NVIDIA is reshaping its business model as pressure builds from hyperscalers seeking to reduce reliance on the company’s AI chips. In a recent post, technology investor and researcher Evergreen Capital said CEO Jensen Huang’s strategy appears aimed at diluting NVIDIA’s revenue dependence on hyperscalers through direct capital allocation moves. After roughly three months of observation, Evergreen concluded that if the transition unfolds as expected, NVIDIA could see a meaningful valuation re-rating.
Why Evergreen focused on one word from the earnings call
Evergreen Capital said its original view dates back to NVIDIA’s earnings release in May. The most important signal, in its view, was not the strength of the reported numbers but the language Huang used on the earnings call.
According to Evergreen, Huang used the word “diverse” 12 times to describe NVIDIA’s customer base, more than the combined total across the previous 12 quarters. Evergreen read that as a strategic declaration. In that framing, NVIDIA is trying to reposition itself from a chip supplier dependent on hyperscaler orders into an AI systems platform provider serving a broad and varied set of customers.
That shift, Evergreen argued, carries an implicit acknowledgment: NVIDIA does not fully control its market share inside the hyperscaler segment.
The “de-NVIDIA” threat still hangs over the stock
Using Porter’s Five Forces as a reference point, Evergreen said hyperscalers, as NVIDIA’s most important buyer group, have a built-in incentive to diversify suppliers and reduce dependence on a single chip source. Whether the goal is lower procurement costs, less supply-chain risk, or stronger bargaining power, developing in-house AI chips remains a rational commercial choice for these companies.
Evergreen wrote, “The ‘de-NVIDIA’ narrative has weighed on NVIDIA’s valuation for a long time.” Even if NVIDIA’s hardware and software stack still leads across nearly all workloads, bears can always point to some future point and argue that the moat will eventually break down. Evergreen’s view is that this is not a debate Huang can win in the narrative arena, and that he appears to understand there is little value in continuing to fight it head-on.
That said, Evergreen did not turn negative on NVIDIA’s near-term hyperscaler business. It said NVIDIA systems still offer the best economics for hyperscalers on tokens per watt and tokens per dollar per watt when measured on total cost of ownership, or TCO. It also pointed to NVIDIA’s distinctive procurement strength in a supply-constrained environment.
Still, Evergreen said NVIDIA cannot control hyperscalers’ long-term willingness to buy, and that risk remains.
SPCX and GPU financing as balance-sheet tools
About three months later, Evergreen said in a newer post that its earlier thesis is now being validated more quickly. It pointed to two concrete steps from NVIDIA: SPCX transactions and a GPU financing program. In Evergreen’s reading, both are meant to reduce revenue concentration tied to hyperscalers while expanding NVIDIA’s reach into the non-hyperscaler enterprise AI compute market.
The GPU financing program stood out in particular. Evergreen said it shows NVIDIA is willing to use its own balance sheet to help smaller enterprises and emerging AI companies secure financing for access to GPU compute resources.
Evergreen acknowledged that some investors may worry about balance-sheet risk. Its answer was that the risk should be relatively manageable because NVIDIA GPU systems have unusually strong residual value. “As the ROIC on token compute continues to rise, the collateral value support behind these systems is actually moving higher,” Evergreen wrote.
A revenue mix shift could change the valuation story
Evergreen offered a clear numerical view. It said the non-hyperscaler enterprise AI compute segment, described as AICE, currently makes up about 50% of NVIDIA’s total revenue. Analysts, it added, expect that share to climb above 70% over the next several years.
For Evergreen, that change matters because NVIDIA’s customer base would be less concentrated in a handful of giant cloud buyers and more distributed across sovereign AI projects, midsize companies, startups, and research institutions. If that happens, a cut in spending by any single hyperscaler would carry less weight.
Evergreen also argued that these non-hyperscaler customers are far less capable of building their own AI chips. Their dependence on NVIDIA’s full stack — including the CUDA ecosystem, systems integration capabilities, and connections with third-party platforms — is higher than it is for hyperscalers. That, in Evergreen’s view, would give NVIDIA stronger pricing power and stickier customer relationships in those markets.
If the transition arrives on schedule, Evergreen said NVIDIA may come to be viewed less as a hardware supplier tied to a small number of outsized customers and more as an AI platform company with a broader customer mix and higher retention. That, it argued, could support a different earnings multiple and a larger repricing window for the stock.
From defending share to building a new market story
Evergreen’s broader argument is that Huang is shifting NVIDIA’s narrative center of gravity away from defending hyperscaler share and toward attacking the non-hyperscaler markets the company can influence more directly.
Evergreen also said the transition is still in progress, and the market will need time to absorb and verify it.