Analytically I believe Nvidia component growth rate is doubling doing for the first time in 30 years what Cyrix and AMD should have done. In Cyrix’s case JV exclusively with TI and then missed a second time with IBM. And in the AMD example JV with Global Foundry 2015/16. That opportunity was certainly on the table? There would have been no supply questions as there are now none discernable on Nvidia Samsung teaming. Samsung for decades needed high margin volume sales as a price margin support against high volume commodity fabrication and now has it. As long as everyone gets along Nvidia supply is unquestionably unrestrained and unlimited.

Back to those basic GPU tenants; bottom bin like Polaris appear to sell for around $43 to $47 a pop think Pentium OEM pricing.

AMD average let’s call it competitive price sell for around $80 and Nvidia gets a premium lets call this top of i3 bottom of i5.

Last two quarters of Nvidia miner sales were not necessarily bottom bin on those SKUs available, many secured from secondary market brokers clearing channel inventory which is as lean as its been in three years. For into future CMP _HX guestimate; $150,000,000 / $80 = 1,875,000 components and on Nvidia q2 guidance $400,000,000 / $80 assuming this is essentially slack from the dice bank = 5,000,000 components. On this method of estimation Nvidia is assuring against commercial mining taking a third of Nvidia quarterly consumer production. Pascal 1060 in 2017 went to mining I’ve estimated 10 M cards as a percent of GTX 10×0 full run. Commercially, the segment I place mining, might have snapped up another 10 M cards which would be equivalent to around one quarter of Nvidia consumer production that currently seems in the 11 M and definitively heading toward 15 M units per quarter range.

Here’s Nvidia component category break down on channel inventory holding data;

Tesla between Accelerators and tip top of science cards = 7.27%. Note that commercial data is somewhat hidden on channel inventory levels that are cards offered by open market broker dealers.

Professional Visualization = 12.18%
Top Shelf Consumer = 71.67%
Mass Market Consumer = 6.75%
Consumer Mobile = 2.13%

So how many and what kind of Nvidia components sold for $5.3 Billion in q1? If we rely on Mr. Huang 100 M units annual the first thing one might notice is average net jumped up from $166.75 to $212 per unit but it’s difficult factoring in developer appliances, compute systems and embedded subsystems.

How about just looking at $2.760 B gaming, $372 M professional visualization and $2.048 B data center revenue in q1 2021 (Nvidia q1 ’22).

Whether AMD aim or Nvidia ability to secure, $180 appears the average net sought for consumer dGPU; $2.760 / $180 = 15,333,333 components. 60 M units annual is approximately the historical top of Nvidia consumer line production.

Professional visualization at $372 M difficult to say on the question of Nvidia professional card sales combined with appliance development and compute system inclusion. Or is it data center that includes appliance and compute systems? Let’s take a guess. Intel likes double Core line OEM average or what would be the equivalent of 1K. lets use Comet Lake 1K across full line $339 and multiple times 3 or 4 for a competitive card product = $1019 to $1395 would be in the consumer MSRP range for AMD Instinct and professional and Nvidia has to be competitive; range 243,844 to 301,275 cards presents a low volume anomaly in relation percent of channel inventory holdings.

Now the hard one, data center and this will be for large memory accelerator. Instinct M125 goes in the channel for $1000 and Nvidia cards currently appear price inflated Tesla A100 developer reference cards in the channel range $11 K to $15 K. Let’s consider RTX A6000 selling for $5500 in the channel / 2 wholesale = $2750. Now add that data center is a commercial sale and data centers have price making power. Let’s double prior $339 dGPU component competitive price to $678 that is Intel full product line economic profit point continuing the thesis all design producers attempt to emulate Intel margin. $678 x 3 for the subsystem wrapper = $2034.
Nvidia q1 data center at 2.048 B / $2034 = 1,006,883 cards and or subsystems.

Finally lets break Nvidia sales aim out by product category on channel data across 120 M units through the year;

Commercial Accelerators = 6,615,246 dGPU

Top of Science = 2,106,215 dGPU

Professional Visualization is the ‘high’ anomaly on this pondering = 14,615,070 units.

Top Shelf Consumer = 86,006,392 cards also thought extending into mobile on 30×0 Max-q.

Mass Market Consumer = 8,098,668 cards.

Mobile as in MX and Max-q and yes it’s back looking and seems low but channel data is good for this = 2,558,410 units.

Mike Bruzzone, Camp Marketing