NVIDIA's earnings guidance is extremely optimistic! Jensen Huang dismisses external concerns about capital expenditure: computing equals revenue

NVIDIA’s latest quarterly financial report shows impressive results, with Q4 revenue reaching $68.1 billion, a record high, and non-GAAP EPS of $1.62, also far exceeding market expectations. After the earnings release, NVIDIA’s stock rose slightly in after-hours trading by 0.19%, reaching $195.94. For the next quarter (Q1 FY27), NVIDIA provided very confident financial guidance and progress updates. Q1 revenue is estimated at approximately $78 billion, well above the market forecast of around $72.8 billion. Gross margin is projected at 75%.

During the conference call, analysts asked about whether the cloud giants’ capital expenditure of up to $700 billion can continue, whether the long-term $4 trillion data center CapEx framework still holds, the significance of acquiring Groq, and issues related to space data centers and gross margins. The following summarizes Jensen Huang’s responses to these topics.

NVIDIA’s latest quarter impresses with record-breaking revenue

NVIDIA delivered an outstanding performance in the latest quarter, surpassing Wall Street expectations in both revenue and profit, setting new records. Q4 revenue reached $68.1 billion, a historic high, significantly exceeding the market consensus of $65.9 to $66.1 billion. Revenue grew 73% year-over-year and 20% quarter-over-quarter.

GAAP EPS was $1.76; non-GAAP EPS was $1.62, beating the market expectation of $1.53. Full-year revenue totaled $215.9 billion, up 65% year-over-year. Full-year GAAP EPS was $4.90; non-GAAP EPS was $4.77. The Data Center segment remains the main growth driver, with Q4 revenue of $62.3 billion, up 22% quarter-over-quarter and 75% year-over-year.

NVIDIA provides highly confident forward guidance

Regarding the next quarter (Q1 FY27) and future product roadmap, NVIDIA issued very confident financial forecasts and progress updates. Q1 revenue is estimated at about $78 billion, well above the initial market estimate of around $72.8 billion. The GAAP gross margin is forecasted at 74.9%, and non-GAAP at 75.0%.

This time, there was no mention of the production timeline for the Rubin platform. According to previous statements, TSMC has entered the tape-out stage, aiming for mass production in the second half of 2026. The new generation Vera Rubin cabinets (such as NVL72) are already in ramp-up production, with shipments to major cloud service providers and partners expected to begin in the second half of 2026.

Cloud giants’ capital expenditure: Agentic AI drives unprecedented compute demand

Bank of America Securities asked whether the cloud customers’ $700 billion capital expenditure can continue. Amid concerns that cloud giants’ cash flow might be under pressure and that next year’s CapEx growth could slow, Jensen Huang expressed high confidence in customer profitability and cash flow growth.

He pointed out that the industry has officially entered a turning point with Agentic AI. The popularity of agents like Claude Code, Codex, and OpenClaw has generated incredible computational demands. In this new AI world, “compute equals revenue.” Without computing power, tokens cannot be generated; without tokens, revenue cannot be created.

Historically, about $300-400 billion annually was spent globally on traditional software CapEx. Now, this capital is rapidly shifting toward AI. Because tokens generated by agent systems can bring real productivity and profit to customers, this enormous compute demand will directly translate into revenue growth for NVIDIA and cloud customers.

Economic issues caused by AI productivity revolution? The key is where the money goes

Huang’s response not only defined the massive compute demand created by Agentic AI but also provided a solution for the cloud giants’ huge CapEx. He also highlighted that capital previously spent on traditional software will now shift to AI.

This reminds the author of Citrini’s view that the productivity revolution driven by AI could lead to a surge in unemployment and economic problems. Vincent, co-founder of Manbiao, pointed out that the issue isn’t that SaaS companies have no revenue or that the economy is doomed. Instead, where did the money paid to SaaS companies go? Is it used for stock buybacks or reinvested?

Suppose a company subscribes to a $10 SaaS service. In the AI era, SaaS companies might see a $10 revenue decrease, but that $10 doesn’t just disappear. Three dollars might become new AI supply chain revenue, and the remaining $7 directly becomes additional profit for the company, increasing profit margins. So, money isn’t lost; the issue is how it is distributed.

(AI too successful causing economic crisis? Institutional forecast for 2028: unemployment over 10%, S&P drops 38%)

From Agentic AI to physical AI, data center CapEx of $4 trillion remains valid

Regarding the long-term framework where data center CapEx could reach $3 trillion to $4 trillion, Huang Huang gave a positive response, explaining this trend through “Token Economics.” He stated that future software will no longer be “pre-recorded” or pre-written but will be generated in real-time based on user intent. This real-time generation requires over 1,000 times the compute of traditional methods. Therefore, every company will rely on AI and have its own AI factory to continuously generate tokens.

Huang emphasized that the current wave of demand is driven by agent AI (such as AI assistants helping engineers write code), which has already reached an inflection point with exponential growth in demand over the past two to three months. The next major opportunity after this wave will be physical AI—integrating AI and agent systems into manufacturing and robotics applications.

What does the acquisition of Groq mean for NVIDIA?

Regarding whether future architecture will favor chiplet designs and the strategic significance of acquiring Groq, Huang Huang said NVIDIA prefers to delay using chiplets as much as possible because crossing chip interfaces introduces unnecessary latency and power consumption. NVIDIA’s CUDA architecture remains dominant because of its highly efficient hardware design.

On Groq and low-latency decoding technology, Huang Huang previewed that more details will be shared at GTC, but he clearly stated that Groq will be positioned as an “accelerator” to expand NVIDIA’s compute architecture, similar to how Mellanox was used to expand networking infrastructure. All NVIDIA GPUs maintain high architectural compatibility, ensuring that software optimization investments can be leveraged across generations, further enhancing customer performance per dollar.

Can NVIDIA maintain its ultra-high gross margin? Huang Huang responds

When asked whether NVIDIA can sustain its high gross margin in the mid-70% range long-term, Huang Huang explained that the most straightforward and critical lever is to continuously provide customers with cross-generation leading advantages. As long as NVIDIA can deliver performance per watt that significantly surpasses Moore’s Law and enable customers to achieve performance gains per dollar far exceeding system costs, it can maintain high margins.

NVIDIA offers a full suite of AI infrastructure products annually (such as six new chips this year, with multiple new products in the next-generation Rubin), combined with top-tier hardware-software integration, continuously providing the most valuable compute power for the exponentially growing token demand worldwide.

Huang Huang on space data centers

Regarding the feasibility and economic benefits of moving data centers into space, Huang Huang admitted that the current economic benefits are indeed poor but will gradually improve in the future. The space environment is very different from Earth, with abundant solar energy and extremely cold conditions. However, due to the lack of airflow and inability to use water cooling, heat dissipation must rely on large conduction heat sinks. Despite the challenges, NVIDIA’s Hopper GPU has already successfully entered space.

Currently, the best application scenario for GPUs in space is “ultra-high-resolution imaging processing,” where AI is used directly in space for denoising, re-projection, and ultra-high-definition imaging, filtering valuable information before transmitting it back to Earth. This approach is far more efficient than transmitting petabytes of raw data for processing on Earth.

This article NVIDIA’s earnings forecast is overly optimistic! Huang Huang dispels external CapEx concerns: compute equals revenue. Originally published in Chain News ABMedia.