drivers for 2H.
General improvement in Blackwell profitability and cost downs remain the primary driver to get margins back to the mid-70%s target by FYE26. Part of this, we believe, is GB300 – for which NVDA may actually recognise a small amount of revenue inside of FQ2 with the real ramp being FQ3. Longer-term, we believe pricing to value remains the key function for NVDA’s margins – this comes down to both hardware and the software overlay of which the release of Dynamo at GTC is a prime example (accelerates inference on NVDA hardware by >30x).
Taken together, Nvidia’s earnings, along with expanding visibility into a multi-year AI infrastructure boom via UBS, suggest that this boom is less about product cycles and more about exponential infrastructure scaling.
Last week, UBS analysts Steven Fisher, Amit Mehrotra, and others noted that the construction boom of AI data centers is not expected to show up in the real economy or provide structural tailwinds until the second quarter of 2026.
“More slowing before reacceleration in 2026,” Fisher wrote in a note, adding, “We expect stimulus and structural forces to drive the rebound, while cyclical factors remain weak.”
Another UBS note outlined the bullish outlook on data center-driven power demand, particularly for natural gas-linked utilities and midstream names.
The AI data center buildout could be viewed as the digital-age cousin of the 1930s “New Deal”—but instead of highways and dams, it’s GPUs and megawatts. It’s reshaping American infrastructure but with Big Tech at the helm.
Tyler Durden
Wed, 06/04/2025 – 06:55
