NVIDIA is shifting towards annual product releases to maintain its leading position in the AI hardware field.

When we last saw Nvidia's roadmap, the company announced that it was abandoning its biennial release schedule for new gaming graphics cards and pushing back the plan by a year. This meant that according to the previous release schedule, it would launch a new GeForce product line in 2024 instead of 2025. Now, the company seems to have reversed this plan, at least for its AI hardware. It has accelerated the release schedule to once a year to maintain its dominant position in the AI field. It is currently unclear how this will affect gamers, but it indicates that Nvidia is not resting on its laurels.

The company provided an updated roadmap during this month's investor presentation, as reported by Tom's Hardware. The title of the roadmap boldly proclaims "From Biennial to Annual Cadence," indicating a significant increase in the speed of product releases. The plan shows the annual progression of upcoming hardware as follows: H -> B -> X. This means that its current H100 accelerator will become H200 next year, followed by B100, and finally X100 in 2025. H200 sounds like a "refresh" version of its contemporary H100 design, which is the gold standard for AI training. Its Grace Hopper GH100 "superchip" will follow the same progression, first becoming GH200, then GB200, and finally GX200 in 2025.

Nvidia's previous roadmap indicated that "Hopper Next" was expected to arrive in 2024, and "Grace Next" in 2025, in a somewhat amorphous discussion of future hardware. However, it appears that Nvidia has released these products ahead of schedule to maintain its leading position in the industry. The original plan for "Ada Next" was set for 2025, and the GeForce version with the Blackwell architecture was planned for 2024, so it is currently unclear how the new plan will affect these plans. The same investor presentation also noted that Nvidia's "installed base needs an upgrade," with only 47% of customers having GPUs that support ray tracing.

The company's move is wise as there is currently more demand for Nvidia products than supply, leading to inflated prices. This scarcity may prompt the company to explore other solutions, including developing custom chips, with reports suggesting that Microsoft is partnering with Athena. Google and Amazon have already embarked on this path with their Tensor GPUs and Graviton chips, but unless Nvidia can convince them, Microsoft may one day run its own chips on its Azure cloud. Accelerating the release schedule and diversifying the product portfolio are important steps in this direction, but we will have to wait for the Blackwell release in 2024 to see how the situation unfolds.