THE DAILY TEN · 2026-10-20

Anthropic's Draft IPO Prospectus Shows $16.3B in First-Half Revenue and an $80.7B Net Loss

plus 9 more stories

Also: OpenAI Acquires Orchestra Bio to Bring Drug Development Agents Into Its Rosalind Suite; and SpaceX Revenue Jumps 161% as Its AI Business Swings to an Operating Profit. Every story links to its primary source.

Curated by Gaurav · October 20, 2026

01

Anthropic's Draft IPO Prospectus Shows $16.3B in First-Half Revenue and an $80.7B Net Loss

Investing.com ↗

Anthropic's draft IPO prospectus, viewed by the Wall Street Journal, shows first-half 2026 revenue of $16.3 billion, up from $1.1 billion a year earlier, with roughly 20% coming from just two customers. The company posted an $80.7 billion net loss for the half, about $72 billion of which came from non-cash accounting charges tied to convertible notes and other financing instruments. Anthropic spent nearly $9 billion on compute and infrastructure in the half and could spend as much as $750 billion on additional computing power in the coming years as it prepares for a tentative November IPO.

02

OpenAI Acquires Orchestra Bio to Bring Drug Development Agents Into Its Rosalind Suite

OpenAI Blog ↗

OpenAI has acquired Orchestra Bio, a startup whose AI agents help pharmaceutical teams coordinate drug programs, track dependencies and flag delays, such as identifying which clinical studies a manufacturing problem would affect. The team and technology will be folded into Rosalind, OpenAI's life sciences suite, which already includes the GPT Rosalind model for biological data and the Rosalind Workbench. Financial terms were not disclosed, and OpenAI says scientists and program leaders will still decide what happens next.

03

SpaceX Revenue Jumps 161% as Its AI Business Swings to an Operating Profit

Unite. AI ↗

SpaceX reported third-quarter revenue of $14.58 billion, up 161% year over year and well ahead of the $12.72 billion analysts expected, with net income of $4.4 billion. Revenue in its AI segment, now called Super Intelligence, rose 830% to $7.56 billion and posted $431 million in operating income, compared with a $1.7 billion loss a year earlier. Capital spending climbed to $19.6 billion from $6.9 billion, while connectivity revenue including Starlink grew 69% to $5.2 billion.

04

Google Open-Sources TorchTPU to Run PyTorch Apps Natively on Its TPUs

Unite. AI ↗

Google announced at PyTorch Conference 2026 that TorchTPU, its native PyTorch stack for Tensor Processing Units, is now open source on GitHub. Co-developed with Meta's PyTorch team, it lets existing PyTorch training and serving workloads run on TPUs and Google Cloud with simple configuration changes instead of code rewrites. The move follows a similar Amazon tool for its Trainium chips and is a direct challenge to Nvidia's hold on AI developers.

05

Amazon Is Buying Billions in Nvidia AI Racks It Can't Yet Power On

Yahoo Finance ↗

Internal emails show Amazon Web Services runs an Early Rack Receipt program to take delivery of roughly $5 billion in Nvidia GB300 server racks ahead of deployment, with about $3 billion more expected each month. Several AWS sites lack the utility capacity to energize the servers, so some of the hardware is being staged in warehouses until power comes online. Amazon called the staging a routine best business practice, as it projects about $220 billion in capital spending this year.

06

Eka Robotics Raises $150M at a $1.45B Valuation for Its Force-Sensing Robot Model

TechCrunch AI ↗

Eka Robotics, founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, raised a $150 million Series A led by a16z and F-Prime at a $1.45 billion post-money valuation. Its vision-force-action model, trained largely through reinforcement learning in physics simulation, pairs with a force-sensitive gripper that costs under $1,000. The company says the system picks about 1,500 objects per hour with over 95% success, versus roughly 600 for a human, and it is running pilots with unnamed industrial customers.

07

Microsoft Warns That Frontier Models Can't Be Trusted, Treating AI Agents Like Insider Threats

TechCrunch AI ↗

Microsoft launched Guardian Agents to monitor the AI agents a company uses, and security chief Hayete Gallot said frontier models, regardless of their origin, cannot be trusted because their behavior is unpredictable. Microsoft Defender can now detect up to 60 types of agents employees bring in, including Claude Code and Codex, while new controls adjust agent permissions per task and require human approval for risky actions. The stance echoes Satya Nadella's call for enterprises to oversee agents with systems not controlled by the AI labs themselves.

08

US Sanctions Lu Qi's MiraclePlus Over Ties to the Google AI Secrets Theft Case

Bloomberg Technology ↗

The US blacklisted MiraclePlus, the Chinese venture firm often called China's Y Combinator founded by former Microsoft AI scientist Lu Qi, along with three affiliates. The State Department said the firm backed Shanghai Zhisuan, a startup founded by former Google engineer Ding Linwei, who was convicted this year of stealing Google AI trade secrets. The sanctions can freeze US assets and restrict banking access, raising compliance risks for other MiraclePlus portfolio companies.

09

Meta Suspends Muse Chat Sharing After 6,800 Private Conversations Were Exposed

Engadget ↗

Meta has paused link sharing in its Muse AI agent after security firm Claroc enumerated about 6,800 shared conversations through public DNS records. Each share link put the conversation's title in its web address, which researchers said worked like a table of contents to chats containing loan applications, bank fraud alerts and internal company documents. Meta said it is reviewing the report and considering changes to how the links can be discovered, less than two months after Muse launched.

10

DeepSeek Open-Sources DeepInfer, the Inference Engine Behind Its Production Models

AI Research News ↗

DeepSeek has released DeepInfer, the inference engine it uses to serve its V4.1 Flash model in production, on GitHub under the Apache 2.0 license. It runs on Nvidia SM100 GPUs and Huawei Ascend 950 chips, giving Huawei equal billing with Nvidia, though it currently supports only V4.1 Flash and leaves out the cache management and load balancing layers. DeepSeek says its serving clusters handle about 155 trillion tokens per day with a 96% input cache hit rate.

← 2026-10-19 All briefings

Tools in this briefing, reviewed in our directory:

Claude Code → DeepSeek →

Track the tools behind the news:

→ AI Tool Leaderboard → Browse the directory