Nvidia Earnings Overview: Networking in Focus
The AI giant continues to climb the wall of worry, and we have the first quarter of networking revenue broken out of results.
Nvidia reported earnings, and as expected, they beat earnings. Did you think that Jensen would fail us now?
On a more serious note, the guide was a bit light regarding buy-side bogeys, but there is almost nothing to complain about this result. Below is the StreetAccount summary, and you can see it’s a clean beat by every metric. They beat revenue by the customary $1.5-2 billion dollars in revenue and continue to have blowout margins.
NVIDIA reports Q1 EPS $6.12 ex-items vs FactSet $5.60
Reports Q1:
Revenue $26.04B vs FactSet $24.59B
Q2 Guidance:
Revenue $28.0B +/- 2% vs FactSet $26.62B
Non-GAAP gross margin 75.5%, +/- 50bp vs consensus 75.6%
Ten-for-one forward stock split effective 7-Jun-24:
The big piece of news is the ten-for-one stock split. While it fundamentally means nothing for the shares, there is probably a benefit from having a nominally lower price. Let’s now discuss what matters this quarter.
Networking Disclosure for the First Time
This is the first time they’ve broken out networking within the data center. ~14% of their revenue in the data center segment consists of networking. Nvidia bought Mellanox when it was at a ~$1.3 annual revenue run rate in 2019, and now, just a little over four years later, it’s almost at a 10x revenue run rate. That’s impressive.
This breakout of networking versus computing will be a key KPI going forward. The ratio of networking revenue to compute will be a new piece of information investors can track quarterly, and I think the perception of their networking moat will be judged by an increase or decrease in this ratio. As they continue to grow ancillary revenue like NVLink and their new Spectrum-X platform, the revenue ratio should go up.
But if you look closer, networking revenue was down quarter over quarter, and the ratio of networking to compute went from 18% of total data center revenue to 14%. As we ramp Blackwell, this should improve, and according to Nvidia, they had a reason why networking was down sequentially. It was all about supply.
Strong networking year-on-year growth was driven by InfiniBand. We experienced a modest sequential decline, which was largely due to the timing of supply, with demand well ahead of what we were able to ship. We expect networking to return to sequential growth in Q2.
This is such a weird answer to me, but apparently, there was not enough availability from players like Fabrinet, as the ramp of 800G was a bit slower than hoped. I’m curious about what this number will look like when GB200 ships because I think there’s a chance for meaningful attach rates.
I want to discuss now the other “big deal” on this call, Spectrum-X.
Spectrum X and Ethernet
I think this is my favorite takeaway from the call because, first, SemiAnalysis pretty much called it, and second, the overreaction in Arista seems like a potential opportunity to me. SemiAnalysis notes that Spectrum-X is a huge deal, as the high Radix switches will likely be a massive player in the Ethernet ecosystem.
They have a strong proof point in the call, as they highlighted a 100k GPU cluster. Gavin Baker points out that it might be the largest cluster in the world, which is super impressive.
In the first quarter, we started shipping our new Spectrum-X Ethernet networking solution optimized for AI from the ground up. It includes our Spectrum-4 switch, BlueField-3 DPU, and new software technologies to overcome the challenges of AI on Ethernet to deliver 1.6x higher networking performance for AI processing compared with traditional Ethernet. Spectrum-X is ramping in volume with multiple customers, including a massive 100,000 GPU cluster. Spectrum-X opens a brand-new market to NVIDIA networking and enables Ethernet-only data centers to accommodate large-scale AI. We expect Spectrum-X to jump to a multibillion-dollar product line within a year.
It’s also no surprise that Nvidia has already considered the competitive threats to Infiniband. I’m sorry to break it to you, Ethernet bulls. The Infiniband killer might just be Nvidia.
Arista took this very negatively because they previously had only guided to ~$750 million in 2025 revenue, while Nvidia thinks Spectrum-X will be a multi-billion dollar product next year. That implies Nvidia is leading RoCE deployments and that many hyperscalers are happy that there’s an ethernet solution with a potential secondary supplier and will still pick the best of breed, aka Nvidia. Nvidia is still leading networking, and it’s up to Broadcom to put any dent in its impressive market share. Good luck.
Speaking of networking, I think the Sovereign cloud opportunity is attractive because it likely implies a solid networking attachment rate for Nvidia. As the world has large customers that are not hyperscalers, each of those customers is more likely to use Nvidia networking than not. Let’s shift the conversation to Sovereign cloud now.
Sovereign Cloud in the High Single Billion this Year
Collette disclosed something new: that sovereign cloud is a “high single billion” business this year.
NVIDIA's ability to offer end-to-end compute to networking technologies, full stack software, AI expertise, and rich ecosystem of partners and customers allows sovereign AI and regional cloud providers to jumpstart their country's AI ambitions. From nothing the previous year, we believe sovereign AI revenue can approach the high single-digit billions this year.
That’s a big new step and an example of the continuing broadening of AI revenue. Hyperscalers are the largest customers here, but even then, they are starting to represent only 40 percent of the business. An example of a significant new player is Tesla, the enterprise customer mentioned in detail this quarter.
Strong sequential Data Center growth was driven by all customer types, led by enterprise and consumer Internet companies. Large cloud providers continue to drive strong growth as they deploy and ramp NVIDIA AI infrastructure at scale and represented the mid-40s as a percentage of our Data Center revenue.
I think Nvidia is firing on all cylinders before producing the essential new product, Blackwell. Collette said the magic words about Blackwell, so things seem to be going well there.
Demand More than Supply (There is No Airgap)
One of the items in the wall of worry was that a heavy revenue shift from Hopper to Blackwell would create an airgap in results. Given that they just guided for sequential revenue growth in Q2, and we know that there will likely be some Blackwell Revenue in Q3, and the revenue crossover likely happens in Q4, I think it’s safe to say that sequential revenue is still on the table.
Importantly, they said demand exceeds supply, which was the wording they used last year before they drove four quarters of sequential revenue growth.
While supply for H100 grew, we are still constrained on H200. At the same time, Blackwell is in full production. We are working to bring up our system and cloud partners for global availability later this year. Demand for H200 and Blackwell is well ahead of supply, and we expect demand may exceed supply well into next year.
What’s more, I think the Blackwell launch will be much larger than Hopper. I believe that the Blackwell crossover in Q4 will be more than enough to lift revenue to sequential new highs.
Blackwell is designed to support data centers universally, from hyperscale to enterprise, training to inference, x86 to Grace CPUs, Ethernet to InfiniBand networking, and air cooling to liquid cooling. Blackwell will be available in over 100 OEM and ODM systems at launch, more than double the number of Hoppers launched and representing every major computer maker in the world.
There will be some concerns about gross margins, as Hopper’s maturity allows ~80% gross margins. Nvidia is guiding to the low 70s in the second half, which is a meaningful step-down, but given supply chain maturity, there is ample room to ramp that margin to much higher numbers next year. All is well in the kingdom of Nvidia.
Yes, we know that the Blackwell TCO is going to create massive cost downs for Hopper, but Neoclouds deploys with long-term contracts. And even with a haircut, it’s pretty easy to make the financial models work for Hopper in the shorter term. If this number even remotely holds, financial capital will deploy in massive amounts even at worse payoffs and even harder at Blackwell. Put differently, we should expect large step-ups when Blackwell comes out, especially if this is what demand looks like during the late Hopper era.
Training and inferencing AI on NVIDIA CUDA is driving meaningful acceleration in cloud rental revenue growth, delivering an immediate and strong return on cloud providers' investment. For every $1 spent on NVIDIA AI infrastructure, cloud providers have an opportunity to earn $5 in GPU instant hosting revenue over 4 years.
It’s an AI gold rush, and the returns on capital are high and easy to underwrite. Build a data center, add power, and rent it out quickly. Yes, Blackwell will change that calculation, but if you have space in your data center, you can deploy that quickly—no need to overcomplicate it with fears of air gaps.
Now, I’m going to turn to another much weirder part of the results: Nvidia is starting to climb the market cap rankings, and in doing so, it’s beginning to have significant and weird outsized impacts on the market. As Nvidia climbs the ranks to become the largest company in the world (a house view here), it will start having an outsized impact on the market.
What it Means to be Megacap
Humor me because this has nothing to do with Nvidia’s results but everything to do with Nvidia's stock. I found the price action on Thursday curious: Nvidia added ~200b+ in market capitalization, and the entire market puked. Maybe it’s just active management chasing the new large-cap weighting, but I found the specific price action odd, and it gave me a lot of food for thought.
I looked for a few explanations here, and one that comes to mind is Momentum1. GE and LLY have to be among the few highest momentum stocks this year, but I still didn’t find this answer satisfactory.
The thing that came out of the woodwork is that there might be some kind of unwind when one specific company does that well, and there is an interesting long-short volatility trade that deals with indexes and stocks that could be an explanation. I don’t know enough to rule this out as the “reason” either.
But the thing I have to ask is, what happens when a large capitalization company takes a lot of share in terms of flows and has excellent results? I would argue that this creates a natural lift from passive buyers. Each time someone buys an ETF for the market, Nvidia will be a more significant part of that purchase in the future. This is possibly one of the reasons for Apple’s continued valuation umbrella. Still, I can’t help but believe that Nvidia racing to the top of the market capitalization leaderboard will help the stock mechanically.
They continue to beat results and show the world what is possible in this new age of AI, but additionally, they will take more of a share of passive voice as they march higher. I continue to believe Nvidia is going to be the largest company in the world, and when as it gets there, incremental passive investing is just another tailwind for shares.
Last, I think the ~40-dollar CY 2025 EPS scenario is in full play. That puts Nvidia's valuation at ~25-27x next year’s earnings. I understand that Nvidia has gone up quite a bit, but in my opinion, Blackwell has many ways to surprise the upside. Despite the rise, Nvidia doesn’t look that expensive, given the context of growth.
That’s all for this week! Thanks for reading, and if you enjoyed this post, please share.
Consider the Momentum peak trade over. I need to be a bit more tactical on updating ya’ll, but that was an excellent short-term call. We are back to all-time highs driven by Momentum. It was a nice little call while it lasted.
Thanks for this. Any idea what the lifespan of a GPU used in a data center is likely to be? Any reason to believe there will be a replacement cycle in 3 - 5 years for GPUs currently being installed?
What's the next wall of worry? :) It looks worries about competition, air pocket, and maybe even ethernet has died off. Has the market realized nvidia is king yet?