In Q1 2026, they said Q2 would miss 8B in sales to China but guided at 45B. So if they didn’t have to stop sales to China they would have guided 53B for Q2. Now that they switched their production to not focus on China they are guiding 54B for Q3. That seems like a dramatic slowdown to me.
The first 2 sentences … completely true. The rest… huh? They are guiding Q3 to $54B factoring in no China sales. You are adjusting one side of this comparison around that ban but not the other … why aren’t you adding back the missed China sales to the Q3 guide if you are making this comparison? You are padding one side of the equation and not the other, and saying it’s evidence of a slowdown.
For H20s to China: They could have sold $7.1B of in Q1 (only did $4.6B, were blocked on the final $2.5B), and could have sold $8B in Q2. Let’s extrapolate that out on the same seq growth and say they could have sold $9B in Q3 (+12.67% seq).
That would make the $54B guide into $63B for an apples to apple comparison. That’d be an +18.8% seq growth had the ban not happened, instead of the quite impressive in its own right +15.5% seq growth they just guided with the ban in place.
Of course, this is a way more nuanced discussion than this as the capacity for H20s is out of HOPPER capacity. They have been switching over to as much BLACKWELL capacity as they can get, a diff packaging process that TMSC is trying to increase capacity of as fast as possible. Beyond that, Blackwell is currently being sold as racks (NVL72) not blades (HGX8), and so is a much more complicated process and series of assembly. Yet they kept some existing Hopper lines running for a trickle of supply that I think was to keep H20s around plus to continue to satisfy a long tail of Hopper demand.
Note how mgmt said Hopper sales increased this quarter. This is a sign they did repurpose some of that supply chain after the ban, to increase the supply of Hopper in the meantime. Yet they also sold $650M H20s to a non-China customer, showing H20s lines are still active (some of it was past inventory written down, that that means ~ 2/3rds was freshly made inventory).
They clearly want to keep Chinese companies happy by keeping the H20s available and a supply chain for more at the ready … at least until they can get a China-specific Blackwell variety approved and made. Of course, geopolitics isn’t letting that happen as 2 major governments are huffing and bluffing, and no customers are buying yet since the ban (and it all goes way beyond AI chips … rare earth magnets and airplane supplies are parts of the trade threats lately too). But if that doesn’t happen, they can easily repurpose it back to Hopper and fill some of that long tail of demand, even as Blackwell supply continues to increase.
Like the others chiming in (Smorg, StockNovice), I do not agree with nor really understand your argument about some mysterious new signs of supply constraints showing up over the past 2Qs. You are throwing up numbers (rather incorrectly, as noted above), and drawing conclusions that are not there.
Of course there are supply constraints… but those constraints have been improving over the past 6 quarters across Hopper and now Blackwell – and management insists that Blackwell and now Ultra continues to ramp up in supply over the next 2Qs as mgmt just noted. From here, Rubin is on target and we’ll see it appear in 2026, and we’ll likely see this same transition from Blackwell/Rubin occur.
If you add ~80% of that missed $8B forward, it lines up closely with their Q3 revenue guidance. To me, this suggests NVDA produced chips that ended up unsellable (because export controls on and off for China) instead of producing other chips they could have sold. That points to constraints on production capacity.
Again… huh? Yes they made chips they couldn’t sell, because of the ban! And yes, $8B is their estimate of what they could sell without the ban. How does this correlate to supply chain in any way? I don’t get it why they are being conflated and how any of the numbers used above are evidence of any new supply chain woes that are worsening. I mean, HBM memory is sold out well into 2026 at this point! Same with TSMC CoWoS capacity! NVIDIA has likely over a year of vision into their supply chain at this point. They know how much they can get, how much customers are committing to, and as StockNovice pointed out, the latter of that continues to be way higher than the former (supply > demand).
As Saul says, you sell when the story changes. For me, the story here has changed—Nvidia still has incredible demand, but its growth is bottlenecked by issues outside of its control.
You are taking what has been a lingering concern for 2+ years now around NVIDIA (supply chain constraints, which has constantly improving) and saying it is now some new thing that just appeared in the last quarter or two, and are presenting the China ban impacts as the evidence. There are plenty of reasons why one might exit here (China sales completely gone, rise of domestic and now Chinese competition, “AI is a bubble”, ever present threat of a Taiwan invasion, the most valuable company in the market, etc) …. but some newly emerging supply chain bottlenecks??? Sorry, I don’t get it.
-muji
