Anthropic’s annualized revenue growth over seven months

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> 300% or 4x growth in ARR over six-and-a-half months

> recent reports confirm Claude most used after ChatGPT + demand for Claude API growing rapidly

> competition with Anysphere (Cursor) - who’s among their biggest customers - also rises after recent poaching of two Claude Code leads

> loss of employees despite Anthropic’s reportedly high salaries ($300,000 to $690,000) for few

> Anthropic’s research found all major AI models resort to blackmail when threatened with shutdown

> 12/16 models blackmail majority of the time (including Llama 4 Maverick when prompt is tweaked)

> not random response - explicitly reasoned through blackmail as the optimal choice

> goal conflict enough for misalignment: leaked sensitive documents even without threat of replacement

> low vs high blackmail rate when they believe situation is a test vs real

> other research by Apollo and METR also confirm AI misalignment: evidence finds models engage in cheating, manipulation and scheming

> mentions spike coincides with AI boom

> US datacenter power consumption projected to 3x from 25 GW in 2024 to 80+ GW by 2030

> to drive nearly half of US electricity growth 2023 - 2028

> tech firms moving to nuclear energy to support AI growth & close $500B power infrastructure gap

> exploring both small modular reactors (SMRs) for near-term needs + fusion tech for long-term