Anthropic Hits $380B Value as Microsoft Eyes Exit [Model Behavior]

[00:00] Nina Park: I'm Nina Park. Welcome to Model Behavior. [00:03] Nina Park: Model Behavior examines how AI systems are built, deployed, and actually operated in real professional environments. [00:11] Thatcher Collins: I'm Thatcher Collins. [00:13] Thatcher Collins: Joining us today is Chad Thompson, who brings a systems-level perspective on AI, automation, and security. [00:20] Thatcher Collins: Chad, great to have you. [00:21] Thatcher Collins: Thanks, Thatcher. [00:22] Thatcher Collins: It is a significant week for infrastructure shifts, particularly regarding how the major labs are funding their next generation of compute. [00:31] Chad Thompson: Let's start with Anthropic. [00:34] Chad Thompson: Yesterday, the company closed a $30 billion Series G round, more than doubling its valuation to $380 billion. [00:44] Chad Thompson: Thatcher, they are also moving into the political arena. [00:47] Thatcher Collins: Exactly, Nina. [00:49] Thatcher Collins: Anthropic announced a $20 million donation to Public First Action. [00:55] Thatcher Collins: This group is pushing for AI guardrails and transparency, [00:59] Thatcher Collins: specifically supporting candidates like Senators Marsha Blackburn and Pete Ricketts. [01:05] Thatcher Collins: It's a distinct regulatory stance compared to the rest of the industry. [01:09] Thatcher Collins: It is interesting because while Anthropic pushes for regulation, [01:13] Thatcher Collins: Microsoft is looking to change its relationship with OpenAI. [01:17] Thatcher Collins: Microsoft AI lead Mustafa Suleiman confirmed they plan to develop their own frontier models [01:22] Thatcher Collins: to reduce dependence on OpenAI's technology. [01:26] Chad Thompson: That shift comes just as OpenAI released GP3 Codex Spark. [01:31] Chad Thompson: This is their first production model to run on Cerebrus hardware instead of NVIDIA chips. [01:37] Chad Thompson: Chad, the performance numbers are notable. [01:40] Chad Thompson: They are. [01:42] Thatcher Collins: Codex Spark is hitting 2,100 tokens per second. [01:47] Thatcher Collins: It's built for real-time iteration rather than deep reasoning, [01:50] Thatcher Collins: which suggests OpenAI is diversifying its hardware strategy to solve inference speed bottlenecks. [01:56] Thatcher Collins: For sure. On the enterprise side, we saw a major deployment this week. Goldman Sachs is now using [02:04] Thatcher Collins: Claude 4.6 agents for trade reconciliation and client onboarding. They report a 30% reduction [02:13] Thatcher Collins: in onboarding times, which is a massive efficiency gain for a Wall Street firm. [02:18] Chad Thompson: While banks are moving toward automation, Hollywood is pushing back. [02:23] Chad Thompson: The MPA has called on ByteDance to stop using copyrighted works in its new C-Dance 2.0 model, [02:31] Chad Thompson: which recently produced a deep fake featuring Tom Cruise and Brad Pitt. [02:36] Thatcher Collins: The conflict between generative capability and copyright is also hitting the campaign trail. [02:44] Thatcher Collins: Democratic candidates like Mallory McMissle are making AI regulation and kids' online safety a central pillar of their 2026 platforms. [02:54] Thatcher Collins: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. [03:01] Thatcher Collins: You can find more detailed reporting at mb.neuralnewscast.com. [03:09] Thatcher Collins: Neural Newscast is AI-assisted, human-reviewed. [03:13] Thatcher Collins: View our AI Transparency Policy at NeuralNewscast.com.