One of the key debates in the artificial intelligence (AI) industry right now is whether or not Nvidia (NVDA 6.18%) will hang onto its first-mover advantage and dominate the AI space for years to come. Given that Nvidia's stock has more than tripled this year and trades at over 111 times earnings, that's a key question among semiconductor investors heading into earnings season.
Nvidia's would-be competitors have made some progress announcements on their AI roadmaps, including Advanced Micro Devices (AMD 2.37%) announcing its MI300 accelerator chip in June, as well as Intel's (INTC -9.20%) promotion of its new Gaudi AI accelerators. And each is enlisting large cloud companies to adopt a more open-source software approach as an alternative to Nvidia's expensive closed-source, vertically integrated CUDA software ecosystem.
But did anyone think Nvidia wouldn't respond in kind? CEO Jensen Huang just published a presentation that laid down a new gauntlet on these would-be challengers.
Moving from a two-year to a one-year innovation cadence
In the past, Nvidia would operate on a two-year cadence, unveiling a new chip architecture every two years. This is what we have seen with the introduction of the H100 data center graphics processing units (GPUs) late last year, two years after the release of the former model, the A100 in 2020.
However, in the roadmap on Nvidia's latest presentation, management now projects a new architecture across a range of its data center chips every single year:
Image source: Nvidia.
As you can see, next year Nvidia will introduce the H200, which is likely a modification of the existing H100 architecture. But in that same year of 2024, Nvidia also plans to move ahead with a new architecture in the B100 chip, code-named "Blackwell." And that new architecture will be followed the following year with the X100 architecture in 2025, which doesn't have a code name yet.
As you can also see, in tandem with the new chip architectures, Nvidia will be updating its entire ecosystem, including the Grace Hopper "superchips" that combine Nvidia GPUs with its new Arm (ARM 4.11%)–based central processing units (CPUs), the enterprise inference-focused L40S GPUs, and its InfiniBand and ethernet-networking systems in combination with the new chip-architecture introductions.
It's not clear what the Grace Hopper NVL chip is, only that it also appears to be a new kind of chip-form factor. Apparently, it will also follow the one-year cadence.
Leaving competitors in the dust or a defensive necessity?
No doubt, this roadmap is quite ambitious. It's not easy to design these chips and get the manufacturing capabilities at Nvidia's foundry partners up to speed. And this accelerated pace of innovation will no doubt be expensive.
There are also a couple of ways for investors to interpret the new cadence. On the positive side, this could be Nvidia leveraging its early lead -- and the robust profits and cash flows that come with that lead -- to outspend its competition, which currently has fewer resources. Nvidia's profits have skyrocketed this year, with the company earning $8.2 billion in net income for the first half of the year and likely on its way to earning $20 billion to $25 billion for its fiscal 2024 year.
In contrast, Advanced Micro Devices and Intel have actually recorded net losses through the first six months of the year, as each has heavy exposure to the PC industry as their main cash cow. So, this could be a case of Nvidia really turning the screws on would-be competitors, which currently lack the resources to innovate at this hyperspeed. These include both Intel and AMD, and also the large cloud companies that are beginning to produce custom in-house accelerators. While the cloud companies definitely have the financial resources, chipmaking isn't their primary business.
On the other hand, a negative interpretation could be that Nvidia may think Intel, AMD, and the cloud companies will develop a strong open-source software alternative to CUDA, and that the software moat CUDA has afforded Nvidia until now may be breached at some point.
If these other companies succeed in developing a strong open-source alternative, Nvidia may need to resort to out-innovating competitors on the hardware front. Hence, the new one-year cadence. While that could be a recipe for long-term superiority, it may also be more capital-intensive, and a rapid pace of hardware innovation isn't as strong of a moat but rather one that requires continued execution.
A strong counter from Nvidia
After Nvidia's meteoric rise this year, challengers answered with their own accelerators and roadmaps this summer. But it appears Nvidia just one-upped the competition again with its stepped-up cadence of innovation.
The AI wars are intensifying, and it will certainly be interesting to see what new innovations come out of it next.
