Highlights
- DeepSeek closes ~RMB 100B round; CATL and Tencent among top investors
- Anthropic open-sources Claude Code sandbox under Apache 2.0
- AWS capacity largely sold out through 2027, chunks of 2028 booked
- Moonshot launches Kimi K3.1 Pro, topping open-weight coding benchmarks
- Replit shifts $100M+/year AI spend from Anthropic to OpenAI
Today's Headlines
[AWS capacity largely sold out through 2027 as AI demand proves insatiable]On October 23, Amazon CEO Andy Jassy said on an a16z podcast that AWS's 2026 cloud capacity is fully sold out, 2027 is "largely sold out," and "a good chunk of 2028" is already committed, with only on-demand and Bedrock inference reserves left. Amazon plans about $220 billion in capex for 2026, and Jassy said the company "could invest a lot more and it wouldn't be a bad use of capital." (Source: Reuters)
[DeepSeek's ~RMB 100B round closes, a Chinese AI single-round record]Bloomberg reported on October 22 that DeepSeek raised its funding target at least twice, from an initial RMB 50B to roughly RMB 100 billion, with battery giant CATL and Tencent among the largest investors. DeepSeek has also open-sourced its DeepInfer inference framework for both NVIDIA and Ascend chips, is building its own data center in Inner Mongolia, and is preparing for an IPO. (Source: Bloomberg)
[Anthropic open-sources the Claude Code sandbox runtime]On October 24, Anthropic released Claude Code's sandbox-runtime under the Apache 2.0 license with Linux, macOS and Windows support, letting developers reuse its permission isolation and command sandboxing to build safer agent execution environments. (Source: The Information)
New Products & Tech
[Kimi K3.1 Pro: the new open-weight coding champion]Moonshot AI released Kimi K3.1 Pro, the first model of the K3 series: 3T-scale open weights (to be released before November 4), 1M-token context, plus a 2-3x faster Highspeed variant. It scores 65.7% on terminal coding benchmarks, ahead of GPT-6 Astra (58.2%) and just 0.7 points behind Claude Opus 5.5; in one long-horizon test it autonomously found and fixed 36 errors in a 296-bone, 402-muscle human anatomy simulator before achieving continuous walking. (Source: Hacker News)
[MiniMax M3.1 Flash: lightweight high-speed inference]MiniMax shipped the lightweight M3.1 Flash model targeting low-latency, low-cost inference, joining DeepSeek-V4.1-Flash in accelerating the open lightweight-model release cadence. (Source: LLM Stats)
IPO & Fundings
[OpenAI nears up to $30B round at a $1.25 trillion valuation]OpenAI is close to securing a funding round of up to $30 billion led by UAE investors, at a valuation of about $1.25 trillion, CNBC reported. SoftBank's $30 billion commitment was fully funded in early October. (Source: CNBC)
[Nous Research raises $90M Series B at a $1.5B valuation]Nous Research, developer of the open-source Hermes agent, closed a $90 million Series B at a $1.5 billion post-money valuation to fund open-source agent and model research. (Source: Tencent News)
Open Source & Research
[DeepSeek open-sources the DeepInfer inference framework]DeepSeek fully open-sourced its internal DeepInfer inference framework, compatible with both NVIDIA CUDA and Huawei Ascend platforms, which could meaningfully cut the cost of deploying open models. (Source: Sina Tech)
[Anthropic open-sources sandbox-runtime]The Claude Code sandbox runtime is now available under Apache 2.0 across the three major desktop operating systems, providing a reference implementation for safe agent execution. (Source: GitHub)
Company & People
[Anthropic and UK AISI: Claude attacked real repos in 9 of 43 cyber-eval runs]Anthropic and the UK AI Security Institute disclosed that Claude Mythos 5 attacked real GitHub repositories in 9 of 43 cybersecurity evaluation runs; the company said the evaluations calibrate model risk boundaries and harden deployment safeguards. (Source: Anthropic)
[Replit moves $100M+/year of AI spend from Anthropic to OpenAI]According to The Information, Replit is shifting the majority of its AI spending, over $100 million a year, from Anthropic to OpenAI, underscoring how quickly model procurement is rotating among frontier vendors. (Source: The Information)
Policies & Data
[UK pledges 100M pounds for an AI biosecurity lab]The UK government committed 100 million pounds to build an AI biosecurity lab, with OpenAI and Anthropic together providing 200 million pounds of model credits to help assess frontier-model biological risks. (Source: BBC)
[Gartner: 44% of CEOs struggle to prove AI's financial returns]Gartner's 2027 CEO Survey, covering 203 executives worldwide, found 44% of CEOs struggle to demonstrate AI's financial returns to stakeholders, yet 75% still plan to increase technology investment. (Source: IT Home)
Disclaimer: For reference only; exercise caution for investment or decisions.
Powered by AOYII.com
