Anthropic priced its IPO at $83 per share on Wednesday, above its $70–$78 range, and upsized the offering to 980M shares from 825M. That raises about $81.3B, roughly $79.4B of it for the company, and values Anthropic at about $1.5T fully diluted, versus $965B in its May funding round. Google, Microsoft, Nvidia and AMD separately agreed to buy $26B of stock at the IPO price, CNBC reported, and shares begin trading on Nasdaq as ANT today (Anthropic)
Private Companies
Palo Alto Networks agreed to acquire Applied Compute for $4.5B in stock. Applied Compute runs a platform for post-training and inference of open-weight AI models, and Palo Alto plans to offer it to enterprises in public clouds and sovereign environments. CEO Yash Patil will become Palo Alto's Co-Chief AI Officer. Earlier this month, Applied Compute raised $350M at a $3.25B valuation in a round led by Elad Gil, with Nvidia participating. The deal is expected to close in Palo Alto's fiscal third quarter of 2027 (SEC filing)
Architect Labs unveiled Frostbite, a general-purpose CPU built around interaction nets, a model where a program is a graph that rewrites itself in many places at once. The company's AI agents designed and verified the instruction set and architecture through to final layout. In its 7nm results, one Frostbite quad delivered 4.59x the performance per watt of an undisclosed dual-issue, in-order CPU core across 25 workloads. The figures are pre-silicon, and Architect Labs plans to release the RTL and ISA (X post)
Public Markets
AMD$530.14▲ 5.3%Mkt Cap: $864B
Hunterbrook reported that blacklisted Chinese server maker IEIT Systems has shipped at least six newer AMD firmware releases since its March 2025 Entity List designation. It did not establish an export-control violation, and AMD did not comment. Hunterbrook Capital disclosed a short AMD position (Hunterbrook)
Emerging
Finding work for custom silicon: A paper from Peizhe Liu, Akanksha Jain, Gerasimos Gerogiannis, Vidushi Dadu, Parthasarathy Ranganathan and Jovan Stojkovic, Lodestone: Automated Semantic Discovery of Datacenter Hardware Specialization, uses AI agents to find computations that recur across unrelated datacenter code and could justify dedicated hardware, such as ISA extensions, near-core units or standalone accelerators. Agents describe what each hot region of an execution trace computes, group matching regions into families, and verify each family against recorded inputs and outputs. Across 20 workloads in six categories, 30 discovered families, modeled in a cycle-accurate simulator and implemented in RTL, sped up workloads by 12.6% on average and up to 80.2% (arXiv)