Nvidia projects $673 billion in sales and signals where AI money is flowing

Nvidia projected 70% revenue growth for its fiscal year 2028, which would put annual sales at around $673 billion if the current Wall Street consensus for fiscal 2027 holds, as for(geeks) reports. The figure would put the chipmaker ahead of Apple and Alphabet in revenue, leaving only Amazon among U.S. tech giants with higher projected sales.

The forecast was delivered by CFO Colette Kress on August 26, 2026, and is well above the 44% average estimate tracked by LSEG. It also marks a shift in posture: the company had never offered a forecast so far out in time, although CEO Jensen Huang had previously given short-term signals about demand for AI chips.

The numbers behind the forecast

The second-quarter fiscal 2027 results provided the basis for the optimism. Quarterly revenue reached $96.2 billion, more than double the same period a year earlier, while data center revenue rose 117%, to $89 billion. Shares rose about 4% in after-hours trading in one report, with another pointing to a 5.6% gain at the session's peak, a difference that reflects the trading range after the release, not a change in the forecast.

The most relevant point for those building software: supply, not demand, is what sets the ceiling in the short term. Huang stated that component shortages, including memory, prevented Nvidia from making an even larger forecast, as AI infrastructure consumes a growing share of global chip and memory capacity.

"Our demand is much higher than 70%. Our supply allows us to deliver 70% with confidence, and we'll keep working with our supply chain to increase that."

What 'tight supply' means for inference costs

For Brazilian developers, the practical reading is straightforward. When the manufacturer itself admits that demand exceeds production capacity, the bottleneck won't disappear over the next few quarters. This puts pressure on GPU prices in the clouds and neoclouds, exactly where training runs and, more importantly, where the inference that powers LLM-based products happens.

Those putting models into production feel this squeeze in per-token pricing and in the availability of GPU instances. The memory shortage Huang mentioned also affects high-capacity boards (such as those used to serve large models), which tends to keep the cost of serving dense models high and reinforces the appeal of alternatives like quantization, smaller models, and the use of an open ecosystem to reduce dependence on cutting-edge hardware.

The customer base spreads beyond the hyperscalers

A recurring investor concern is that Nvidia's growth depended too heavily on a small group of hyperscalers building data centers for a handful of frontier AI labs. Huang stated that the next phase involves a broader set of buyers: regional AI companies, neocloud providers, startups, and traditional enterprises.

Nvidia groups these customers under the ACIE label and sells them more than GPUs, offering much of the data center stack to organizations that can't assemble the system on their own. According to Huang, this category was previously "largely invisible."

"At this time last year, a single lab was driving the buildout; today we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem, and physical AI coming online."

This diversification matters for Brazil because it's precisely the segment of neoclouds, startups, and conventional companies that offers accessible GPU capacity to teams outside the major global providers. A broader buyer base also pushes Nvidia to sell networking, systems, and other data center components alongside its processors, though the reported results don't detail how much revenue comes from these products.

The effect is already showing up outside the U.S.: European semiconductor stocks rose after the results, with ASML gaining about 2.5% and STMicroelectronics, Infineon, and BE Semiconductor climbing between 2% and 4%.

The open question: Nvidia finances those who buy its chips

The company's role went beyond selling hardware. Nvidia is investing in model developers such as OpenAI and Anthropic, backing neoclouds that rent out computing power based on its chips, and helping structure financing for data center construction. One example is $105 billion in financial support for a large computing campus under construction in Ohio, where OpenAI is expected to be a tenant. The company also announced a partnership with major Wall Street banks to arrange up to $500 billion in data center financing.

These arrangements have raised concerns about circular financing: Nvidia helps finance customers or infrastructure, and that money can flow back to Nvidia through purchases of its products. Huang defended the strategy, arguing that frontier AI companies need unusual amounts of capital before they have balance sheets solid enough to raise money cheaply.

"This is the first generation of startups that has needed tens of billions of dollars to fund itself. (...) The cost of building AI, the cost of deploying AI, is very capital-intensive."

Huang also stated that Nvidia's infrastructure can be redistributed among customers and workloads should an AI company fail, which he says limits the risk. It's worth noting that this is Nvidia's own defense, not an independent assessment of the financing risk.

The question that remains unanswered is whether the ACIE customer base can sustain the projected growth without Nvidia's balance sheet continuing to subsidize the expansion. For those building software in Brazil, this is the point to watch: if the GPU capacity boom depends on credit structured by the manufacturer itself, the price and supply stability developers expect from cloud providers is tied to a financial mechanism not yet tested through a downturn.