JPMorgan: The CUDA moat is crumbling, and Nvidia faces real challenges beyond the CEO's departure
JPMorgan points out that Nvidia’s share in the 2026 compute protocol will drop to just 28%, with strong competition emerging from AMD, Google TPU, and others. Meanwhile, CUDA’s moat was breached by AI in only three days. The open-source market is facing tightened regulation and increased risks of export compliance. Although Nvidia’s current valuation remains attractive, both major drivers supporting high growth expectations have been undermined.
Nvidia is currently facing structural pressures from both ends, which are as significant as, if not more profound than, the potential departure of founder Jensen Huang.
JPMorgan strategist Michael Cembalest pointed out in his latest report, "Nvidia and the Next Step for AI," that cutting-edge AI labs are diversifying their procurement focus among multiple vendors such as AMD, Google, and Amazon. Nvidia’s share in compute contracts post-2026 will be only 28%.
Meanwhile, the open-source model ecosystem is facing regulatory scrutiny from Washington. For investors, this means the two core growth engines that previously supported Nvidia's lofty valuation—pricing power in AI chips and exposure to the open-source market—are now both under simultaneous pressure.
Cembalest’s conclusion is that the shifting competitive landscape "is in some respects as important as changes in management, if not more so."
Leading Labs Vote with Their Feet, Nvidia’s Share Drops to 28%
Cembalest compiled data on major leading AI labs’ announced compute contracts for deployments after 2026, which total about 44 GW (gigawatts). Of these, Nvidia GPUs account for only 12.1 GW, representing a 28% share. The compute distribution is as follows:
Broadcom custom chips: 23%;
AMD: 18%;
Amazon Trainium: 16%;
Google TPU: 14%;
Cerebras: 2%.
This number stands in sharp contrast with Nvidia's dominant 80%-90% share of AI chip revenue from 2024 to 2025.
Among the seven major accelerators' purchasers, only SpaceX has yet to advance its in-house chip plans, while all other players are actively looking to bypass Nvidia’s gross margin as high as 75%.
In October 2025, OpenAI signed a multi-gigawatt compute deal with AMD. On that day, AMD shares posted their largest single-day gain in nearly a decade—an early signal of this trend.
CUDA Moat: Broken by AI in Three Days
Nvidia’s previous core competitive advantage lay in its CUDA software ecosystem—the engineering costs to migrate hardware platforms were high, making it difficult for customers to switch.
But Cembalest cites conversations with a leading lab indicating this advantage is fading:
A year ago, migrating models to AMD GPUs required 6 to 12 months, incurring high and complex engineering costs. Now, with the assistance of the latest AI models in engineering, migrating new models to AMD GPUs only takes three days.
In terms of performance, Epoch AI’s inference benchmarks show that AMD MI455X scores about 2.9 times higher than Nvidia GB300 overall, and about 2.2 times higher than Nvidia’s own Vera Rubin VR200.

Google TPU7x also outperforms Nvidia GB300 NVL72 in terms of compute per dollar, except at the lowest speed tier.
Cembalest concedes that Nvidia's Rubin Ultra "may change this situation upon its expected launch in 2027," but under current market dynamics, that is a long wait.
Open Source Community: From Core Market to Regulatory Target
As leading labs gradually diversify their procurement, Cembalest identifies the open-source weight community as "one of Nvidia’s core markets." This group consumes most inference compute on OpenRouter, and their models are rapidly catching up to the frontier of closed-source capabilities.
SemiAnalysis estimates that 1.5 GW of restricted Nvidia chips has already flowed into overseas markets, with "Nvidia being a key part of the supply chain, whether intentionally or not."
Cembalest also cited data showing that the proportion of Nvidia's revenue from Singapore addresses rose from 8% to 22% after Biden administration export controls were enacted, and Nvidia plans to stop disclosing this figure separately starting July 2025.
Goldman Sachs analyst Jim Schneider warned last Friday that Nvidia may face "more scrutiny in specific jurisdictions."
Cembalest quoted former Stratechery writer Andrew Sharp to pinpoint the core issue:
While Nvidia has enjoyed the benefits of the U.S. market, laws, and customers, it doesn't operate in a way that is accountable to the U.S. government as an American company should.
Valuation Is Cheap, But There’s a Catch
Cembalest is not predicting a crash.
From 2023 to 2025, Nvidia earned 83% of AI chip share, with a gross margin as high as 75%, and chip design cycles are long. He believes that for long-term holders, Nvidia—trading down 32% from its all-time high with a PEG of just 0.2x—remains attractive and is the lowest-valued among large AI stocks.
However, the 0.2x PEG is predicated on consensus expectations for growth being met, but both pillars of that growth are now wobbly: frontier labs are accelerating their purchases of competing chips, and open-source buyers now need licenses to continue procurement.
As Cembalest himself summarizes: the shift in the competitive landscape "is, in some respects, as much a risk for investors as changes in management."
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
Zoom (ZM.US) gets a bullish rating from Citi ahead of earnings: AI monetization signals emerge, target price raised to $131
Citibank has issued a forecast ahead of Zoom's Q3 financial report: Two major signals from channel partner surveys and Analyst Day both reinforce its view of a "moderate acceleration in the second half of the year."

Bridgewater's $23 Billion Portfolio Reshuffle: Drops Google (GOOGL.US) and Amazon (AMZN.US), Increases Holdings in AI Hardware and Energy
Bridgewater's Q3 portfolio was restructured, with a total market value of $23 billion. The fund significantly increased holdings in AI hardware and energy stocks such as Qualcomm and Micron, reduced positions in Google, Amazon, and S&P 500 ETF, and fully exited industrial stocks like Union Pacific, with a notable increase in the weighting of the healthcare sector.

180,000 People Petition, Musk Voices Support! But NVIDIA (NVDA.US) Board Reportedly Has No Intention to Reinstate Jensen Huang
Sources familiar with the matter revealed that Nvidia (NVDA.US) Board of Directors is currently not considering the possibility of him returning to the CEO position.

