According to the Wall Street Journal, Nvidia's lead is fading as newcomers such as Google and Amazon start flogging their own high-end chips to outside punters, claiming power and efficiency that can match Nvidia’s best gear.
Smaller rivals, including AMD, Qualcomm and Broadcom, are rolling out silicon designed for AI data centres. Nvidia’s biggest clients, such as OpenAI and Meta, are crafting homegrown chips, which threatens the outfit’s ubiquity.
A mass customer walkout is unlikely, but firms rushing to diversify suppliers could make Nvidia’s future growth look a touch less magical than the cocaine nose jobs of Wall Street expect.
The market keeps mutating as giant infrastructure deals appear each week and fresh slabs of silicon hit the shelves.
Nvidia chief Jensen Huang likes to claim it is more than a chip shop, talking up “rack-scale server solutions” and marketing data centres as “AI factories.” Even so, its basic product is accelerated computing, and that is exactly what everyone else in the AI game wants.
From February to October, Nvidia sold $147.8 billion worth of chips, networking kit and assorted hardware, a hefty jump from $91 billion in the same stretch last year.
The outfit hit $4 trillion in value in July and briefly sprinted past $5 trillion before AI bubble worries clipped its wings. Even with that correction, it is worth more than twice Broadcom, which sits at $1.8 trillion.
Nvidia has sped up its release cadence. Late last year, it shipped its Grace Blackwell servers, the most powerful kit it has built so far, and the stock was snapped up almost instantly. At an October event in Washington, Huang said the outfit had sold 6 million Blackwell chips in 2025 and had orders for 14 million, worth more than half a trillion dollars.
Challenges linger. Nvidia has been effectively banned from flogging chips in China for three years, which is a problem because Huang insists the country holds half the world’s AI developers. Without those billions, its growth hits a ceiling, and Chinese firms will likely grow accustomed to domestic silicon.
AMD chair and chief executive Lisa Su said “insatiable demand for compute” would continue to rise, and she retooled AMD three years ago to take on Nvidia. AMD’s market cap has nearly quadrupled to more than $350 billion, and it has scored major deals with OpenAI and Oracle.
Broadcom has muscled into the ring. Once a slice of Hewlett-Packard, it ballooned into a $1.8 trillion monster through mergers and now makes XPU as well as networking kit that links data-centre racks.
Troubled Chipzilla mostly snoozed through the AI boom because of strategic faceplants but has poured money into design and manufacturing in the hope of coaxing clients to buy its new data-centre processors. Chipzilla will need all the goodwill it can get.
Qualcomm, better known for mobile chips, saw its shares jump twenty per cent in October after announcing its AI200 and AI250 accelerators, which it says excel in memory capacity and energy efficiency.
Google and Amazon have mountains of cash, and their internal chip projects are now attracting interest from third-party customers.
Google has long used its own tensor processing units and first opened them to outsiders in 2018. Firms including Meta, Anthropic, and the Fruity Cargo Cult Apple now buy or negotiate access to TPUs to train their models. Dylan Patel of SemiAnalysis said the trend might signal “the end of Nvidia’s dominance.”
Amazon is expanding a cluster for Anthropic that will house more than one million Trainium chips, and AWS has begun wider sales of custom silicon that it says is faster and more efficient than Nvidia’s equivalent gear.
Even Nvidia’s customers are nibbling at its market by commissioning application-specific integrated circuits. OpenAI and Broadcom have a multibillion-dollar deal to develop custom chips. Meta bought startup Rivos to push its in-house training chips, and Microsoft’s chief technology officer has said the outfit will rely more on custom accelerators. Elon Musk’s xAI is hiring chip designers to build hardware architectures for model training.
Nvidia insists its systems are more flexible than these bespoke options, and most analysts reckon it will keep its lead, but the AI chip world has far more than one player in the fight.