Nvidia is accelerating. It’s already showing up in Q4 FY26, and their CFO is guiding for “sequential revenue growth throughout calendar 2026.” Q4 revenues were up 73% y/y and 22% q/q to $68B vs. 63% y/y and 24% q/q growth in Q3, its second straight month of acceleration.
Nvidia’s Q4 Data Center revenues, worth $62.3B, accelerated by 75% vs. 66% growth in Q3. Within Data Center revenues, Networking is leading the charge, growing at a 3.6x rate vs. The 2.6x growth rate last quarter, while Nvidia’s Compute sales (GPUs and GPU systems) continue their acceleration, growing 58% to $51.3B vs. 56% and 50% growth in Q3 and Q2 of last year. That’s extremely impressive for a company the size and scale of Nvidia.
Even margins are robustly expanding and are on course to expand this year. While Nvidia exited CY25/FY26 with gross margins of 71.1%, management expects margins to climb to ~75% levels for CY26/FY27 starting with management’s Q1 projection, where they expect Q1 gross margins to reach 75%. That’s a full 400 bp climb, which is mighty.
Blackwell is currently ramping up with strong shipment volumes, while Rubin will begin to ramp up volume shipments in H2. According to management, every hyperscaler and the two major frontier LLM startups, OpenAI and Anthropic, will be using Vera Rubin to process AI workloads this year.
Further, Nvidia now sees >$500B in forward revenues coming in this year from rising shipments of its Grace Blackwell and Vera Rubin platforms, an upgrade from last quarter’s revelation of ~$500B in GPU backlog orders.
This time around, Nvidia did not have any specific updates for its Spectrum-X switches, the company’s Ethernet networking switch platform used in scale-out networks. Last year management revealed the Spectrum-X switch platform had grown to a $10B ARR business. Analysts speculate the Spectrum-X platform is possibly now a ~$12B ARR business as of CY25, and this could grow substantially once Nvidia starts shipping its Spectrum-X Photonics switches (with CPO) in the bottom half of this year.
Nvidia has significantly ramped up its inventory purchases and manufacturing capacity commitments to $95B from just $16B a year ago. That’s actually bullish for Nvidia.
A quick note on Nvidia’s Gaming segment : Management did say that they are facing supply constraints in their Gaming division, which might be a headwind to growth in this division going forward. No specific highlights were made as to whether the supply constraints had anything to do with the ongoing memory shortages.
Nvidia’s CFO did reveal that Nvidia’s top customers, all top 5 American hyperscalers, accounted for 50% of Nvidia’s DC revenue. But she strongly hinted at the next growth curve coming from its non-hyperscaler client cohort.
In our view, this should give some more clarity into the belief that the underlying breadth of this AI infrastructure buildout could be expanding beyond the usual suspects (hyperscalers). This new growth cohort for Nvidia represents clients such as neo-clouds (CoreWeave $CRWV , Nebius $NBIS , IREN $IREN , etc.), frontier model makers (OpenAI, Anthropic, xAI, etc.), enterprises, and even sovereign nations.
And Nvidia continues to expect $3-4T of AI capex by 2030.
By now, analysts on Nvidia’s Q4 call were still not done yet, and questions continued on the capex outlook.
Analysts kept tripping up Jensen Huang through the call on his $3-4T AI capex estimate. This eventually triggered Huang at the end of the call to wax poetic for 7 straight minutes as to why there is enough potential for AI capex growth at a 50% CAGR to his $3-4T estimate by CY ’30.
The short version of Huang’s AI manifesto delved into the economics of running AI models and that software models would now 100% depend on token generation.
This model of token economics was not present in the era of “classical computing” (all the years until 2023-2025). And the sudden surge in token generation over the past 3-4 months due to model inference will, according to Huang, push cumulative hyperscaler capex over $700B this year (55% growth) and anchor the 5-year capex CAGR rates at +50%. Because AI companies need that much more compute to support token generation and processing.
This backdrop in the coming surge in token economics is also the fundamental reason why Huang shifted the company’s focus on being just a GPU company towards enabling his vision of “AI factories.”
Huang eventually tied down his manifesto into three waves of AI—the maturity of the Model Training wave, the current inflection of the Model Inference wave, eventually culminating in the Golden Age of Physical AI, “where we take AI and these agentic systems into the physical applications.”
While all these points were very bullish for Nvidia, its AI peers, and the broader AI infrastructure ecosystem, we did notice a few areas of weakness in Nvidia’s earnings, which will cast a shadow on Nvidia’s shares in the interim period.
In conclusion, we explain Nvidia’s Q4 report while comparing these observations with some Conviction Score ratings and the eventual impact to Nvidia outlook.
