NVIDIA Rubin platform mass production imminent, Bitcoin miners accelerate shift to AI infrastructure track

NVIDIA CEO Jensen Huang revealed at the CES conference in Las Vegas that the next-generation Vera Rubin platform has entered mass production, with its AI computing power approximately five times that of the previous generation. This news has not only attracted significant attention within the AI industry but is also profoundly influencing the long-term trajectory of Bitcoin miners and the crypto infrastructure market.

According to NVIDIA’s plan, the Rubin platform will be officially launched later this year, with its core focus on AI inference, the fastest-growing segment, which helps users efficiently generate output results from pre-trained large models. Huang disclosed that the Rubin flagship server will integrate 72 GPUs and 36 CPUs and can form ultra-large clusters through high-speed interconnects, with a single cluster capable of housing over 1,000 Rubin chips, targeting data center-level deployment.

The emphasis of this release is on efficiency leapfrogging. NVIDIA stated that the Rubin system can improve generative AI token processing efficiency by about 10 times, primarily due to its self-developed data types and system architecture optimizations. Notably, despite only a roughly 1.6-fold increase in transistor count, overall performance has achieved a leapfrog growth, further raising the industry’s technological threshold.

Huang described the current AI development as a race for infrastructure. Faster computing power means reaching the next model capability inflection point more quickly, forcing competitors to continuously increase investments in chips, networking, and storage layers.

This “computing power arms race” is also reshaping the business models of Bitcoin miners. More and more miners are no longer solely positioning themselves as crypto mining enterprises but are transforming into power, rack, and data center operators, proactively selling energy contracts, cooling capacity, and existing site resources to AI clients. Compared to the highly cyclical profits of Bitcoin mining, hosting AI workloads often provides more stable cash flow during bear markets.

However, the AI boom has also raised the survival threshold. High-quality data center resources are becoming scarce assets, with large-scale cloud providers and AI startups driving up rents and equipment costs. This means that miners with scale, power advantages, and financing capabilities are more likely to succeed, while small miners heavily reliant on mining profits may face greater pressure by 2026.

Additionally, NVIDIA has simultaneously launched new network switches using co-packaged optical modules to support high-speed interconnection of thousands of servers. The company stated that CoreWeave will be among the first customers to adopt the Rubin system, with Microsoft, Amazon, Oracle, and Alphabet also viewed as potential users. This trend further confirms that AI infrastructure is becoming a new competitive frontier for crypto computing power capital.