Blackwell

Nvidia projected more than $1 trillion in Blackwell and Rubin chip sales through 2027, underscoring Blackwell’s role as the commercial bridge to its successor.

Who they are

Despite its person classification, coverage identifies Blackwell as Nvidia’s AI-chip generation, introduced in 2024 with the GB200 superchip combining two B200 GPUs and a Grace CPU. It appears as the foundation for Nvidia’s data-center, workstation and sovereign-AI deployments, as well as the performance baseline against which the company presents its next Rubin platform.

The recent arc

Coverage peaked in 2025’s fourth quarter as Blackwell shifted from a forthcoming architecture into a focal point of Nvidia’s demand narrative, product access debate and deployment cycle. At GTC, Jensen Huang said Nvidia expected to sell 20 million Blackwell chips, compared with 4 million Hopper chips; the same period also brought reporting on Trump and Xi Jinping’s discussion of chipmaker access to China, with Trump saying Blackwell sales were not discussed.

The recent arc

In 2026, the emphasis moved from Blackwell’s rollout to its place in Nvidia’s succession plan. Nvidia launched Vera Rubin claiming sharply lower training and inference costs than Blackwell and said some LLMs could be trained with roughly one-fourth as many chips. Yet Blackwell remains present in major deployments: ByteDance was reported to be working on a Malaysian installation of roughly 500 systems, Taiwan opened a Blackwell-based supercomputing center, and Eli Lilly cited a data center with 1,016 Blackwell chips.

The tension

The central tension is between extraordinary demand for Blackwell and the pace at which Nvidia itself is moving beyond it. Early coverage included reported engineering and rack-overheating challenges, AMD’s MI325X positioning against upcoming Blackwell chips, and China-access scrutiny; later coverage increasingly frames Rubin as both Nvidia’s successor product and a claimed efficiency leap over Blackwell.

Why it matters

If deployment stories and Nvidia’s sales projections continue to materialize, Blackwell will stand as the installed base that ties AI infrastructure spending to Nvidia’s accelerated product cadence. The uncertainty is whether customers can sustain the upgrade cycle as Rubin promises materially better efficiency, while geopolitical access constraints and competing accelerators may determine how broadly Blackwell-era systems spread.

Related: Nvidia · Jensen Huang · China · AMD · Nvidia launches the Vera Rubin platform, saying it will offer dramatic · Nvidia announces Blackwell, a new generation of AI chips available lat