In April, the U.S. government imposed new export controls prohibiting the shipment of high-bandwidth memory (HBM) GPUs to China, effectively sidelining Nvidia’s H20 chip. As a result, Nvidia incurred a one-time $5.5B write-off of inventory in the April quarter, and Jensen suggested they’ll lose about $15B in sales this year. Deepwater estimates that about 25% of Nvidia’s FY25 revenue, or $50B in CY25, was effectively China-related. That means that about $20B has been going through authorized channels of H20 sales, and another $30B has been going to third parties, many stationed outside of China, that supply China with Nvidia GPUs. In other words, putting curbs on H20 only partially fixes the problem of China’s access to the most advanced GPUs.

To recapture this revenue segment, the company is reportedly fast-tracking a “Blackwell-lite” chip that could ship as soon as June. Each quarter that “lite” chip is in the market, would add back about $5B in revenue.

Some specifics: For the July guide the buy-side models have already marked revenue estimates down to roughly $38B, down $7B vs. the imprint estimates of $45B, to reflect the U.S. curbs.

The key on the call will be any commentary from the company that suggests, outside of the impact from the China curbs, the demand from non-China customers is exceeding expectations. In other words, management doesn’t have to guide the July quarter higher than expectations, but rather show that the dip is temporary and tied to export paperwork, not end-market weakness. The long-shot positive commentary would be any suggesting that the curbs eventually will ease as part of a broader U.S.–China trade deal. My sense is that’s in the works, but outside of anything the company will want to speculate on.

Even if China temporarily pauses, most other data points suggest the AI infrastructure spending ramp is intact. xAI’s ‘Colossus’ data center is reportedly targeting roughly one million Blackwell GPUs, which would account for about $30–$40 billion worth of Nvidia hardware over the next two years. If Musk comes through with those orders, xAI alone would account for almost 8% of revenue over the next two years. Additionally, OpenAI has publicly commented that it is capacity-constrained and needs more GPUs. Orbiting the tech capex buildout conversation is what Apple plans to do to close the AI gap. It’s becoming increasingly clear the company needs to step up investment from its current $5B AI run rate. I believe we will hear more about Apple’s increased AI investment on the June quarter earnings call, which could add a new 2–4% customer to Nvidia’s book. Any way you cut it, it’s clear that hyperscale build-outs are still in the very early innings.

The Street calls for 56% revenue growth in CY25 and just 26% in CY26, but hyperscaler capex plans suggest CY26 growth could accelerate, not decelerate. In other words, whatever tomorrow’s headline number, structural demand for accelerated compute remains intact.

Our thesis is simple: owning Nvidia remains the cleanest way to invest in the AI buildout, which despite all of the hype remains in its infancy. Street revenue expectations for CY26 are roughly $245B, but we believe they are 10% too low, implying Nvidia trades closer to 20x forward EPS once the upward revisions flow through. If management reiterates that late in CY25  revenue will re-accelerate as additional Blackwell supply hits the channel, then multiple on NVDA has room to expand.

Disclaimer