Nvidia expects to sell $1 trillion in AI chips through 2027 — and it's pushing further into inference

At the recent GTC conference, Nvidia's CEO Jensen Huang announced an ambitious new inference system, marking a pivotal moment in the company's strategy to maintain its leadership in the rapidly evolving AI sector. With the demand for AI technology surging, Nvidia now anticipates a staggering $1 trillion demand for its AI systems, specifically the Blackwell and Rubin architectures, by the year 2027. This figure represents a significant increase from the previous estimate of $500 billion through 2026. The highlight of the event was the introduction of the Nvidia Groq 3 LPX, a groundbreaking chip designed to enhance inference workloads by as much as 35 times. This innovative chip combines technology from AI chip startup Groq with Nvidia's own Vera Rubin architecture, and its production will be handled by Samsung, with shipments expected in the latter half of this year. Huang emphasized the importance of this launch, stating, "The inflection point of inference has arrived." This announcement is a continuation of Nvidia's $20 billion partnership with Groq, which involved licensing their technology and recruiting top talent from the startup. The collaboration had been hinted at during Nvidia's previous earnings call and was later confirmed by reports from The Wall Street Journal. While Nvidia's graphics processing units (GPUs) continue to dominate both training and inference tasks in AI, the competitive landscape is evolving. A growing number of companies, from major cloud providers to emerging chip startups, are developing specialized and more cost-effective systems tailored for the repetitive nature of inference tasks. The increase in AI agents—tools that automate tasks for humans—could further escalate the demand for efficient inference solutions. In response to this evolving market, AI firms such as OpenAI have begun exploring alternatives to Nvidia's hardware. Reports suggest that OpenAI was not fully satisfied with Nvidia's inference chips, leading to a significant $10 billion partnership with the inference chip startup Cerebras earlier this year.