NVIDIA Q4 2026: Expansion into agentic and physical AI

NVIDIA continues to deliver steady quarterly growth, transitioning from a specialty chipmaker to the primary architect of a new industrial era. This expansion is driven by the global shift toward AI Factories, where data centres are being completely rebuilt to generate intelligence rather than just store data. Amid rapid adoption of its Blackwell architecture and development of its next-generation Rubin platform, the company is successfully turning massive computation into revenue for its customers.

Revenue during the quarter accelerated from Q3 and increased 19.5% sequentially. NVIDIA’s annual research and development budget neared US$20 billion. In FY2026, the company returned US$41 billion to shareholders through dividends and opportunistic stock repurchases, representing 43% of free cash flow. However, the company’s focus remains strategic reinvestment.

NVIDIA serves a wide range of customers, ranging from cloud service providers and AI model makers to large businesses and sovereign nations. Demand for the Blackwell architecture is strengthening as customers move beyond just training AI to actively running it (inference). This growth is fuelled by a global shift toward accelerated computing and the rise of agentic and physical AI. Sustained demand continues to outpace supply across NVIDIA’s entire product roadmap despite its move to increase inventory by 8% and expand purchase commitments significantly. This provides visibility into supply and capacity for the next several quarters.

NVIDIA’s GB300 NVL72 system delivers up to 50x better power efficiency and 35x lower costs per AI token than competitors. Through constant software updates, NVIDIA improved performance by 5x in just four months on the same hardware.

NVIDIA’s software platform, CUDA, remains the dominant industry standard. Its fungibility (flexibility) is a key advantage, as nearly all 1.5 million AI models on Hugging Face are optimised for NVIDIA hardware. This makes it easy for sectors like robotics and telecommunications to adopt their technology. Further, deep ties with leaders like OpenAI and a US$10 billion investment in Anthropic ensure NVIDIA stays at the centre of the AI frontier.

Total revenue in the upcoming quarter is expected to reach US$78 billion, with a potential 2% deviation. The revenue continues to be led by Data Centre without any contribution from China. Gross margin in Q1 2027 and the full year is expected to maintain around the mid-70s.

Data Centre

Data Center full-year revenue grew 68.2% year-on-year to US$193.7 billion, driven by the rollout of the Blackwell architecture. This represents a nearly 13x increase since the launch of ChatGPT three years ago. Segmental quarterly revenue increased 21.7% sequentially.

Despite record production, demand outpaces supply. This applies not only to the newest Blackwell chips but also to older models like Hopper and Ampere. Nearly nine gigawatts of power on Blackwell is now operational. The Grace Blackwell (GB200) systems alone now account for two-thirds of all Data Centre revenue.

Revenue from Sovereign AI business tripled to over US$30 billion in FY2026, fuelled by national infrastructure projects in Canada, France, the Netherlands, Singapore, and the UK. Conversely, revenue from China remains at zero due to US export restrictions. It is worth noting that Chinese competitors are making progress following recent IPOs.

NVIDIA began sending samples of its upcoming Rubin platform to customers. Production shipments are scheduled for the second half of 2026. Rubin is designed to be significantly more efficient, potentially reducing inference costs by 10 times and the number of GPUs required for training by 75% compared to Blackwell.

Management expects revenue to grow every quarter throughout 2026 (calendar year) despite ongoing challenges with global power constraints.

Gaming

Gaming full-year revenue rose 41% to a record $16.0 billion. The segment grew due to strong demand for the new Blackwell chips and better availability. These graphics cards remain the top choice for gamers and digital creators. During the quarter, NVIDIA improved performance on AI PCs, making AI tasks like language models run 35% faster. Despite strong demand and healthy inventory, the company expects the segmental growth to be limited by supply constraints beyond the upcoming quarter.

Professional Visualisation

Professional visualisation full-year revenue rose 70% to a record $3.2 billion. Although a smaller revenue contributor, NVIDIA expanded its workstation lineup with the RTX PRO 5000 Blackwell (professional-grade desktop chip). This hardware allows developers to build and test AI models locally before scaling them to the cloud.

Automotive

Automotive full-year revenue rose 39% to a record $2.3 billion. Alpamayo is a recently introduced open-sourced portfolio of AI models and datasets for autonomous vehicles that is capable of complex reasoning. Alpamayo will be integrated into the upcoming Mercedes-Benz CLA.

In FY2026, revenue from physical AI exceeded US$6 billion, driven by demand for self-driving solutions from providers like Waymo, Tesla, and Uber. This sector is expected to scale to millions of vehicles over the next decade, requiring significantly more compute.

Key analyst questions

NVIDIA expects global spending on AI data centres to reach US$3-$4 trillion by 2030, driven by generative AI. CEO Jensen Huang thinks that agentic AI has reached an inflection point. While the current boom is driven by digital agents (like coding assistants), physical AI in areas like autonomous manufacturing and self-driving fleets is seen as the next phase of major growth.

The top five cloud service providers contribute to about 50% of Data Centre revenue and are projected to spend nearly US$700 billion on AI infrastructure in 2026. Huang expressed confidence in continued capital expenditure growth among its customers. He reiterated that ‘compute equals revenues’. For cloud service providers and hyperscalers, capital expenditure drives compute capacity, and compute capacity drives revenue. Without investing in AI infrastructure today, revenue growth would be constrained. The demand for high-performance infrastructure is expected to remain robust through 2027.

CFO Colette Kress responded to an analyst that NVIDIA’s gross margin can be sustained as long as the company continues to deliver significant gains in performance per watt and per dollar to customers.

AI token demand is growing exponentially as agentic AI requires faster processing to function. This has kept demand high for all hardware, including older models. However, as GPUs get faster, traditional processors (CPUs) have become a bottleneck. To solve this, the Vera CPU is specifically designed to handle heavy AI data rather than everyday computer tasks, ensuring the entire system stays fast.

Despite challenges like poor economics and heat dissipation, Huang expressed optimism in the development of space data centres in the future, driven by the presence of abundant solar energy and the need for high-speed processing.

The fifth perspective

NVIDIA is establishing itself as the essential AI platform by investing across its entire hardware and software stack to support a diverse ecosystem of startups and model builders. By designing its chips and software to work together, the company delivers the performance gains necessary to maintain steady pricing. The fungibility of the NVIDIA platform across both cloud and edge environments ensures its industry leadership.