There Are Fixes for AI's Power Crisis. Here Is Why They Are Not Happening.
OpenAI shut down Sora partly because it consumed more power than the company could sustain. The US grid is out of headroom. Solutions like renewable expansion and nuclear are well-understood, but each faces years of permitting delays and political obstacles. Meanwhile, a Cambridge team has built a neuromorphic chip that could cut AI energy use by 70 percent, if it ever makes it to market.
Read full story →Anthropic Overtakes OpenAI in a Market Where None of the Normal Rules Apply
Anthropic hit $1 trillion on secondary markets this week, surpassing OpenAI's $880 billion on the same platform. But secondary market prices reflect scarcity and investor FOMO, not business fundamentals. Anthropic's last formal funding round valued it at $380 billion. Someone reportedly offered to swap their house for shares. The underlying revenue growth is extraordinary; the price is something else entirely.
Read full story →The Workplace Where Humans Will Soon Be the Minority
Slack's GM predicts that within two years, AI agents will outnumber human users on the platform. A Gartner survey of 469 CEOs backs him up: most expect transformative change by 2028. But a ground-level survey finds half of enterprises still stuck in pilots, blocked by talent gaps and uncertain ROI. The destination seems agreed. The road is murkier.
Read full story →DeepSeek V4 Arrives With a Price Tag That Changes the Conversation
DeepSeek released V4 with 1.6 trillion parameters and output pricing of $3.48 per million tokens, against $25-30 from Western rivals. Trained on Huawei's Ascend chips rather than Nvidia hardware, it is both a competitive model and a statement that China's AI supply chain no longer depends on US exports. The market noticed immediately.
Read full story →GPT-5.5 Bets on Autonomy, Not Raw Intelligence
OpenAI's GPT-5.5 leads on agentic benchmarks but trails rivals on abstract reasoning, and that is by design. The model is built to execute multi-step tasks with minimal guidance rather than to reason more brilliantly. Combined with OpenAI's "super app" vision, it signals a deliberate pivot from intelligence as a product to agency as a service.
Read full story →Claude Opus 4.7 Makes a Quiet Leap on Vision and Code
Anthropic's Opus 4.7 lifted its vision benchmark score from 54.5 to 98.5 percent and tripled its production task resolution rate on SWE-bench. Released April 16 with pricing unchanged, it is one of the larger capability jumps in a single model iteration this year, though Anthropic's own docs note a more powerful Mythos Preview already exists above it.
Read full story →The Billion Messages Nobody Sent
Slack just passed a billion messages a day, and a growing share of them were sent by AI agents, not humans. Peter Harrison argues this is the moment the labour displacement story moves from abstract economics into the communication layer itself. Agents coordinate work. They don't earn wages. They don't buy anything. The loop breaks quietly, one chat at a time.
Read opinion →