DeepSeek V4.1 Flash 总拥有成本

Groq is one of the most interesting chip stories in AI. Nvidia paid $20B to license their IP and hire most of their team structured as a licensing deal rather than an acquisition to sidestep regulatory scrutiny. It close

Memory is taking over Hyperscaler CapEx. In CY23 and CY24, memory was ~8% of total Hyperscaler spend. We estimate it hits 30% in CY26 and moves higher in CY27. That's a near-4x shift in just four years. (1/4) 🧵

It doesn't make sense to say a subscription plan is worth $ X in isolation. The way to think about subscriptions is that your monthly payment grants you some numbers of “credits”. Each (model, token type) combo consumes

People think AI inference margins are a race to the bottom. Anthropic's gross margins were -94% in 2024. MiniMax was -25%. The narrative made sense (1/5)🧵

MONEY PRINTER ALERT🚨 NVIDIA vLLM B200 CAN GENERATE UP TO💰️$15 BILLION💰️OF ANNUAL PROFITS PER GIGAWATT serving the open DeepSeekv4.1 Flash model at the official interactivity & official selling prices. Using Engram DR

NVIDIA's Evolution🧵 Most people focus on FLOPs and headline specifications when comparing NVIDIA GPU generations, but the real story lies in the chip floorplan, which is a remarkable piece of engineering. With each gen

When will AI systems be able to carry out long projects independently? In new research, we find a kind of “Moore’s Law for AI agents”: the length of tasks that AIs can do is doubling about every 7 months.

NEW: The actions of Trump and Vance in recent weeks highlight something under-appreciated. The American right is now ideologically closer to countries like Russia, Turkey and in some senses China, than to the rest of th

GPT-5.6 Sol and Luna are ahead of Terra at every point on the Intelligence vs Cost per Task chart. GPT-5.6 Luna stands out as a particularly cost efficient model Charting the Artificial Analysis Intelligence Index shows

What’s the evidence on rent control? I charted the findings from 112 studies: Great for rent-controlled tenants ✅ But: Pushes up rents for other people ❌ Pushes down new housing construction ❌ Pushes down building qua

We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy bec

New from me: Young people in Britain are some of the most well-educated in the rich world, but economic stagnation means that breeds not growth but frustration. In fact, US workers who are functionally illiterate now e