Anthropic released Claude Opus 4.5 last week, a new large language model aimed at code generation that beat analysts expectations by scoring 80% on the SWE verified benchmark and earning the no. 1 spot on the ARC AGI test. The newsletter describes the model as capable of doing nearly any of the work required of a talented entry level software engineer and, in some cases, work at an advanced level. Anthropic is framed as taking a leadership position in applied Artificial Intelligence coding tools and the enterprise customer segment, a strategy the author says puts the company on a path to profitability in 2028 and projects ‘?B in revenue’.
Google also features prominently. Reports that Meta is in talks for a multi-billion dollar deal to use Google TPUs sent Google’s share price to a 52-week all-time high amid ongoing supply constraints for Nvidia. The piece notes Anthropic’s existing 1 million TPU deal to train the next version of Claude and highlights Ironwood, a recent TPU rumored to deliver a 1.5-2x performance improvement per dollar spent on training. Those developments are presented as elevating Google as a credible challenger to Nvidia’s ‘?T market cap’, particularly for niche TPU use cases aligned to coding workloads.
On enterprise implementation and public policy, Salesforce survey findings are summarized: full Artificial Intelligence implementation has increased 282% since 2024, from 11% to 42%; CIOs are dedicating 30% of budgets to agentic Artificial Intelligence; 96% say their company is either using AI agents or plans to within two years; and 75% of CIOs report feeling more confident in their role than a year ago. Separately, the U.S. Genesis Mission is described as an initiative that opens government scientific data, including 70 years of classified material, to labs. OpenAI, Anthropic, and Google reportedly received petabytes of experimental data from 17 U.S. National Laboratories spanning 1953 to 2025, and the order directs the Department of Energy to build a closed-loop Artificial Intelligence experimentation platform linking supercomputers, datasets, foundation models, and robotic laboratories, a development labeled the ‘AI Manhattan Project’ in the article.
