Yeah, I'm concerned about the future of Arc now. If Celestial can legitimately hit 50% margin on it's own merits, it'll continue being a line. But what if it can't? Do they give up and drop out of GPUs again? If they just crank up the MSRP to non-competitive levels that'll kill it off too.
Let me propose a slightly less...depressing future.
Intel decides to take a page from the Nvidia and AMD playbook. They take their dGPU market, and subdivide its cost between the GPU market, the integrated graphics market, and a third market that we call parallel compute. The cost of products does not have to have a gross margin of 50%, but the products that they deliver to the market do. That means competing with the like of Nividia that can slap "AI accelerator" onto anything and sell it for premium levels of profits, and your gross margin is based off of the price of the licensing for the silicon production and design...which can be a relatively low percentage of the overall cost of the final packaged product. IE, owning the IP and manufacturing the processor on both a consumer GPU, the GPU core on an integrated GPU, and a surprisingly high margin (but awfully low actual sales figure) accelerator card that makes the whole shebang work as a cumulative 50% gross margin model.
Let me do that for those who don't speak sales fluently. Let's say that Intel projects that they want a GPU division. Their sales people state that the GPUs will require 50 million to develop and their annual sales are 1 billion , and that they subsequent sales will be 200 million units. Those units are broken down into 50% co-processors for CPUs, 20% GPUs, and the 30% other processors. The sales people project that the iGPUs only make 30% margin, the GPUs can make 50% margin, and the other processors make 70% gross margins. Their total sales then would be (0.5*.3)+(0.2*.5)+(0.3*.7) = 0.46 margin. ((1000-50)*(.46))/1000 = 43.7% gross margins... which means a dead GPU division.
Now...let's do some sales math. The 50-20-30 split isn't fair. That's what sales may look like, but imagine if that 30% at 70% margins was inflated. If it instead was 50% of the projected sales instead of 30% because it suddenly cost much more...while remaining the same actual sales figures, and it can now easily be a 50% gross margin...because all of a sudden the higher margin product represents more of the projection because of a sales number fabricated by something like the existence of Nvidia GPU scalping in the Chinese market. This launches the product, because the projections match the 50% gross margin requirement...but three years down the line the real gross margins are lower. Welcome to the discrepancy between projections and reality, which everyone buries because they all know that a 50% gross margin requirement is bat crap crazy of a requirement and will price you out of any market where the competition can provide even a slightly reasonable alternative.
TL;DR
Sales forecasting is the fine art of lying and making the numbers say what you want with enough plausible deniability that whenever reality differs nobody wants to question because they have been a part of the lie too long. This is how we get crap like Harley Davidson, who projected huge savings by moving production over seas...and got sales cratering because they forgot that their premium customers were only willing to pay for a premium if it came packaged with an identity that "made in China" could not support.
This won't end anything but the craziest projects...because a good sales person can explain exactly how much swiss cheese the moon is made out of. If they're great they can do this to a NASA scientist to help justify the cost of the next moon mission.
