For years, if you wanted to build an AI system, you needed Nvidia chips. Almost no one else could supply them at scale, and the company used that position to become one of the most valuable businesses on earth. That near-monopoly era is not over. But it is getting complicated.

Nvidia reports earnings Wednesday, and analysts expect a 79% jump in quarterly revenue, its fastest growth rate in more than a year. Adjusted profit likely came in around $43 billion for the quarter. Big Tech is pouring more than $700 billion into AI this year, up from roughly $400 billion in 2025, and a large share of that money flows through Nvidia's order books.

So why is the company's stock up only 19% this year, while AMD, Intel, and Arm have each roughly doubled, and Alphabet is up 27%?

The market is shifting under Nvidia's feet

The original gold rush in AI was about training, the phase where companies feed enormous datasets into models to teach them patterns and reasoning. Nvidia effectively owned that market. The processors required were expensive, power-hungry, and Nvidia built the ecosystem, software included, that made them work.

The next phase is called inference: the moment the model actually does something useful. When you type a question into a chatbot, search a database, or ask your phone to summarize an email, that is inference. The inference market is much larger than training, but it also rewards different things. Many of those tasks are smaller, faster, and more cost-sensitive. That opens the door to chips that were never competitive in the training race.

AMD and Intel are pushing processors tailored for exactly these workloads. Alphabet has struck deals worth tens of billions of dollars for its own custom chips. Amazon's in-house chip business is gaining ground. These are not scrappy startups. They are companies with the engineering talent, the customer relationships, and the financial depth to build alternatives that actually get used.

"It's less so Nvidia versus TPUs, Nvidia versus AMD," said John Belton, a portfolio manager at Gabelli Funds, which holds Nvidia shares. "I think it's more: is the Nvidia ecosystem as dominant moving forward, as some of these new inference workloads start to proliferate."

Nvidia is not standing still. In March it unveiled new chips built around technology from Groq, an inference-focused startup it acquired. Those chips are not yet included in Nvidia's own forecast of $1 trillion in sales from its Blackwell and Rubin platforms by the end of 2027, which means investors are watching closely for signs that the new product line can become a real growth engine.

What could go wrong

A few risks stand out beyond the competitive pressure.

Data center space is becoming a bottleneck. One analyst, Chaim Siegel of Elazar Advisors, put it plainly: "The customers just simply don't have place to put the GPUs. They want to buy as much as they can, but they don't really have the data centers to put them into." Demand is real, but physical construction takes time, and a slower buildout could limit how fast Nvidia can convert orders into revenue.

China is a separate problem. Nvidia has not been allowed to sell its most advanced H200 chips there, and Beijing is actively pushing local alternatives. CEO Jensen Huang recently traveled to China alongside President Trump, raising hopes for some easing of restrictions, but nothing has been announced.

Profit margins, currently around 74.5%, could also face pressure later this year as memory and chip packaging costs rise and Nvidia ramps up its next-generation Rubin chips.

The company is still the most important piece of infrastructure in AI. Wednesday's report will likely confirm that. But the more interesting question is whether its advantage in training, which was nearly unassailable, will translate cleanly into inference. The answer to that will shape not just Nvidia's future but the cost and accessibility of AI for everyone who eventually uses it.