Thanks to Nvidia Corp.’s practice of reporting earnings outside of the typical cycle for technology companies, the question of whether the almost $5 trillion company will record strong demand in 2026 had already been safely answered well before its latest announcement on Wednesday.
Investors already knew that the hyperscaling AI companies are collectively forecasting around $650 billion in capital expenditures this year, an increase of about 60% from 2025 — and Nvidia will get a lot of it.
Preempted by its customers, Nvidia thus needed its own fresh good news to trump what investors already knew. Hello, margins. Adjusted gross margin in the November-January period was 75.2%, the highest it has been since the second half of 2024. The company forecasts that number to be roughly the same in the current quarter. What’s unclear is just how long Nvidia can maintain this extraordinary profitability as the AI landscape matures.
One lingering question going into the company’s closely monitored analyst call was about supply. The company won’t be able to avoid the rising cost of memory, even if it is at or near the front of the line for the crucial component compared with most other companies in the electronics world. The company’s chief financial officer, Colette Kress, said that the company had “strategically secured inventory and capacity to meet demand beyond the next several quarters” but that it expected “tightness” in supply to persist. The leading makers of the components have warned the shortages will linger through 2027 and maybe longer. The demand for the hardware behind AI is still growing faster than the infrastructure needed to produce it.
Another question is how resilient Nvidia is against competitors seeking a slice of the AI chip business. To reapply Jeff Bezos’ (possible) phrase, Nvidia’s margin is its competitors’ opportunity — and they are starting to grasp it. In the final quarter of last year, Alphabet Inc.’s stock price rose when it started to become clear its own AI chips — called Tensor Processing Units, or TPUs — were handling a significant portion of workloads for its Google Cloud clients as well as for its own AI services like Gemini. Amazon.com Inc. notched a win for its own AI chips by bringing on Anthropic as a client.
The availability and pricing of alternative chips has made diversifying worth the constraints. According to Bloomberg Intelligence, the average selling price per unit of a Google TPU is $8,000 to $10,000 compared with $23,000 or more for Nvidia’s H100 chip or $27,000 and above for its newer Blackwell system.
