📝 usncan Note: Nvidia And Today’s Chip Landscape
Disclaimer: This content has been prepared based on currently trending topics to increase your awareness.
Futuristic central processor unit. Powerful Quantum CPU on PCB motherboard with data transfers.
getty
After the dominant U.S. tech company with the largest market cap lost around 5% over the last couple of weeks, analysts are trying to figure out why, and what this augers for the company’s financial future. Much of the recent coverage cites wishy-washy export signals as having a major effect on Nvidia’s stock with the on-again, off-again nature of H20 sales to the Middle Kingdom shaking things up. But what about competition from U.S. companies?
To be clear, Nvidia isn’t going anywhere – its hold on the AI hardware market is dominant, with Blackwell and H100s and AI accelerators par excellence, the firm is towering above the crowd. But that said, there are signs of competition, and some of them are coming from unlikely places.
There’s Intel and its line of Gaudi chips, named after famed architect Antoni Gaudi and his flamboyant style of building. Experts explain that the Gaudi chips are not competitive with Nvidia’s GPUs in terms of power – rather, Gaudi is for cost-conscious customers looking for a pared-down chip at lower expense.
Another contender, though, is not a traditional chip maker.
It seems that Google is continuing to develop its own Axion line of processors running in what the company calls a “Titanium” ecosystem – a framework for hardware performance that sort of sits in the tech stack next to the processors.
Now, there’s not a lot of public information about what Google is doing with Axion online – at least not this year. There was an explosion of articles in late 2024 about Axion’s origin, and how it could threaten Nvidia’s monolithic dominance. But one of the very few indicators that this is still in play is a new article at The Information suggesting that Google people might want some kind of parley with Nvidia to talk about things.
Double Threat
There is, however, some other suggestion that Google might also be making inroads with its TPUs.
“While Chinese rivals have gotten some notice for recent efforts to compete with Nvidia, investors may want to pay more attention to a major domestic rival that doesn’t get much buzz for its work in semiconductors,” writes Britney Nguyen Sept. 2. “Alphabet Inc. has closed the gap in the past year to become ‘the best alternative’ to Nvidia.”
Citing input from D.A. Davidson analysts, Nguyen theorizes how a “spinoff” of the AI accelerator division could be lucrative for the parent company. Specifically, Nguyen describes market interest in Google’s sixth-generation Trillium TPUs, which were made generally available in December, and the seventh-generation Ironwood TPUs, which debuted this spring. Noting Anthropic’s hiring of TPU engineers, Nguyen suggests that Google is gaining ground in taking business away from other vendors.
So the TPUs are one alternative, and Axion is another. It’s helpful to note, though, that the TPUs are only made for Google’s own ecosystem, and not competitive across the entire market.
Up and Down Movement
In the stock market, there’s no indicator that any competition is driving Nvidia stock down right now. After that share price cut, which seems generally related to export policies, the stock has gained about 1% in recent cycles. The suggestion of more robust domestic competition is a longer-term concept. When those in tech media report that Elon Musk and xAI might want to buy TPUs, or that a selection of customers are choosing some non-Blackwell chip builds, they’re pondering a longer time horizon, where the only constant in this kind of fast-paced market is change.
“There’s little argument that chipmaker Nvidia has been the biggest beneficiary of the advent of artificial intelligence (AI),” wrote James Brumley at Motley Fool last week in a piece that I think articulates this well. “Its processors are the centerpiece of most of the world’s AI computing platforms, and artificial intelligence-related products now account for the vast majority of Nvidia’s sales. … the next five years aren’t likely to look quite like the last five, though. The AI industry is maturing, and so are Nvidia’s competitors. Indeed, even AI data center operators are now looking for very specific solutions for their unique needs. Names other than Nvidia will benefit from this ongoing evolution.”
Neither Intel nor Google features in Brumley’s recent analysis. Instead, he mentions chip rival Qualcomm, and two other lesser-known companies: Marvell, with its CXL (Compute Express Link) motherboard extenders, and NXP Semiconductors, which he describes this way:
“(NXP’s) i.MX RT700 mini neural processing unit paired with NXP’s eIQ Neutron NPU AI/ML accelerator turns wearables, smart home devices, and even portable medical tech into true AI ‘edge computing’ products. At the same time, the company makes advanced driver assistance systems equipment, industrial automation solutions, and more.”
So – some of the change is based on new models and ways of building hardware systems. We know that edge computing is taking over as engineers find feasible ways to run more powerful LLMs on unconnected devices. Maybe the new world of the future will be reliant on vibrant unit-to-unit decentralized networks – and Nvidia’s brighter moments will be associated with obsolete top-down data center design. But data centers are still popping up, so again, this is a longer term anticipation.
We’ll be seeing more analysis of these trends in conferences and events planned in the MIT community through the fall. We’re closing out a year that brought us more than most of us could have imagined in terms of tech advances. What will the next five or ten years look like?
