Nvidia's top management has made it clear that the demand for its new acceleration platform is high. Blackwell It continues to grow. During a recent stay in Taiwan, Jensen Huang He indicated that orders are progressing rapidly and that the company needs more production capacity to serve data center and project clients. Artificial Intelligence.
At a TSMC event in Hsinchu, the executive stressed the need to bolster wafer supply and emphasized that the company's activity is growing month by month. From the Taiwanese manufacturer's side, CEO CC Wei acknowledged this request and cautioned that, despite very strong sales prospects, the Available capacity remains limited.
More wafers for TSMC amid the wave of orders
Huang asked his long-standing partner to increase the volume of wafers destined for Blackwell to alleviate supply chain strain. According to Wei, the request exists—without specifying figures—and is a response to the accelerating adoption of AI in businesses and consumer markets. Meanwhile, TSMC is working to balance supply and demand, although pressure on its manufacturing lines will continue. high in the short term.
The Nvidia executive also emphasized TSMC's role in their technology roadmap. He made it very clear that the success of their next-generation accelerators is inseparable from TSMC's support. world leader in foundry, a message that points to a strategic relationship that will continue to intensify.
HBM memory and key partners: SK Hynix, Samsung and Micron
A critical part of Blackwell's rollout is high-bandwidth memory. Huang noted that SK Hynix, Samsung, and Micron have significantly increased their production to support growth, and that Nvidia has already received samples of the most advanced versions of these components.
SK Hynix recently confirmed that it has committed all of its production for next year and plans to increase investment in anticipation of a potential AI-driven chip supercycle. Meanwhile, Samsung It is holding talks to supply next-generation HBM (HBM4), while Micron It also accelerates its roadmap, which alleviates some of the bottleneck but does not eliminate the tight nature of the market.
How it affects Europe and Spain
For European cloud operators, integrators, and technology companies, this context translates into more demanding delivery deadlines and a need for plan ahead the capacity of their data centers. Blackwell's availability will depend on both TSMC's wafer production schedule and the arrival of HBM memory, so upgrade schedules in Spain may require fine-tuning and staggered purchases.
In practical terms, generative AI and advanced analytics projects in Europe will need to coordinate infrastructure procurement and deployment, closely monitoring the evolving capacity of Asian suppliers. Even with improved supply, technology managers will need to consider scenarios with longer delivery windows and prioritization by manufacturers.
Blackwell: chip family and associated ecosystem
Huang pointed out that Nvidia doesn't just manufacture GPUs: the Blackwell platform also relies on CPUs, networking, and switching, multiplying the number of components involved in each system. This ecosystem means that any stress on wafers or HBM memory has a cascading effect on entire servers, cabinets, and clusters intended for training and inference workloads.
Sales to China and a restricted market
When questioned about the Chinese market, the executive indicated that there are no active talks underway for market Blackwell thereIn light of current US restrictions on exports of advanced technology, this regulatory framework determines the fate of the shipments and the geographical assignment of the supply, reinforcing competition between regions to ensure supply.
The current landscape combines a very strong demand With an industrial machine still fine-tuned: TSMC forecasts rising sales, memory manufacturers are increasing production, and Nvidia is looking to increase the availability of its platform, while European customers are calibrating timelines and budgets to avoid missing the Blackwell train.
