Collection window: UTC 2026-10-19 20:48 – 2026-10-20 20:48 (last 24 hours). Publication times were checked on each article page and converted to UTC. Stories covered in the previous brief (AWS TorchNeuron, Tenstorrent, Mercury 3, Meta/No Kings) were excluded as duplicates. Mistral Voxtral 3 was left out because its page shows no publish time, and a Politico story on OpenAI and China export controls was left out because its timestamp could not be verified.
Anthropic’s draft S-1 reportedly shows $16.3B in H1 revenue and an $80.7B net loss
According to a draft IPO prospectus viewed by The Wall Street Journal, Anthropic booked $16.3 billion in revenue in the first half of 2026, up from $1.1 billion a year earlier. The $80.7 billion net loss is mostly accounting: about $72 billion stems from non-cash adjustments tied to convertible notes and other financing instruments. Compute and infrastructure spending reached nearly $9 billion in the half, more than all of 2025 ($7.3 billion), and the company has roughly $450 billion in cloud commitments over the next decade, including $111 billion with Google Cloud, $110 billion with AWS and $31 billion with Microsoft. About 20% of H1 revenue came from two customers, and the filing warns that damages from a severe AI security incident could exceed insurance limits.
Reference: Investing.com summary of WSJ report (2026-10-20 15:47 ET = 19:47 UTC) · WSJ original
Google open-sources TorchTPU, following Amazon in chipping away at Nvidia’s PyTorch advantage
Google open-sourced TorchTPU, which lets PyTorch apps run on its Tensor Processing Units with a configuration change rather than a rewrite. It worked with Meta, Hugging Face, Inferact (maker of vLLM) and RadixArk (maker of SGLang), and also shipped a TPU agent suite to help developers migrate and fix errors, plus a LiteRT update with better PyTorch support. The move lands one day after Amazon released TorchNeuron for Trainium, signaling a coordinated push by cloud providers to loosen Nvidia’s grip on PyTorch, which by some estimates underpins about 85% of AI apps.
Reference: TechCrunch (2026-10-20 11:38 PDT = 18:38 UTC)
OpenAI acquires drug-development agent startup Orchestra Bio to bolster its Rosalind suite
OpenAI acquired Orchestra Bio, a startup building AI agents that help pharmaceutical companies manage drug development. Its platform lets scientists hand off work such as study monitoring, timeline tracking and flagging potential delays to agents, and the team and technology will be folded into Rosalind, OpenAI’s life-sciences models and tools. Terms were not disclosed. The deal sharpens the biotech race with Anthropic, which bought Coefficient Bio for about $400 million in April and launched Claude Science in June.
Reference: Axios (2026-10-20 17:55 UTC)
Meta’s Muse passes 10 million U.S. downloads in its first 40 days
Sensor Tower data shows Meta’s AI companion and agent app Muse has reached 10.9 million downloads since its September debut, 10.2 million of them in the U.S. Only 11 other apps have crossed 10 million U.S. downloads within 40 days of launch, the most recent being Threads in July 2023. Over the same window ChatGPT saw 3.7 million U.S. downloads, Grok 500,000 and Claude 200,000. Muse’s 7-day retention of 30% beats the 19% rival average, daily actives topped 2.5 million for the first time on October 16, and Meta put 39% of its October ad spend behind the app.
Reference: TechCrunch (2026-10-20 10:12 PDT = 17:12 UTC)
DeepSeek open-sources DeepInfer, the inference engine behind its API, with Huawei Ascend support
DeepSeek published DeepInfer, its in-house production inference framework, on GitHub under Apache 2.0. It includes high-performance kernels, multi-chip distributed inference, speculative decoding and structured outputs, and runs on both Nvidia SM100 data-center GPUs and Huawei Ascend 950 chips. It currently supports only DeepSeek-V4.1-Flash and leaves out a cache system, prefill/decode disaggregation and cross-machine load balancing. Built mainly on vLLM with ideas from SGLang and TensorRT-LLM, the repo passed 8,000 stars by the afternoon of release.
Reference: The Next Web (2026-10-20 14:08 UTC) · DeepSeek on X (09:30 UTC)
Agent sandbox cloud Daytona hits a $1B valuation with a $60M Series B
Daytona, which provides cloud sandboxes where AI agents run code, browse the web and use software, raised a $60 million Series B led by FirstMark at a $1 billion valuation. The company says it can spin up a new sandbox in 60 milliseconds and that its revenue run rate grew 25-fold from January to September. Customers include Anthropic, OpenAI, Nvidia, Cursor, Cognition, Meta and Cohere. CEO Ivan Burazin argues that agents will drive demand for CPU-heavy infrastructure and expects a shortage.
Reference: Fortune (2026-10-20 08:00 ET = 12:00 UTC)
Physical-AI robotics startup Eka raises $150M Series A at a $1.45B valuation
Eka Robotics, founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, raised $150 million led by a16z and F-Prime. Its core is a vision-force-action model that leans on force sensing, paired with a force-sensitive gripper that costs under $1,000; the company says its robot picks about 1,500 objects per hour (versus roughly 600 for a human) with over 95% reliability. Training happens via reinforcement learning in physics simulation, generating thousands of hours of robot experience in minutes. Eka is running pilots with industrial customers on pick-and-place and wire manipulation.
Reference: TechCrunch (2026-10-20 06:00 PDT = 13:00 UTC)
ByteDance, Alibaba and Tencent order 3M+ Nvidia RTX PRO 5500 cards worth about $35B
Reuters reports that China’s three tech giants have ordered more than 3 million Nvidia RTX PRO 5500 workstation GPUs at about $11,400 each, a combined sticker price above $35 billion, and that total orders from Chinese customers exceed 6 million. Because the B300 cannot be sold to China and the H200 is licensed case by case, buyers plan to use this less powerful but exportable card mainly for AI inference. Nvidia has told customers it cannot guarantee supply, with one source saying partners may receive only 5% to 10% of requested volumes. Samples are expected in December, with commercial deliveries starting in Q1 2027.
Reference: The Edge Malaysia / Reuters (2026-10-20 17:10 UTC+8 = 09:10 UTC)
Today’s Summary
- Anthropic’s draft S-1 puts numbers on the AI-lab model: explosive revenue growth paired with enormous compute commitments.
- Open-source stacks that reduce Nvidia dependence keep arriving (Google TorchTPU, DeepSeek DeepInfer), even as Chinese giants bet $35B on Nvidia cards.
- Capital and users are flowing across the agent ecosystem: infrastructure (Daytona), biotech (Orchestra Bio), robotics (Eka) and consumer apps (Muse).
Leave a Reply