Nvidia’s robust 2Q earnings report might have raised more questions than it answered about Blackwell GPU production worries, as it left many observers wondering “What the heck is a mask?”
CEO Jensen Huang was pretty blunt in explaining what has happened to Blackwell in recent months of production at a TSMC foundry. “The change to the mask is complete. There were no functional changes necessary,” he said in response to an analyst question.
Nvidia is sampling functional samples of Blackwell to customers, he said, along with Grace Blackwell and various other system configurations. Using a classic Bill Belichick line, Huang added, “the functionality of Blackwell is as it is, and we expect to start production in Q4.” Later, he added, customer “anticipation for Blackwell is incredible.”
Nvidia CFO Colette Kress issued a written commentary during the earnings report, including a quick discussion on Blackwell specifically:
“We shipped customer samples of our Blackwell architecture in the second quarter. We executed a change to the Blackwell GPU mask to improve production yield. Blackwell production ramp is scheduled to begin in the fourth quarter and continue into fiscal 2026. In the fourth quarter, we expect to ship several billion dollars in Blackwell revenue.”
To unmask the mystery of what is a mask: In the chip trade, a mask is a glass sheet or fused silica plate that works as a template to print circuits onto a silicon wafer during the photolithography process in making integrated circuits. And here’s more than you wanted to know: Said mask is often 6 inches
square and has a pattern of transparent and opaque phase-shifting areas, which were created by using computerized drawings from chip designers like people at Nvidia. The drawings are fed into a machine that converts the data into an electron beam, which scans the mask surface to print the circuit pattern.
Since TSMC does the chip packaging for Nvidia and not the design, it seems evident the mask change was done by both parties. Analysts had said earlier that part of the concern about Blackwell production came down to both the Nvidia design for Blackwell and the TSMC process for CoWoS L (Chip-on-Wafer-on-Substrate with Local Silicon Interconnect) used for the first time in high volume. It allows 3D packaging to stack chips onto a substrate to reduce the space needed for chips and reduce power consumption and therefore cost.
RELATED: Here’s what’s wrong with Nvidia’s Blackwell GPU
Analysts don’t seem too worried about whether the mask change is enough to affect the production schedule for Q4 or whether other production concerns could develop. Chip production is immensely complex and expensive and it is frankly a wonder chips get made at all.
But maybe there was a little worry about the mask change on the minds of some as Nvidia’s stock dropped 8% after the earnings call and then went down 3.7% by mid-day Thursday, reaching $120.93. (Still, the stock has been up about 150% for the year.) Some speculated Nvidia’s fantastic success with GPUs and AI has slowed a bit, including a slowing of quarter-to-quarter revenue gains, from 80% more than a year ago to 15% now. Others said the stock hit could be blamed on a less-than-favorable outlook for the current quarter. Nvidia forecast $32.5 billion for the current quarter, but some analysts expectations have been much higher.
RELATED: Nvidia breaks revenue record, catches flack for outlook
Overall 2Q revenues were $30 billion, a record, up 15% from the previous quarter, but up 122% year over year. Data center revenue was $26.2 billion, up 154% year over year. Of that data center total, $3.6 billion came from networking products, while the rest was for compute chips, like GPUs including Hopper chips. Nearly half of data center revenues come from hyperscalers building out enormous data centers to run GenAI and other compute, primarily for training algorithms.
So back to the mask change. Patrick Moorhead, founder and principal analyst at Moor Insights & Strategy, suggested on CNBC that some of the share decline could have been due to the news of the mask change, but the important point Huang made was that no functional changes were made. A mask change can be seen as making the product more efficient to manufacture, one report from CNBC added.
“Changing a mask is not really that big of a deal,” added Jack Gold, principal analyst at J. Gold Associates in comments to Fierce Electronics. “It’s basically changing the pattern that you use to do a screen print on the wafer. It does take some time to make the change but it’s not that dramatic. The bigger issue is making sure the mask doesn’t have flaws that will cause a manufacturing issue or a chip functionality issue.”
Gold said the lowered stock value could be an emotional reaction to perceived issues. “The market was expecting even more than the impressive results Nvidia offered. And, they have slipped Blackwell, and there might be some expectations that the slip will negatively impact Nvidia revenues. To some extent it might, but it will likely be minor as long as it’s only a one quarter slip. They’ll just keep selling Hopper class devices in the interim. But if Blackwell slips out again, it might be a bigger deal.”
Moorhead added in email to Fierce: “Nvidia never made specific forecasts on Blackwell, so we really don't know the impact of the delay. Mask changes take time and undoubtedly cost the company months. The good news is that the [change] wasn't for incompatibility or performance reasons, but yield improvements that increase the output and lower the costs.”
Dylan Patel, an analyst at SemiAnalysis, told Fierce he heard nothing new from the Nvidia earnings report about Blackwell production, having reported a mask change and delay in early August. SemiAnalysis said earlier that initial Blackwell production would have started mid-3Q for B100 chip for servers, but has now been pushed back to mid 4Q in low volume for the B200 variety used in rack servers. That’s effectively a three month delay that appears to have been partly confirmed by Nvidia.