Microsoft CEO Satya Nadella Just Shared Bad News for Nvidia and Great News for This Rival Chipmaker
I'm LongbridgeAI, I can summarize articles.Microsoft plans to shift Azure data centers from Nvidia GPUs to custom Maia chips starting next year, positioning Maia as the primary inference system. This move benefits partner Marvell Technology, which designs the chips, while reducing reliance on Nvidia. Microsoft aims for multi-gigawatt deployments of Maia 200 and 300 series, signaling a broader industry trend among hyperscalers toward custom silicon for AI workloads.
Microsoft (MSFT +1.08%) is planning a big change to its Azure cloud computing data centers starting next year. It will rely much more on its custom-designed silicon than on leading graphics processing units (GPUs) like those from Nvidia (NVDA -1.84%). It's already made significant progress in expanding workloads for its Maia 200 chips, and it expects its next-generation Maia 300 chips to deliver significant improvements when it starts deploying them next year.
"Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services," CEO Satya Nadella said. That's great news for its custom chip partner, Marvell Technology (MRVL -1.65%), and bad news for Nvidia.
Image source: Microsoft.
Nvidia is losing data center market share
Nvidia's extreme dominance of the data center GPU space during the first few years of the artificial intelligence (AI) boom left its market share with nowhere to go but down. Its biggest customers, including the hyperscalers and leading AI frontier labs, have developed custom solutions that can handle AI workloads, and that can deliver better price performance on specific tasks than Nvidia's GPUs.
That said, GPUs offer far greater flexibility in terms of the types of tasks they can perform, making them particularly valuable for companies deploying servers to sell compute capacity to third parties with dynamic demands. Notice that Microsoft is starting its Maia expansion by focusing on its own products and services, not on leasing Maia compute capacity to third parties looking to cut costs on their OpenAI GPT-based applications.
NVDA - Nvidia
NASDAQ · As of 4:00 PM ET 10/15/26
Moneyball Superscore97/100Key data points
Their use, though, could expand over the long run. Alphabet (GOOG +0.37%) (GOOGL +0.45%), which has been building its own custom AI chips for longer than any of its competitors, has seen strong adoption of its Tensor Processing Units (TPUs) by its Google Cloud customers for training and inference. Apple used TPUs to train Apple Intelligence, and Anthropic is using gigawatts worth of TPUs to develop its large language models.
While Nvidia can continue to grow sales extremely quickly, other chipmakers' growth rates are likely to outpace it. That's reflected in its P/E valuation, which now sits below those of its biggest rivals.
Microsoft could be Marvell's fastest-growing customer
During its 2026 investor day this month, Marvell said it expects that it will grow its share of the custom AI accelerator market over the next five years. Its work on the Maia 300 is likely a big reason why. Microsoft's plan to make that processor the primary inference system for its products and services is in line with earlier reports that it will make a big step up in custom silicon orders starting next year.
The revenue opportunity for Marvell is huge. Microsoft will spend approximately $175 billion in capital expenditures on data centers over the next year. Roughly two-thirds of that will go toward servers and networking equipment. If Marvell takes just a sliver of that spending, it would amount to a huge boost to its revenue, which totaled just $9.5 billion over the last four quarters.
MRVL - Marvell Technology
NASDAQ · As of 4:00 PM ET 10/15/26
Moneyball Superscore88/100Key data points
Marvell believes the combination of its custom AI chips and its networking chips will help it grow revenue tenfold over the next five years. That is to say, Microsoft's ramp-up in Maia chips is just the start. Marvell also signed a deal with Google earlier this year to design and sell inference-optimized versions of its TPU chips. The deal also gives Alphabet the option to take a stake in Marvell.
Considering Marvell's growth potential, the stock's forward P/E of 63 looks inexpensive. If you look out to its fiscal 2029 (which mostly takes place in calendar 2028), the stock trades at just 21 times earnings estimates. Given its potential to capture a significant slice of the market for data center AI processors, that seems like a good price to pay.