NVIDIA looks set for its next growth chapter

‘Everything is sold out’ was the key message from NVIDIA’s latest earnings call. Geopolitics aside, the AI leader continues to see insatiable demand for its products, and looks to remain the prime beneficiary of the company’s estimate of US$3-4 trillion of AI infrastructure spending over the next five years.*

It was a fairly uneventful quarter (second quarter, fiscal year 2026), marked by a continuation of strong growth. The market is becoming increasingly blasé to NVIDIA’s financial ascent, which in July 2025 became the first public company in history to reach US$4 trillion market value. With Blackwell rack level teething issues in the rear-view mirror and the company refuting any delays to Rubin (the successor to Blackwell), investors can look forward to a powerful innovation curve. NVIDIA (NASDAQ: NVDA) is providing a solution to the power challenges of AI data centres as well as delivering return on investment (ROI) by helping customers deliver agentic AI with its associated exponentially higher computational requirements at much greater energy efficiency per token.

Here are five key messages that we took away from the earnings call:

1. Epicentre of geopolitics

The key near-term variable is China H20 sales given shifting geopolitical sands, which remains a dynamic situation. NVIDIA removed US$8 billion of sales from the July quarter and writing off US$4.5 billion in inventory in expectation of denial, post the US government requiring export licenses to China for H20 chips. Despite the US government issuing some licenses in August to select Chinese customers, NVIDIA has so far not shipped any H20 chips since then, and has not included H20 sales to China in its guidance. Potential H20 China sales this quarter was quantified at US$2-5 billion if “geopolitical issues recede”,* while the company remains hopeful the US government will ultimately allow them to export some form of Blackwell product to China.

With ‘NVIDIA GPU access the US carrot to every trade negotiation’, NVIDIA will likely remain at the epicentre of geopolitical volatility, and consequently China will continue to push domestic companies to localise their AI compute. However, it’s worth noting China remains a source of long-term upside for NVIDIA versus the current low single-digit percentage of its data centre sales.

2. Sovereign AI ramping up

Despite the China headwinds, NVIDIA has said it is on track to more than double sovereign AI sales (sales to countries/governments) this year to over US$20 billion.* This is a new customer base for the technology sector and marks a transition from the internet era, given every country’s desire to be less reliant on the US and have its own AI infrastructure and capability due to the importance of this new technology wave.

3. Full stack innovation solving for power challenges

While the focus is always on the chips, this ignores the huge amount of software innovation driving greater power efficiency in AI data centres. The ability to train AI models at lower precision maths when combined with the latest Blackwell hardware is driving 50x greater energy efficiency per token today vs the prior Hopper generation. NVIDIA’s networking business also doubled year-on-year as NVIDIA continues to come up with solutions for ever larger and more efficient clustering of GPUs. This includes its GB200 NVL72 rack as AI data centres scale up to hundreds of thousands of GPUs to deliver both frontier models, and inferencing workloads, as reasoning models and agentic AI drive a surge in token processing.

4. No Rubin delay

NVIDIA continues to follow an accelerated roadmap and refuted recent reports of a delay to the next generation Rubin architecture. The company noted that production is in progress in TSMC semiconductor factories, along with the five new next generation AI superchips announced at its GTC event in March. This includes the new Vera generation ARM-based CPU as well as multiple networking chips that enable a Rubin third generation rack level product. The product creates a high hurdle for the competition, notably rival AMD, which aims to deliver its first rack level offering in 2026, as well as ASIC suppliers.

5. New US$60 billion share buyback

In one of this year’s great understatements, US Treasury Secretary Scott Bessent said this week when asked about the potential for a similar US government Intel stake in NVIDIA, “I don’t think NVIDIA needs financial support”.1 With almost US$50 billion in net cash on the balance sheet,2 NVIDIA will be making more in interest income than most companies make in total profit. That strong profitability and free cash flow generation underpinned the announcement of a new US$60 billion share buyback.*

Written by Richard Clode, CFA