When discussing the results for the second quarter of the fiscal year of 2026, the president of Nvidia pointed out that Global outlays on AI infrastructure can reach $ 3-4 for the end of the decade. This is the thesis of a huge weight for the company's valuation, because with such an investment scale, even maintaining the current position in the value chain may translate into a long -term, powerful stream of revenues. The quarter itself ended on July 27, 2025 brought Nvidia $ 46.7 billion. sales, by 56 percent more a year to year, which confirms the dynamics of demand for systems based on Blackwell architecture.
Huang declaration by 3-4 trillion This is not an isolated opinion – the same market -size projection is also supported by analytical comments and relations from the capital market after the publication of the NVIDIA report. Colette Kress, the company's financial director, explicitly said that the scale and range of “AI factories” created today is building a long -term growth path before Nvidia.
For shareholders, it is important that the company is not based solely on a one -time equipment replacement cycle, but on a growing, distributed demand – from hyperskalers, through the so -called Neo-chums, up to enterprises and sovereign projects.
Check also: The minister of digitization reveals how much Gigafabryka AI will cost in Poland
Hard financial data
Solid financial data has fastened the narrative about Hipwzrost. The whole fiscal year 2025 closed $ 130.5 billion with records with records. revenues, more than twice as much as a year earlier, and the data centers segment became the main business motor.
Special offer
In the second fiscal quarter of 2026, the company once again raised the bar, including thanks to the maturation of the Blackwell platform and the growing scale of NVL system supply. It is this foundation that allows Huang to talk about the next stages of AI factories with such certainty.
The strength of convincing in this story is not only demand, but also the pace of innovation. Nvidia has gone through the annual rhythm of the premieres of systems, which hinders the rivals “jumping” of the generation. After Blackwell and this year's Blackwell Ultra in 2026, Rubin will enter the market, and recently announced Rubin CPX was designed especially for the “massive context” of AI models, i.e. tasks such as video generation or long -text programming.
In the company presentation Rubin CPX and the full wardrobe NVL144 CPX are to provide up to 8 exflops power for inference, a NVIDIA estimates that an investment of $ 100 million. It can generate up to $ 5 billion in such a system. revenues from tokens. This is an example of how the annual term of products translates into new, monetized cases of use.
Added to this is the advantage of full-stack-from GPU and network, through software, to ready platforms that make it easier for customers to quickly run services. All this is provided by NVIDIA.
It is worth adding that the latest results MLEPERF Inference V5.1 have shown that GB300 NVL72 Systems on Blackwell Ultra have established new bandwidth records in the tasks of reasoning, and NVIDIA boasts not only harsh performance, but also clear progress in the cost of cost and efficiency at Wat – that is, parameters that directly improve the economics of factories. From the CIO point of view, what matters is that Higher efficiency per wat and per dollar reduces the total cost of possession and increases the margins of AI services, which promotes loyalty to the platform.
Is the vision of billion revenues realistic?
The market begins to discount: some of the analysts allow the scenario in which NVIDIA will become the first company with revenues of $ 1 trill until the end of the decade. Annually – just as the main beneficiary of the investment in the AI infrastructure.
Such a path would require a high share in the expenditure of data centers and further Expanding the offer beyond the accelerators themselves, but the scale of the demand that hyperskalers are talking about, gives you the basis.
On the other hand, shareholders should be aware of the cyclicality of expenses. There are already forecasts that after the explosion in 2025, the pace of investment in 2026 may slow down – at least in some players – under pressure of depreciation and guarding of profitability. NVIDIA, with a technological advantage and calendar of the Prime Minister, can still win in the portfolio, but short -term fluctuations in orders or export restrictions (e.g. lack of sales of H20 to China in the last quarter) can cause variability of the course and results.
However, the long -term investment thesis is based on a stable trend: an increasing number of generative AI applications and the transition of companies from the experimenting phase to the monetization phase.
In this context, “good news” from Huang is not just a marketing narrative, but a preview of a real, long -term investment series in which NVIDIA strengthens the role of the default infrastructure supplier to AI. The company closes record quarters, maintains the annual rhythm of innovation and expands the platform, and the market still looks very absorbent.
Nvidia shares have increased by 32 percent since the beginning of the year.
Note: information posted in the text is only information and does not constitute a recommendation for the purchase or sale of financial products. This text does not constitute an investment recommendation or investment consulting activities within the meaning of §3 of the Regulation of the Minister of Finance of 19 October 2005 on information constituting recommendations regarding financial instruments, their issuers or exhibitors (Journal of Laws 2005 No. 206 item 1715).
