Nvidia Navigating Elevated Expectations
After a historic surge in revenue growth, Nvidia’s revenue expectations for the second half 2025 and 2026 face downside risk.
Despite strong growth and continued execution of its product roadmap, Nvidia's stock has lost momentum after a spectacular two-year run, underperforming the S&P 500 by approximately 15% since mid-June 2024.
While Nvidia remains the dominant AI player with a formidable economic moat, the lack of meaningful upside in the company’s April quarter revenue guidance challenges its beat-and-raise narrative. This lack of upside is notable given the expectation of potential pull-forward demand from China ahead of the Framework for the Diffusion of Advanced Artificial Intelligence Technology export restrictions, which took effect on January 13, 2025, with most provisions requiring compliance by May 15, 2025.
Following the recent sell off in the stock, I expect a short-term rebound leading up to Nvidia’s GTC event, which begins on March 17th with CEO Jensen Huang’s keynote address on March 18th, where the company will share its outlook and what’s next in AI.
However, looking beyond the near term, what lies ahead for Nvidia's stock? Risks to both direct and indirect (via Singapore) China revenue raise concerns about the sustainability of Nvidia’s growth trajectory in the second half of 2025, especially given elevated investor expectations. While the ramp of Nvidia’s new Blackwell platform reinforces near-term growth, limited upside to revenue and earnings forecasts coupled with risks to China-sourced revenue point to downside risk in the stock heading into and following the company’s next earnings report in May.
Below are five key points highlighting why Nvidia’s valuation is under pressure, the evolving AI investment theme, and the growing risk of AI compute capacity outpacing demand—heightening the likelihood of a cyclical slowdown in AI infrastructure spending and downside risk to Nvidia’s second half 2025 and 2026 revenue forecasts.
1.) The market is implicitly assigning a lower multiple on China earnings, making this the dominate factor behind the stock’s valuation compression.
While concerns over a cyclical peak in AI capex spending have weighed on Nvidia’s stock, a more tangible and quantifiable risk is its direct and indirect (via Singapore) sales exposure to China. Singapore and China sourced revenue collectively accounted for 31% of Nvidia’s total revenue in the fiscal year ending January 2025. This helps explain why despite stronger-than-expected 2025 capex guidance from leading cloud players directly countering fears of a cyclical peak in AI compute investment, Nvidia’s stock has remained under pressure.
In FY25, sales to Singapore surged 3.5x year-over-year—outpacing Nvidia’s overall revenue growth, which more than doubled. In FY25 Singapore became Nvidia’s second largest geographical source of revenue, accounting for 18.1% of total revenue up from 11.2% in the prior year. This surge in growth is widely attributed to the use of Singapore as a transshipment hub, enabling China to indirectly acquire advanced GPUs despite U.S. export restrictions.
The U.S. government's Framework for the Diffusion of Advanced Artificial Intelligence Technology aims to restrict China's access to advanced GPUs and close existing loopholes, posing a significant risk to this revenue stream and could create an immediate headwind for Nvidia’s growth outlook. While the Trump administration may introduce new rules that supersede this framework, it is unlikely to fully reverse restrictions on China's access to more advanced GPUs.
2.) Similar to the telecom bubble of 2000, portions of AI compute infrastructure spending face quality-of-demand concerns.
While the largest cloud service providers dominate AI infrastructure spending, neocloud players represent a meaningful share of demand for Nvidia GPUs and AI compute systems. This cohort of companies comprises a mix of reinvented cryptocurrency miners that have opportunistically pivoted to exploit AI data center growth, existing smaller cloud players shifting focus to AI compute services, and more typical startups.
Neocloud companies have raised significant capital through equity and debt financing, often using chips and AI hardware systems as collateral. This funding has fueled investments in GPUs and AI systems, and the build-out of new AI compute capacity.
Much of this AI compute capacity is built in anticipation of future demand, making these companies speculative expenditures particularly vulnerable to any oversupply in AI compute capacity.
Additionally, there are concerns over the quality of AI compute demand from portions of cloud players customer base, as much of it is fueled by venture capital-backed high-cash burn startups rather than established, revenue-generating enterprises. This raises questions about the sustainability of the current AI compute spending growth trajectory. Nvidia CEO Jensen Huang highlighted this source of demand for cloud players in Nvidia’s August 2024 earnings call saying - “The number of generative AI startups is generating tens of billions of dollars of cloud renting opportunities for our cloud partners.”
3.) Measuring AI compute capacity relative to demand is challenging, as capacity expands not only through new data center builds but also via efficiency gains.
Efficiency gains in model design, system optimizations, and de-bottlenecking efforts that boost GPU utilization, effectively expand existing AI compute capacity. Algorithmic advancements reduce the compute requirements for training AI models, with the potential for continuous improvements in AI algorithm efficiency.
This is not inherently negative for Nvidia’s future demand, but is a difficult to quantify risk. In fact, algorithmic and model efficiency improvements lower AI compute costs, potentially accelerating AI application adoption.
However, the shift from the AI training infrastructure buildout to inference-driven demand may not be seamless, and could result in a period where AI compute supply outstrips demand.
4.) As the AI investment theme shifts from infrastructure spending to applications, Taiwan Semiconductor Manufacturing Corp (TSMC) presents a more compelling opportunity than Nvidia, which trades at nearly a 40% price-earnings multiple premium to TSMC.
As highlighted in my post Surf's up- Catch the AI Smartphone Wave, technological breakthroughs in AI model efficiency are poised to make larger AI models more viable for on-device processing. This will enable more advanced on-device AI features and applications, igniting growth in AI smartphones and consumer devices. With 35% of TSMC’s revenue coming from smartphones, this trend will expand TSMC’s AI addressable market beyond AI infrastructure, positioning it as a key beneficiary of the next wave of AI adoption.
Additionally, regardless of whether demand for merchant GPUs or custom AI chips grows at a faster pace —or whether Broadcom and Marvell or Taiwanese competitors like Alchip and Mediatek are gaining ground in the custom AI chip market—TSMC remains the ultimate winner as the near-monopoly supplier of advanced foundry manufacturing.
5.) Nvidia has both a track record of huge success, but also one of significant cyclical corrections in revenues, with no real signal from management of this risk in advance of the last two downturns.
In the past six years, Nvidia’s revenue has fallen short of original expectations about one-third of the time, with seven quarters during this period experiencing year-over-year revenue declines. This includes a 24% YoY revenue drop in the January 2019 quarter, followed by three successive quarters of declines. Similarly, after a disappointing July 2022 quarter, revenue declined by 17%, 21%, and 13% YoY in the following three quarters.
Different variables led to each of these cyclical downturns, the point being demand for Nvidia is dictated by external factors and a positive outlook from the company should not be simply taken at a face value. Nvidia's ultimate revenue end point still illustrates an extremely strong long term growth trajectory, however one that still can manifest significant and difficult to predicate cyclical volatility.