QumulusAI stock crashes 15%, erasing premarket rally despite $240.9 million Nvidia Blackwell deals

NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter model, on October 15, offering open weights and datasets. On October 21, NVIDIA and the German government formalized an AI ecosystem partnership, focusing on infrastructure. The model is adopted by companies like Scale AI and Salesforce, aiming to drive enterprise adoption and infrastructure demand. NVIDIA's strategy involves giving away the model to capture value through hardware and infrastructure.

Nvidia (NVDA) shares rose 2% to $220 after announcing a $1.1B advance to KYEC for U.S. chip-testing facilities, part of $279B in supply commitments. CEO Jensen Huang emphasized capacity constraints. Nvidia expects testing capacity to double by 2027, with significant investments in suppliers like IBIDEN, Samsung Electro-Mechanics, and Amkor Technology (AMKR).

Laura Fay, Nvidia's VP of global enterprise marketing, retired in July after a decade. She was recruited by CEO Jensen Huang to expand Nvidia's marketing beyond gaming. Nvidia hired Alison Wagonfeld as its first chief marketing officer in February. Several other long-term Nvidia executives have also retired this year.

Amazon (AMZN) is purchasing billions in Nvidia (NVDA) AI hardware, but some unpowered server racks are sitting in warehouses due to insufficient electricity in data centers. Amazon introduced an 'Early Rack Receipt' program to secure Nvidia chips amid competition. AWS CFO John Felton noted the AI business is 'hugely free cash flow negative,' with company-wide capital spending at $220B this year.

Bolt plans to use Nvidia's Alpamayo Stack software for autonomous driving in 25,000 Lucid vehicles in Europe, starting with a data-collection program in Madrid in 2027. Bolt and Nvidia will adapt the system to European road conditions. Lucid's potential robotaxi volume reaches 60,000 vehicles with existing partnerships.

Anthropic has committed nearly $90 billion to secure compute infrastructure, including $44.6 billion confirmed in an SEC filing and another $45 billion in negotiations. The deals target the Monarch Compute Campus in West Virginia, highlighting the shift in focus from model weights to power control. DeepSeek and llm-d released open-source software, reducing costs and commoditizing the inference stack.