Jensen Huang resigns from Nvidia

Stork's LLM price catalog recorded a massive price hike for Mistral Large (latest) today. First-party list prices jumped from $0.50 to $0.70 per million input tokens, and from $1.50 to $2.10 per million output tokens. This represents a flat 40% increase across the board for developers buying directly from the lab.

Price cuts have been the industry norm for the past year, making a 40% hike a glaring outlier. Mistral is signaling that they either need to improve their margins immediately to offset compute costs, or they believe their model quality commands a premium that users will swallow.

For builders routing heavy traffic through Mistral Large, your runway just got shorter. You now have to decide if the performance justifies the sudden premium, or if it is time to route queries back to cheaper frontier alternatives.

CoreWeave reported Q3 revenue of $3.66 billion, a 168% year-over-year increase. The infrastructure provider also disclosed it has 10x as much qualified demand as available capacity through the end of 2027. In a related move, UK self-driving startup Wayve signed a multi-year deal to train its AI Driver models on CoreWeave's hardware.

The numbers highlight a brutal reality for anyone trying to secure GPUs: the waitlist is only getting longer. CoreWeave is printing money because they have the hardware everyone else is desperate to rent, and locking in autonomous driving companies like Wayve ensures their clusters stay saturated.

The winner here is CoreWeave's valuation, but the loser is the mid-market AI startup. If you do not have the capital to sign multi-year commitments right now, you are going to be fighting for scraps in the spot market.

Anthropic has officially launched its IPO roadshow, offering 825 million Class A shares at $70 to $78 each on the Nasdaq under the ticker ANT. The offering aims to raise roughly $64 billion and is reportedly six times oversubscribed. During the roadshow, the company also claimed it took just three days to port Claude to AMD GPUs, a process that previously required months of engineering.

Going public at this valuation puts Anthropic in direct financial competition with the biggest tech incumbents. The six-times oversubscription indicates Wall Street is desperate for pure-play AI equities, especially ones that can demonstrate hardware flexibility in a constrained GPU market.

The rapid AMD port is a direct threat to Nvidia's software moat. If Anthropic can easily switch compute providers in a weekend, it strips away vendor lock-in and gives them immense leverage in future hardware negotiations.

Three senior national security personnel at OpenAI were prank-called by Russian duo Vovan and Lexus. The pranksters posed as Ukraine's spy chief for 17 minutes, engaging the OpenAI executives in discussions about Elon Musk and potential military AI assistance. Separately, OpenAI informed lawmakers it plans to support bipartisan bills restricting China's access to AI chips and chipmaking equipment.

The juxtaposition is staggering. OpenAI is actively positioning itself as a geopolitical asset to the US government, pushing for aggressive tech-nationalism against China. Yet, their own senior security team failed a basic identity verification test from known pranksters.

Lawmakers relying on OpenAI for national security advice should be nervous. If foreign pranksters can bypass internal vetting to reach senior personnel, actual state-sponsored threat actors will have no trouble infiltrating the human layer of their security apparatus.

SpaceX announced plans to construct $74 billion worth of AI computing sites. The massive infrastructure rollout includes a $50 billion facility located in Arizona and a $24 billion gas-powered complex situated near the company's Starbase in Texas.

This is a brutal escalation in the AI infrastructure arms race. By using SpaceX's capital and physical footprint, Musk is bypassing traditional cloud providers to build proprietary compute clusters. The gas-powered complex specifically highlights how energy constraints are forcing AI builders to become their own utility providers.

Traditional hyperscalers should be on high alert. SpaceX is not just launching satellites anymore; they are aggressively acquiring the physical land and energy required to train the next generation of frontier models entirely in-house.

The FBI successfully disrupted a North Korean cyber operation dubbed the FaceStealer campaign. Hackers utilized fake Zoom-style meetings, deploying recordings of known individuals to steal over $400 million in cryptocurrency.

The attack vector represents a terrifying maturation of deepfake technology. Attackers no longer need to compromise passwords; they just need to clone a trusted face and voice to authorize unauthorized transfers in real-time video calls.

Every compliance and finance team needs to update their verification protocols immediately. If your authorization process relies solely on a video call confirmation, you are entirely vulnerable to the exact tactics North Korea just used to drain half a billion dollars.