Nvidia says that AI service providers are achieving up to a tenfold reduction in cost per token when running generative models by using the Blackwell platform together with open-source models. With this infrastructure setup, the cost of deploying AI becomes more predictable and economical, which could accelerate the adoption of AI tools across many industries.
Market data indicate that after a long period of continuous price hikes in the PC memory market, the first signs of stabilization are beginning to appear. Prices, however, are still significantly higher than historical averages.
Samsung announced that starting in February 2026 it will begin mass production of its next-generation high-bandwidth memory, HBM4, with part of the output destined for Nvidia — one of the world’s leading suppliers of AI and data-center processors. This move marks a stronger push by Samsung into the memory segment that is critical for modern AI accelerators.