On Monday, Microsoft (MSFT -1.80%) unveiled its latest AI chip architecture, the Maia 300.
And the company went out of its way to point out the ways in which the Maia 300 was superior to graphics processing units (GPUs). "This is our biggest bet yet on our own silicon," said CEO Satya Nadella.
This was seen as a shot across the bow to the world's premier GPU maker, Nvidia (NVDA -3.16%).
So is the Maia 300 the "Nvidia killer" that everyone has been anticipating for years? It might be, but not in the way you might think. Here's what's really going on.
Image source: The Motley Fool.
Cutting the cost of AI
The Maia 300 is an upgraded version of the Maia 200, which was released in January.
The Maia 200 is currently in production and has already been deployed at Microsoft facilities in Iowa and Arizona, where it's used instead of GPUs and other components from Nvidia, Intel (INTC -4.56%), and Advanced Micro Devices (AMD -5.73%).
Microsoft claims the new, upgraded Maia 300 chip is even more economical and can process 40% more tokens per megawatt when compared to GPUs. A higher token-per-megawatt ratio results in lower AI costs because the biggest expense for an AI data center is electricity.
But even if it's cheaper to operate, it's certainly not going to match the power of Nvidia's top-of-the-line systems. Won't that be an issue?
It's not always about being first
Perhaps more than any other company, Microsoft has done very well by coming in second. Its Azure cloud platform is the second-largest, compared to Amazon's Amazon Web Services. Its Surface tablets have never outsold Apple's iPad. Its Xbox has always played second fiddle to the Sony PlayStation.
So it shouldn't surprise anyone that Microsoft isn't trying to beat Nvidia at its own game, but to compete on a more favorable playing field: cost. "Moving data efficiently, rather than raw compute power, is increasingly the bottleneck in large inference systems," Microsoft said in its blog post announcing the Maia 300 system. The company hopes to leverage this to its advantage.
Image source: Getty Images.
Austin Lyons, an analyst for Chipstrat, thinks there's an opportunity here in optimizing different AI processors for different tasks. "The advantage of custom silicon is that Microsoft can optimize for the latency its own workloads actually require," he wrote. "A meeting summary and a coding agent don't need the same response time. Maia targets more throughput at each of those latency points, translates to more useful tokens served per megawatt and lower cost passed for customers."
Pat Moorhead of Moor Insights & Strategy agrees, writing:
Microsoft is now moving toward deploying Maia at meaningful production scale. At that scale, custom silicon isn't just about performance. It can become an important margin driver by improving the economics of delivering Microsoft's own AI services.
The AI race is changing
Lyons and Moorhead are echoing commentary from across the industry that suggests the AI race is evolving away from a zero-sum contest to create the fastest, most powerful processors. Hyperscalers and other major players are starting to think about the economics of AI and what customers will be willing to pay in an era where the tech giants start to monetize the new technology.
By running lower-difficulty, in-house AI queries on processors that operate more cheaply, Microsoft is taking another step in proactive cost management, which is one of the key reasons investors rewarded the stock earlier this year.
NASDAQ: MSFT
Key Data Points
So while the Maia 300 isn't an Nvidia killer on its own, it's evidence that the most powerful AI processors -- currently Nvidia's GPUs -- may not be everyone's universal first choice in the future for every application. Routing an AI query to the cheapest infrastructure possible for the difficulty of the problem presented may become the norm.
In other words, even if Nvidia continues to produce the best tech in the world for decades to come, its sales volume might suffer regardless.
Nvidia will still be in the mix
The Maia 300 doesn't even begin production until sometime in 2027. And even then, Nvidia AI infrastructure is almost certain to remain a critical part of Microsoft's AI mix.
Just last week, Microsoft started taking preorders for a new Surface Laptop Ultra with a starting price of $2,599. That hefty price tag is thanks to its powerful processor that can run AI models without requiring cloud access.
The maker of that incredibly powerful chip? You guessed it: Nvidia.
