Blog

Anthropic and OpenAI both cut prices and leapfrogged each other’s flagship models — on the same day. If you’re budgeting AI spend or picking a vendor, capability is converging faster than pricing is stabilizing: don’t assume the model you paid the most for stays ahead for long. On September 22, Anthropic released Claude Opus 5.5 at roughly 20% lower per-token pricing ($4/$20 per million input/output tokens) while claiming Astra-level performance at about a fifth of the cost; OpenAI answered the same day with GPT-6 Sol and Luna, cutting API prices in half ($2/$10 and $0.10/$0.50 per million tokens) and claiming Sol beats Claude Opus 5 at 9% of the cost per task.

Google, OpenAI, and Anthropic are building their own AI safety regulator — and Meta, xAI, and Nvidia are calling it self-serving. This is the leading labs’ answer to last week’s political backlash over their call to slow down: rather than wait on Washington, the three biggest labs are moving to write frontier AI’s safety rulebook themselves, and whether that rulebook holds up matters to anyone who will build on, buy from, or eventually regulate these companies. The proposed body, the Standards Authority for Frontier AI (SAFA) — modeled on FINRA, the Wall Street self-regulator, per an idea Google DeepMind CEO Demis Hassabis floated in July — would run pre-deployment third-party testing and standardized safety-incident reporting, but it has no government enforcement power and would be funded by its own member labs; Meta, xAI, and Nvidia have publicly objected that three companies don’t speak for the industry.

Anthropic just committed $11.6 billion to Akamai for cloud capacity — with room to grow to roughly $20 billion. This is one of the clearest dollar figures yet on what it actually costs to run a frontier AI company, and it shows compute spending rippling into unexpected corners of the market: Akamai’s stock jumped as much as 17% after hours on the news. The seven-year deal covers CPU (not GPU) infrastructure for Anthropic’s growing agent workloads, includes warrants convertible into roughly 5% of Akamai’s stock, and follows similar Anthropic compute commitments to Amazon, Google, Microsoft, and AMD.

DeepSeek’s revenue just crossed a $1 billion annual pace, and it’s raising $7.5 billion toward a Shanghai IPO. China’s leading open-weight AI lab is no longer just a low-cost disruptor: real revenue and a credible path to public markets change the calculus on whether the frontier AI race stays a two-lab, U.S.-only story. DeepSeek’s annualized revenue run-rate has more than doubled to $1 billion after August price increases of 2.3x to 4.5x (its prices remain among the cheapest of any major model), and it is finalizing a roughly $7.5 billion round at about a $75 billion valuation ahead of a planned Shanghai Stock Exchange listing.

Meta is betting its next AI platform is something you wear, not an app: $1,299 VR glasses, cheaper Ray-Bans, and a keychain AI pendant. For anyone thinking about where AI product strategy is headed, Meta’s Connect keynote is the clearest signal yet from a major platform company that it sees the next interface as hardware-native AI, not another chat window. At Connect 2026 (Sept. 23 to 24), Meta unveiled $1,299 Meta VR Glasses (100 grams, launching spring 2027) built around its Muse AI agent, a refreshed Ray-Ban Meta Gen 3 ($449+, available now) with a dedicated AI button, and a new palm-sized “Muse Charm” pendant for voice- and camera-based AI interaction on the go (pricing not yet set).

Law firms are now bidding against Anthropic and OpenAI for engineers — at salaries that rival what they pay first-year associates. This week’s jobs story: the AI talent premium is spilling out of the tech sector and into every professional-services firm that touches white-collar work, which previews the kind of hiring competition for AI-fluent talent that MBAs entering banking, consulting, and law-adjacent industries should expect to see more of. Reuters reported this week that major law firms — Cooley (which spun up a 14-role “Cooley AI” unit), Kirkland & Ellis ($500 million committed to a proprietary AI platform), and Morgan & Morgan ($1 billion pledged over a decade), among others — are paying $220,000-plus for AI engineers and up to $438,000 for a director of applied AI, even as legal-analytics firm Firm Prospects counted 46 lawyers moving the other direction, from top-200 firms to AI companies like Harvey, Anthropic, and OpenAI, in the first half of 2026.