Other Recent Articles to Read:

On the hyperscaler side of things, AWS and Nvidia announced a new 2M GPU contract that also includes Vera CPUs and ongoing availability of Nvidia models through Amazon Bedrock. Amazon is also adopting Nvidia’s physical AI suite for robotics-based automation across its logistics network as the two companies get closer. For the ACIE segment, neocloud growth was the biggest contributor. As a reminder, they think that this segment’s growth story is even better than the hyperscaler side, as these smaller customers are less likely to build their own chips or hardware like we’ve seen its mega-cap customers do.

Reporting Change:

Nvidia reorganized its reporting segments into Data Center and Edge Computing. They’re doing this as “AI” becomes an increasingly vague term to describe all of its accelerated compute business. The change allows the company to better communicate its growth drivers and more cleanly segment pieces by use case.

Within data center, they’ll now report a “hyperscale” sub-market that includes the big public cloud vendors and massive consumer internet companies like Meta. They’ll also begin reporting an “AI Clouds, Industrial & Enterprise” (ACIE) sub-market that includes sovereign AI, neoclouds and everything else. Each is about 50% of data center revenue. Edge Computing is all agentic and physical AI-powered devices, which encompasses robotics, automotive, gaming and personal computers. This change signals Nvidia believes that edge processing and inference are ready to become much larger businesses, with the company poised to benefit.

Nvidia also guided to 70% revenue growth for next fiscal year (starting in two quarters). Demand would enable 100% Y/Y growth excluding capacity constraints, but they’re still dealing with those and will be through the end of next fiscal year (18 months). That compares to 45% consensus growth expectations, shattering the consensus view. It’s incredible that they’re beating mean estimates by that wide of a margin after so many quarters of consistently doing so and considering how many analysts religiously track this name.

The company trades for 21x forward EPS. EPS is expected to grow by 90% this year and by 45% next year.

Data Center – Chips:

Starting on the GPU side, Nvidia’s Blackwell platform continues to lead all MLPerf and AgentPerf rankings across the industry. And perhaps unsurprisingly, the newest Rubin GPU is fully expected to top leaderboards as well. This is the main character in the new Vera Rubin AI compute platform, which entered full production on schedule and is enjoying strong demand throughout all client cohorts – hyperscalers… neo-clouds… governments… large enterprises… everyone. 

While GPUs get most of the attention and are undoubtedly the star of the show, there is a large supporting cast of hardware components that are vital in making Nvidia’s performance best-in-class and its unified platform all the more sticky. The mega-cap is also positioning itself to take significant market share on the CPU side, both in terms of direct sales and (newly) inclusion of these chips in Vera Rubin racks (Vera is the CPU). This is what will help them grow from a $5B CPU business to $20B in the coming years. Their confidence is understandable, considering leadership thinks this CPU is 80% faster than the next best substitute.

As we’ve been talking about for a couple quarters, AI directly supports demand for CPUs. That happens in two distinct ways. First, while GPUs are wildly impressive when it comes to overall potential and capacity to do work, they need guidance. CPUs provide the orchestration of this important work to make sure it’s organized rather than chaotic. And furthermore, agents use traditional software tools to conduct work, and those tools are run on a lot of CPU-based, again directly augmenting demand. SpaceX, Oracle and AWS were announced this quarter as a new customer and, generally speaking, “broad adoption” should follow.

But wait, there’s more. Nvidia’s Data Processing Units (DPUs) are yet another important chip component in its server racks. This manages traffic, data processing and other cumbersome tasks so CPUs can focus on facilitating AI work and GPUs can get that work done. Nvidia’s new DPU (Vera BlueField 4 STX) is the product to know here.

Lastly, as Nvidia sees all hyperscalers and frontier lab companies make progress in building their own AI chips, the company wants a piece of that market as well. Their new custom accelerator (Groq 3 LPX) mimics the benefits of other custom chips by focusing on a smaller subsection of AI workloads to specialize in. Shedding the need to cater to general purpose use cases means this can perform much better in certain inference and other machine learning-related tasks, much like we’ve seen from Alphabet and others on the market. These are becoming an increasingly popular data center option, and it’s good to see Nvidia meaningfully participating with a product they think offers 4x token throughput vs. the next best option.

Data Center – More:

Its new Spectrum 6 Switches are (shockingly) set to deliver industry-leading performance and are being deployed in data centers as we speak. This should help the company build on its global networking revenue lead. Amazing to think considering it entered the space just 6 years ago with its Mellanox purchase. Between this proliferation, its custom chips, CPUs and DPUs, Nvidia has raised its revenue opportunity per gigawatt from $18B to $40B in just two generations.

Next, Nvidia just launched its Design, Simulation and Operations (DSX) platform called Nvidia DSX. It treats data center operations like end-to-end platforms, with all components, software tools and support provided in one place. It gives structured guidance on how to optimally run an Nvidia-powered data center, while also providing significant room for customization, as (again) needs across clients are far from monotonous. This customization is done with direct support from Nvidia and its army of developers to help make more projects successful.

AI Industry Maturation:

There were 2 welcome signs of improving industry health this quarter. First, Nvidia enjoyed a significant broadening in demand from more customers. As discussed on the call, we are 1 year removed from OpenAI essentially driving their entire AI growth engine, with hyperscalers representing the rest. Now, many more labs, governments and large companies are contributing to its top-line momentum. This includes both in the USA and everywhere else and diminishes some of the customer concentration risk tied to OpenAI. Certainly not all… but some. Evidence of this encouraging trend can be seen in ACIE data center growth comfortably outpacing its hyperscale segment. 

Second, as covered recently, Nvidia signed a $500B financing agreement with several large institutions. This effectively creates financial markets that provide much better liquidity from a lot more sources other than Nvidia’s balance sheet. It frees Nvidia from needing to fund so much of the buildout and so many supply chain partners on its own. That naturally reduces balance sheet risk. It’s great to see the partners stepping in to create these new markets, which should embolden more overall investment and demand for AI hardware.

Sovereign AI:

Nvidia views neoclouds as highly strategic partners for sovereign AI proliferation. As they explained it, regulatory red tape slows down the deployment process across foreign governments in ways that more local neocloud vendors don’t deal with. Partnering with these companies expedites time to market and gives Nvidia a better chance of dominant  hardware presence in global AI factories. This is why they’ve been so determined to help companies like CoreWeave and Nebius grow with equity investments and some friendly business arrangements too. Those arrangements include revenue sharing structures allowing these cash scarce companies to tap into a take or pay format (Nvidia pays for some of the compute if it’s not used) instead of needing to secure 100% capacity under long-term contract before getting any financing. This increases lender confidence without Nvidia effectively originating loans, while Nvidia takes a share of future revenue generated by the facility.

Capacity and Financial Guarantees:

Supply chain commitments since last quarter spiked from $119B to $279B. This was related to securing needed memory supply to support the next couple years of growth. On top of this $279B they have another roughly $40B in commitments tied to future data center construction and cloud service agreements to procure capacity for its operations. They see demand signals remaining very strong, and are putting their money where their mouth is with this step-up in future commitments. In addition, Nvidia is also using its cash to assist the supply chain in financing the land and power required to support planned data center growth. That represents another $56B. 

There’s also $50B being directly invested into frontier AI labs. That, paired with the $500B in financing platforms being created by Apollo, BlackRock, Blackstone and others, should allow these companies to find affordable capital, which is a current bottleneck in the way of even faster growth. And again, that can now happen without direct Nvidia funding. Nvidia stepping in to help younger, less credit-worthy companies (like frontier model labs) naturally makes it easier for them to collect the cash they need to buy more Nvidia hardware. Yes, this is a form of circular financing that has gotten companies in trouble in the past. 

Management addressed the elephant in the room head-on. They truly do think this time is different (I get nervous just typing that) and see the AI labs becoming the “largest technology companies in the world.” They’re confident in returns and view the associated risk tied to this reward as palatable. If OpenAI or Anthropic failed for whatever reason, Nvidia would be on the hook for some hefty commitments (more on that in a moment), but they could also sell this compute to other customers to fund those payments. Their view is that this is a once-in-a-generation platform shift, and they’re willing to get aggressive to take a bigger piece of it. They also see overall demand remaining strong enough to make all of these cash uses quite attractive. Time will tell. 

And there’s one more thing to cover here. Nvidia is co-signing on OpenAI’s Ohio data center campus. SB Energy is the site owner, OpenAI is the 20-year lease tenant and Nvidia is the guarantor of $108.5B in total OpenAI commitments. Through the SB Energy partnership, Nvidia locked up land, power and shell (LPS) capacity and will see its chips exclusively used in that facility. The arrangement could equate to 1.5M GPUs per chip generation or $175B in total revenue coming Nvidia’s way. They’re also assisting Anthropic (I think) on 2 gigawatts of compute in addition to the capacity they’ve already built without Nvidia’s financial support. As long as OpenAI and others can pay these bills, the returns on the arrangement should be quite compelling. As long as they can pay these bills.  This all makes it a little more clear why I think the $500B maximum commitment already discussed is so encouraging for Nvidia. Nvidia is rapidly adding exposure elsewhere, so methods of offloading investments or making fewer of them is good.

Edge Segment:

Edge computing refers to running AI on the device rather than through a remote data center. Big news this quarter included Microsoft deciding to revamp its PC architecture using Nvidia’s RTX Spark. As a reminder, this is its superchip (CPU + GPU + memory) that should create a lot more potential within consumer experiences.

On the autonomous vehicle side, the Drive Hyperion platform (Nvidia’s reference hardware and sensor kit for robotaxi-ready cars) added Foxconn, VinFast, Uber and HUMAIN as partners.  Alpamayo 2 Super is now commercially available. This is its open-source reasoning model specifically for autonomous cars. Cosmos 3 is its newest fully open multimodal model, and one that is grounded in the laws of nature and physics. This will be heavily utilized for training robots, alongside the Isaac GR00T Reference Humanoid Robot, which is an open hardware blueprint for anyone building humanoids. And finally, several open-source agent tools debuted to help developers turn all of this raw capability into useful outcomes at commercial scale.

Generally speaking, physical AI is the end market to focus on within this bucket. Humanoid and other robotics markets are going to explode in size in the coming years, which is why Nvidia reorganized their reporting segments. They wanted to more directly show that proliferation in their financial results. You don’t make that move unless you’re excited. 

More:

This was a great quarter. I think sometimes we forget how special this fundamental performance is because we’ve seen Nvidia deliver it every quarter over the last few years. Execution like this would be passionately celebrated and rewarded for pretty much any other company on the planet. But for this one? Expectations are higher because of how impressive they’ve been. As I talked about in the Discord, the entire world knew it was going to be a great quarter. That’s no surprise.  To deliver that positive surprise, Nvidia needed to offer more visibility looking out further into the future. As everyone juggles with the idea of when this supercycle will finally peak, it’s visibility and confidence like this that matters most for allowing a potential next leg higher to occur. And they generated this vital dose of investor confidence via FY 2028 commentary.

This is the chip and data center hardware performance leader by a wide margin. While cheaper GPUs and inexpensive custom processors are finding some market share, there is a massive amount of overall demand and plenty of appetite to pay up for the best option out there. This company will  continue to thrive for as long as this cycle keeps rocking, and commentary during the call points to us having at least another year of that being the case. The multiple is attractive. The team is legendary. The runway still looks good. I think bulls should be pleased. For my specific portfolio, this makes me very happy to continue owning the sector exposure that I do. I am not interested in trimming (or adding to) my SOXX position.