I actually got a lot more upbeat about the potential of AI after watching this keynote more so than any other demonstration or presentation.
Blackwell is still based on N4, a derivative of 5nm. We know we will have N3 GPU next year, and they should be working with TSMC on capacity planning for N2. Currently Blackwell is pretty much limited by capacity of either TSMC, HBM or packaging. And unless the AI hype dies off soon ( which I doubt will happen ), we should still have at least another 5 years of these GPU improvements. We will have another 5 - 10x performance increase.
And they have foreshadowed their PC entry with MediaTek Partnership. ( I wonder why they dont just acquired Mediatek ) and may be even Smartphone or Tablet with Geforce GPU IP.
My guess is mediatek’s margins are much lower so an acquisition would tank Nvidia’s stock price by leading to lower returns. That and/or not wanting the distraction of operating mediatek’s many businesses that aren’t aligned to Nvidia core competence.
A major chip acquisition by Nvidia will probably never get global regulatory approval. I guess the question is what does Mediatek provide in the partnership. Nvidia is well versed in developing ARM SoCs already.
Market access, shares and relationship to all vendors, most of them never worked with Nvidia. Ready made 5G solutions. Know how in product category that is not as simple as buying IP and making a chip. Shipping Nvidia CUDA GPU IP to all other domains. For acquisition benefits they share the development cost of leading node and volume. Which is part of the wider SemiConductor industry strategy aka Broadcom.
But yes. I guess China would be the first one against it. I never quite understand why Two Company in country A and B would need Country C 's approval for them to merge or be acquired.
Country C's regulators can sanction the companies entities in C including denying access to the market in C. (C.f. how the US got Swiss banks to comply with financial transparency laws.)
It would not matter if C == Lichtenstein, but China is currently the world's biggest semi-conductor market ...
yeah, those are just modules that anybody can just buy. NVIDIA already makes jetsons which are basically SBCs with decent GPU. so again, what does Mediatek provide?
What do you mean by modules, exactly? In a smartphone, which is the largest market, and where Mediatek is pretty strong, many of those functions are integrated as part of a system on chip. Power consumption also required good interplay between the various subsystems. So it's relatively tightly integrated. You also must have good Android drivers/BSP. And phone vendors are not that interested in changing vendors alle the time. There are also a bunch of patents that one must license, which Mediatek likely already does. Often cross licensing with what is typically competitors (like Qualcomm, maybe even Apple), which takes a bunch of time to set up.
looking back at the news releases: Mediatek seems to bring CPU design expertise on the table, which makes sense, they have much more experience and know-how in this than Nvidia. Mediatek is a top tier company with excellent engineering talent. Ideally Nvidia would acquire such a company so that they can challenge AMD and Qualcomm better but this doesn't seem to fit for their strategy to keep it small and focused. Hemce such a partnership works better for them.
I was taking these sort of improvements for granted. They'll certainly make ai cheaper, but most of the cost is up-front for generative models, and the models don't seem to be getting linearly better.
>I was taking these sort of improvements for granted.
A lot of people would think so. In reality had Smartphone ( iPhone ) not taken off. The PC market alone would not be able to sustain the cost of innovation and Moore's Law would have ended at 14nm and we would have to wait 3 to 4 years for every node. Smartphone along with all the adjacent work on Internet infrastructure scaled us to 3nm. I wrote about both of these in 2014 and was expecting some slow down by 3nm. Now AI will provide the Capex to scale us to 1nm and beyond.
The flagship is an outlier there since it went from 384bit to 512bit memory. There was no such bus width increase for the rest of the stack so the gains are more modest, 4080 to 5080 is only a 34% improvement.
the solution is people will figure how to burn their models onto ASICs cheaply. apple model on iphone, google model on android, etc. This is tantalizing to businesses (1) you have YOUR company's model on lockdown as you see fit (2) iterations or improvements to the model mean you'll need to buy buy buy
The acquisition of ARM would have put Nvidia in the position of being able to impose to competitors uncompetitive restrictions on a quasi monopolistic market. Is it the case with MediaTek?
Mediatek is the largest smartphone SoC manufacturer, so yes. They're a supplier for Apple and the biggest competitor to Qualcomm, the two companies that were the most vocal in opposing the ARM acquisition.
Is GPU improvement is driven more by gaming then by AI hype? Gaming is the biggest industry and there is real money coming from that. Does speculative money from VCs actually overshadow actual money spent by consumers?
I know stock prices is driven by AI hype but how much does it actually effect the bottom line of Nvidia? I think GPU improvement happens regardless of AI.
Datacenter revenue alone is ~10x of gaming. The datacenter revenue is thought to have literally ~100x the earnings all up (H100 and 4090 have similar transistor counts but the H100 sells for over $30k while the 4090 sells for $2k which indicates huge margins).
Gaming is pretty much insignificant for nvidia. That’s why nvidias stock has 10x’ed recently and their PE looks better now than it did 5 years ago despite that stock increase. They found a new market that dwarfs their old market.
NVIDIA’s net income grew ~580% year-on-year in their 2024 fiscal year. FY2025 is on track for 100%+ growth, essentially 14x in the last 2 years. This is not coming from gaming, “AI hype” is having a huge effect on NVIDIA’s bottom line.
It all depends on whether AI companies can continue to find significant improvements to their models this year. Are transformers reaching their limits? Can researchers find the next level of performance or are we headed for another AI slump?
Interpreting your question about "GPU improvement" from a product perspective, my read is that NVIDIA is of course targeting AI applications and the datacenter. To that end it just focuses on silicon that makes most sense for AI compute, and not so much for gaming.
Of course, the GPUs for the datacenter and for gaming are the same designs, so my read is that in gaming NVIDIA makes up for lack of actual performance for traditional rendering by pushing technologies that can utilize tensor cores like AI upscaling, frame prediction, ray tracing + denoising, etc.., that don't actually contribute to game graphics as much as they could have if they did an architecture tailored to gaming needs instead, with the technologies that they have. It's also sexier in theory to talk about exclusive AI-powered technologies proprietary to NVIDIA than just better performance.
"NVIDIA’s financial report reveals that its data center business revenue in FY2Q25 grew by 154% YoY, outpacing other segments and raising its contribution to total revenue to nearly 88%."
Gaming almost doesn't even register in Nvidias revenue anymore.
But I do think Jensen is smart enough to not drop gaming completely, he knows the AI hype might come and go and competitors might finally scrounge up some working SDKs for the other platforms.
Gaming is effectively irrelevant to nvidia now. The stock appreciation over the last 8 years that brought them from a niche to a global dominant company is all from revenue that first came in from crypto and then got absolutely dwarfed by AI.
ML was big before LLMs and nVidia was already making a killing from selling expensive GPUs that would never draw a single polygon ca 2015. They've been hobbling FP64 (double precision) support in cheaper "consumer" GPUs, to prevent their use in most data centers, for a long time too.
Looking into this a bit, it seems that nVidia was still making most of its money from "gaming" as recently as early 2022. I'd suspect that, if crypto mining could be separated out, the transition point actually happened a couple of years earlier, but nevertheless the datacenter segment didn't become dominant until about 2-2.5 years ago. It existed well before then, but wasn't that big.
Blackwell is still based on N4, a derivative of 5nm. We know we will have N3 GPU next year, and they should be working with TSMC on capacity planning for N2. Currently Blackwell is pretty much limited by capacity of either TSMC, HBM or packaging. And unless the AI hype dies off soon ( which I doubt will happen ), we should still have at least another 5 years of these GPU improvements. We will have another 5 - 10x performance increase.
And they have foreshadowed their PC entry with MediaTek Partnership. ( I wonder why they dont just acquired Mediatek ) and may be even Smartphone or Tablet with Geforce GPU IP.
The future is exciting.