Nvidia put up a stellar report, yet the market seems to be numb to the strong fundamentals coming out of the company. Not only did Nvidia report record QoQ data center growth at $11.1 billion, but the CFO also stated they expect to see QoQ growth every quarter this year with visibility into CY2027. Management also confidently stated they will exceed the $500 billion estimate they had provided for Blackwell and Rubin combined.
Management also stated that Nvidia is now the world’s largest Ethernet networking company, accomplishing this in just a few short years of Spectrum-X deployments. Networking growth was 263% this quarter, up 100 points from last quarter’s growth rate of 162% YoY growth.
The Q&A session centered around whether capex can continue to grow, if the CUDA moat is still intact, and the company’s strategy for keeping margins elevated. The analysts poked yet there was no chink to be found in the armor.
Nearing $700 Billion – can Capex Continue to Grow?
About four months ago, our firm published on the AI Monetization Supercycle here and here. The point of those articles was to say that debates around an AI bubble are loud at this exact point in time because of Big Tech has spent an exorbitant amount of money for the training phase, which is a foundational stage rather than a revenue-generating stage. As a result, investors see soaring capex, and meanwhile, the revenue streams are still in their infancy, which fuels the bubble debates.
My contention has been that we are too early in the cycle for these concerns to carry weight. Inference is synonymous with monetization, and we know the inference stage is incoming with architectures like Rubin and Helios (AMD) optimized for high-throughput and low latency at scale. This means they are not only designed to train models, but rather to deploy them into applications where revenue is generated, which is supported by the sheer amount of memory these incoming generations of GPUs offer.
To add to this, Nvidia recently acquired Groq, a startup with specialized inference hardware called LPUs or language processing units. LPUs allow trained models to generate answers very quickly at several hundreds of tokens per second. Groq offers a cloud platform for Ai inference-on-demand to access language models and AI capabilities in the cloud, and also a rack cluster for on-premises AI inference for data centers that want LPUs. Jensen Huang stated to expect announcements regarding how Nvidia plans to strategically integrate Groq : “What we’ll do with Groq is you’ll come to see GTC, but what we’ll do is we’ll extend our architecture with Groq as an accelerator, in very much the way that we extended NVIDIA’s architecture with Mellanox.”
Back to whether capex can increase, there was a question on this in the Q&A session, to which Huang’s answer lines up with my understanding of how the return on capital for infrastructure is set to improve:
Here was the question from Antoine Chkaiban from New Street Research:
“[…] Jensen, I’m curious, you know, when you look at your top cloud customers, cloud CapEx close to $700 billion this year, many investors are concerned that it would be harder for this level to grow into next year, and for several of them, their cash flow generation capability is also getting compressed. I know you’re very confident about your roadmap, right? And your purchase commitments and whatnot, but how confident are you about your customers’ ability to continue to grow their CapEx?
Jensen Huang:
“I am confident in their cash flow growing. The reason for that is very simple. We have now seen the inflection of agentic AI and the usefulness of agents across the world in enterprises everywhere. You’re seeing incredible compute demand because of it. In this new world of AI, compute is revenues. Without compute, there’s no way to generate tokens. Without tokens, there’s no way to grow revenues. In this new world of AI, compute equals revenues. […]
Now, I am certain at this point that we are at the inflection point. We’ve reached the inflection point, and we’re generating profitable tokens that are productive for customers and profitable for the cloud service providers. The simple logic of it, the simple way to think about it, is computing has changed. What used to be software running on computers, modest amount of computers, you know, call it $300 billion-$400 billion worth of CapEx each year, has now gone into AI. AI, in order to generate tokens, you need compute capacity, and that translates directly to growth, and that translates directly to revenues.”
The Defensibility of CUDA during the Inference Phase
One of the more intriguing questions on the call was whether CUDA matters as much as dollars shift from training to inference. Jensen Huang’s answer was essentially that CUDA turns inference from raw compute into a monetizable stack as CUDA offers TensorRT-LLM tools and NVLinkn optimizations, which in turn, lead to a higher performance-per-watt. This also ties back to the notion that inference equals revenue as Huang is stating agentic AI can push tokens into the thousands to hundreds of thousands per session.
Here was the exchange:
Atif Malik, Analyst, Citi:
Thank you for taking my question. Jensen, I’m curious if you can touch on the importance of CUDA, as now more of the investment dollars in AI are coming from inference workloads.
Jensen Huang:
Without CUDA, we wouldn’t know what to do with inference. The entire stack from TensorRT-LLM that we introduced a few years ago, which is still the most performant inference stack in the world.
Optimizing it for NVLink requires us to discover and invent new parallelization algorithms that sits on top of CUDA to distribute the workload and the inferencing to take advantage of the aggregate bandwidth across NVLink Switch. NVLink Switch has enabled us to deliver generationally 50 times more performance per watt. It’s just an incredible leap, and it’s sensible. NVLink Switch is a great invention. It was hard to do.
The, the creation of the switching technology, disaggregating the switches, building the system racks, all of that, you know, we did it all in plain sight, and everybody knew how hard it was for us to do. The, the results are incredible. You know, performance per watt is 50 times. Performance per dollar, 35 times. The leap in inference is incredible. It’s very important to realize that inference equals revenues now for our customers. Because agents are generating so many tokens, and the results are so effective. When the agents are coding, it’s off generating thousands, tens of thousands, hundreds of thousands, because they’re running for, you know, minutes to hours. These systems, these agentic systems, are spawning off different agents working as a team.
Durability of Gross Margins
Nvidia has impressive gross margins that are high compared to previous eras where the gross margins were below 60% and sometimes as low as 40%. The gross margin reported tonight was 75%, thus analysts are wondering if this can be sustained, especially from the rising costs of memory.
Huang provided his typical approach, which is to not directly answer the question if he doesn’t care for the angle. Instead of discussing what effects memory prices will have on the gross margin, he stated that gross margins are sustained when Nvidia delivers “generational leaps” that create a step-up in performance per watt and performance-per-dollar that outpace what Moore’s Law alone could deliver.
“The single most important lever of our gross margins is actually delivering generational leaps to our customers. That is the single most important thing. If we could deliver generationally, performance per watt, that exceeds dramatically what Moore’s Law can do. If we can deliver performance per dollar dramatically more than the cost of our systems, than the price of our systems, then we can continue to sustain our gross margins. That’s the simple, most important concept.”
Data Center QoQ Revenue Sets a New Record
Data Center revenue in the quarter was $62.31 billion, with YoY growth accelerating nine points to 75% YoY although QoQ growth moderated 3 points to 22% QoQ, due to the larger revenue base. This compares to 25% QoQ growth last quarter, marking two strong back-to-back quarters from Blackwell Ultra shipping in volume.
One key takeaway is that the company delivered a higher QoQ growth on a dollar basis in Q4 on a larger revenue base, highlighting just how rapidly Blackwell and networking platforms are ramping. Data Center revenue increased $11.1 billion sequentially, surpassing Q3’s $10.1 billion growth. Assuming similar mix in Q1 on the $78 billion guide, Data Center revenue could come in around $71.4 billion at the midpoint, or up another $9.1 billion QoQ.
Within Data Center, Compute revenue rose 58% YoY and 19% QoQ to $51.33 billion, slowing from 27% QoQ in Q3 though YoY growth accelerated 2 points. Networking revenue accelerated more than 100 points to 263% YoY to $10.98 billion, with QoQ growth accelerating 21 points to 34%.
For the full year, Data Center revenue was $193.74 billion, up 68% YoY; this was driven by a 59% increase in Compute revenue to $162.36 billion, and a 142% increase in Networking revenue to $31.38 billion.
Looking ahead through FY27, Blackwell Ultra and related networking will be the main driver of this near-term QoQ growth, but the bigger story will be Rubin and if the upcoming platform is fully priced in, considering it is expected to carry a significant price premium over the GB300s. For example, the Street is suggesting Rubin could carry up to a 40% to 50% premium versus the GB300’s estimated $3.5 million price tag.
While the timing of the ramp remains a bit more fluid at the moment, Wolfe Research projects Rubin to ramp smoothly in 2H, and quickly meet a similar volume profile to Blackwell in the 1,000 rack/week range. Under a rough projection that Rubin ramps from ~8,000 racks/quarter to 13,000 in 2H, for total volume of ~21,000 racks in FY27, this estimated ASP uplift alone could drive incremental revenue of ~$12.6 billion and $20.5 billion.
Financials
Revenue Accelerates to 73% YoY
Nvidia’s Q4 revenue accelerated to 73.2% YoY from 62.5% YoY in Q3, while QoQ growth moderated slightly to 20% QoQ from 22% in Q3, due to Nvidia’s increasing revenue base. Revenue for the quarter was $68.13 billion, beating estimates by 2.9%.
For Q1, Nvidia guided for revenue growth to accelerate further to 77% YoY, forecasting revenue to be $78 billion, +/- 2%, coming in well ahead of consensus for $72.03 billion. Sequential growth would again moderate to 14.5% QoQ at the midpoint of guidance, though again this is simply due to the law of large numbers as dollar growth is projected to be nearly $10 billion QoQ, versus $11.1 billion in Q4.
For FY26, Nvidia reported revenue of $215.94 billion, up 65.5% YoY, with this growth driven by Data Center revenue increasing 68.2% YoY to $193.74 billion. Nvidia is not providing full year guidance for FY27, but considering that FQ1 is already $6 billion ahead of estimates, the current estimate for $330.76 billion for 54.6% growth may quickly move tens of billions higher.
Networking Steals the Spotlight with 263% Growth
Networking revenue was once again quite an outlier in Q4’s report, with growth accelerating sharply on both a YoY and QoQ basis in the quarter. Networking revenue was $10.98 billion in Q4, up 34% QoQ and 263% YoY – this represents more than a 100 point acceleration from 162% YoY growth in Q3, while QoQ growth accelerated 21 points. Nvidia said the strong growth stemmed from the introduction and ramp of NVLink compute fabric for both GB200 and GB300 systems, as well as growth in Ethernet, InfiniBand and Spectrum-X.
On the call, it was stated that Nvidia is likely the largest Ethernet company in the world: “The amount of switching that we do per rack is really quite incredible. We’re also now the largest networking company in the world. If you look at Ethernet, we came into the Ethernet market about a couple of years ago, into Ethernet switching, and I think that we’re probably the largest Ethernet networking company in the world today, and surely will be soon. Spectrum-X Ethernet has been a home run for us. You know, we’re open to however people want to do, networking.”
This rapid acceleration in 2H drove networking revenue up 142% YoY to $31.38 billion, contributing more than 16% of Data Center revenue, up from more than 11% in FY25. To put in perspective the sheer size of Nvidia’s Networking segment, it alone was >1.5X of Broadcom’s entire AI revenue from FY25, and just $3 billion shy of AMD’s entire annual revenue last year.
Turning to Nvidia’s other segments:
Gaming revenue increased 47% YoY but declined (13%) QoQ to $3.73 billion, with YoY growth driven by strong Blackwell demand and the QoQ decline due to channel inventory moderating after the holiday season. However, Nvidia expects supply constraints to be a headwind on Gaming revenue in FQ1 and beyond.
Pro Viz revenue showed an unusually strong 159% YoY and 74% QoQ increase to $1.32 billion in Q4, accelerating from 56% YoY and 26% QoQ in Q3. Nvidia said this was driven by ‘exceptional’ Blackwell demand, likely for its DGX Spark desktop supercomputer as was the case in Q3.
Automotive revenue was soft in Q4, up just 6% YoY and 2% QoQ to $604 million, driven by adoption of Nvidia’s self-driving vehicle platforms.
OEM and other revenue rose 28% YoY but declined (7%) QoQ to $161 million.
For FY26:
- Gaming revenue was $16.04 billion, up 41% YoY.
- Pro Viz revenue rose 70% YoY to $3.19 billion.
- Automotive revenue rose 39% to $2.35 billion.
- OEM and other revenue rose 59% to $619 million.
Margins Expanded in Q4
Nvidia’s gross margins moved higher in Q4 to the 75% range, with GAAP operating margin following this expansion and moving back to the 65% level. To note, starting in Q1, Nvidia’s adjusted margin figures will include SBC.
Q4 GAAP gross margin was 75%, slightly ahead of management’s guidance for 74.8% and expanding by 2 points YoY and 1.6 points QoQ, with the sequential expansion driven by Blackwell and better product mix. Adjusted gross margin was 75.2%, expanding 1.7 points YoY and 1.6 points QoQ. Looking ahead to Q1, Nvidia guided for a minimal QoQ contraction to 74.9% for GAAP gross margin and 75% for adjusted gross margin.
Q4 GAAP operating margin was 65%, also slightly ahead of guidance for 64.5%, and expanding 3.9 points YoY and 1.8 points QoQ, showing a hint of operating leverage. Adjusted operating margin was 67.7%, expanding 2.8 points YoY and 1.5 points QOQ, coming in slightly above the guidance for 67.3%. For Q1, Nvidia guided for operating margins to be flat at 65% and since the adjusted margin figures will include SBC, it will be lower sequentially at 65.4%.
Q4 GAAP net margin was 63.1%, expanding 6.9 points YoY and 7.1 points QoQ, while adjusted net margin was 57.2%, up 1.1 points YoY and 1.5 points QoQ.
For FY26, as evidenced by the chart above, gross and operating margins contracted YoY as Nvidia faced inventory charge impacts related to the H20 China ban earlier last year. FY26 GAAP gross margin was 71.1%, down 3.9 points YoY, while adjusted gross margin was 71.2%, down 4.3 points.
GAAP operating margin was 60.4%, down 2 points YoY, and adjusted operating margin was 63.6%, down 2.9 points. Despite this contraction, GAAP net margin as relatively unaffected at 55.6%, down 0.3 points, through adjusted net margin contracted 2.7 points to 54.2%.
EPS
GAAP EPS saw a rather large beat in Q4, coming in at $1.76, up 98% YoY and beating estimates for $1.47 by more than 19%.
Adjusted EPS was $1.62, up 82% YoY and beating estimates by just over 5%. Growth accelerated from 60.5% in Q3 and marked Nvidia’s fastest adjusted EPS growth in the last five quarters. Looking ahead to Q1, current consensus estimates point to this acceleration continuing to nearly 108% YoY, though this is likely to be revised higher considering the scale of the revenue beat.
For FY26, GAAP EPS rose 67% YoY to $4.90, while adjusted EPS increased 60% YoY to $4.77; for FY27, initial estimates point to strong EPS growth continuing, with adjusted EPS forecast to rise 67.3% to $7.86.
Cash Flows and Balance Sheet
Cash flows were robust in Q4, with cash flow margins improving significantly on both a YoY and QoQ basis.
Operating cash flow was $36.2 billion in Q4 for a 53.1% margin, up 10.9 points YoY and 11.4 points QoQ. For the full year, OCF was $102.7 billion for a 47.6% margin, down 1.5 points YoY due to the softer Q2 print.
Free cash flow was $34.9 billion for a 51.2% margin, up 11.7 points YoY and 12.4 points QoQ. For FY26, free cash flow was $96.6 billion for a 44.7% margin, down 1.8 points YoY.
Cash and equivalents totaled $62.6 billion, while debt was $8.47 billion.
Inventories increased more than 8% QoQ to $21.4 billion, though more importantly, Nvidia’s supply related commitments surged — we highlighted this last quarter as a key sign that the strong data center QoQ revenue inflection would continue.
In Q4, Nvidia’s supply-related commitments surged nearly 90% sequentially to $95.2 billion, a major step-up from the prior ~$28-30 billion range through late FY25 and the first half of FY26. Nvidia says it is strategically securing inventory and capacity to meet demand beyond the next several quarters, which we believe serves as a key sign that the current accelerated QoQ data center growth of ~$10 billion will likely persist as Blackwell Ultra continues ramping and as Vera Rubin ramps.
Conclusion:
The financials were flawless while the Q&A during the call cleared the pressure points that the bears have leaned on for months, such as sustainability of AI capex, durability of the CUDA moat, and whether memory/input costs can compress margins.
Management’s answers pointed to the same conclusion, which is that demand is broadening, utilization is tightening, and NVIDIA’s ability to deliver step-function gains in performance per watt and performance per dollar will preserve pricing power.
The most important takeaway is that inference is no longer a future event; rather it is the revenue engine for customers today while agentic workloads are increasing token demand. The market will continue to play tug-o-war on whether the AI economy is an investable opportunity or a bubble, but the results this evening continue to make one thing clear: the fundamentals are driving forward record growth and profits with Nvidia remaining an obvious choice for participating in the AI monetization cycle.
Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in NVDA at the time of writing and may own stocks pictured in the charts.
Recommended Reading: