Daily AI news, Friday, 23 October 2026

Good afternoon, here's today's AI, tech and VC briefing. OpenAI is close to raising up to $30 billion at a $1.25 trillion valuation, which would make it the most valuable private company ever. SpaceX is reportedly lining up 80,000 US cell sites for its own mobile network, and telecom shares fell hard. AI stocks are wobbling after Meta's results, Anthropic is handing its Agent Skills standard to an open foundation, Microsoft is rolling out a prompt-to-app builder, and a new benchmark puts Claude Code on top. Plus a big week for funding, and a16z drops "crypto" from its crypto fund's name.

LEAD: OpenAI nears up to $30 billion at a $1.25 trillion valuation

Background: OpenAI last raised $122 billion in March at an $852 billion valuation, and filed confidential IPO paperwork in June. Last month Bloomberg reported it was seeking about $30 billion at roughly $1.4 trillion, and we covered on Monday that it was racing to raise before Anthropic goes public. What happened: CNBC reported late on Thursday that OpenAI is close to finalising a round of up to $30 billion at a $1.25 trillion pre-money valuation, led by investors from the UAE alongside Coatue. OpenAI told investors it aims to get signatures by the end of next week, and the final size is still being set. Bloomberg says the valuation in talks came down by $150 billion from $1.4 trillion. OpenAI is also negotiating a deal for more computing capacity in the UAE. The round is being pitched as an "opportunistic" bridge before an IPO. Why it matters: Even after the trim, this would be the highest valuation ever for a private company, and it sets a very high bar for OpenAI's eventual listing. The lower number is also a sign that investors are pushing back on price as AI stocks swing. Sources: CNBC · Bloomberg · TechCrunch

2. SpaceX reportedly wants 80,000 US cell sites for a fourth national carrier, and telecom shares slide

Background: SpaceX already sells Starlink internet and has been buying nationwide low-band spectrum for phones, with approval for thousands of mobile-capable satellites. What happened: Bloomberg reported on Thursday that SpaceX is in talks with tower companies, including Crown Castle, to get access to at least 80,000 US sites, about half of all purpose-built US cell towers. That would put it on a path to becoming a fourth nationwide mobile carrier. AT&T, Verizon and T-Mobile fell about 5% each before the US market opened on Friday, while Crown Castle rose about 10% and American Tower about 7%. TD Cowen estimates a nationwide ground network would cost more than $80 billion and take years. Why it matters: SpaceX is fresh off a strong quarter and is now aiming at the US phone market itself, not just satellite add-ons. A carrier that mixes satellites with towers would be a real threat to the three incumbents. Sources: Reuters via Investing.com · Forbes · Reuters markets

3. AI stocks sell off after Meta's first cash outflow in over a decade

Background: AI shares have swung all month, first on worries about CoreWeave's debt and the cost of funding AI, then rallying on Wednesday after results from SpaceX and Super Micro. What happened: On Thursday the Nasdaq fell nearly 3%, and Meta dropped almost 8% after reporting its first cash outflow in more than a decade because of AI spending. US stocks were set to open lower again on Friday. Microsoft, Alphabet and Amazon all report next week. Why it matters: Investors are now asking how the giant AI build-out gets paid for, not just whether demand is there. Next week's results from the three biggest cloud companies will be the real test. Sources: Reuters via Investing.com · Reuters on AI spending

4. Anthropic gives Agent Skills to an open foundation, and Claude gets a new Google Cloud option

Background: Agent Skills is Anthropic's open format for teaching an AI agent a task: a folder with a SKILL.md file plus any scripts it needs. It came out of Claude Code. What happened: Anthropic is donating the standard to the Agentic AI Foundation, run under the Linux Foundation, which it co-founded with Block and OpenAI. It says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, already support it, with about 1 million skills published and nearly 280 million installs. Separately, Google Cloud made an Anthropic-run version of Claude generally available, so customers get new Claude features such as Managed Agents on day one while still paying through Google Cloud. Why it matters: Handing the format to a neutral body makes it easier for rivals to keep adopting it, which helps skills become the standard way agents learn company workflows. The Google deal widens Claude's reach inside big cloud customers. Sources: Anthropic · Google Cloud

5. Competitor product moves

Microsoft Copilot Code: Microsoft has started rolling out a "Code" tab in the Copilot app to its early-access Frontier users. People describe a dashboard, internal tool, website or script in plain words and refine it through chat with a live preview; it can use Microsoft 365 work data and connect to GitHub. This puts Microsoft directly into the prompt-to-app space of Replit, Lovable and Cursor. Microsoft Learn New coding-agent leaderboard: Artificial Analysis launched version 2.0 of its Coding Agent Index, which now uses Cognition's FrontierCode test (graded on whether a maintainer would merge the change), Terminal-Bench 4.0 and a code Q&A test. Claude Code with Opus 5.5 leads at 61.7, at about $11.42 per task, ahead of Claude Sonnet 5.5 at 59.3. Codex running DeepSeek V4.1 Flash scored 47.3 for about $0.39 per task. Artificial Analysis · AITOP Catch-up on Cursor: The Information reported last week that Cursor staff investigated, in late July before the SpaceX deal closed, whether employees testing an internal assistant could have sent confidential customer or partner data to SpaceX's AI unit for training. No exposure has been confirmed; Cursor responded by requiring company-email logins and stopping training on employee-generated data. Data handling is a trust issue every coding tool owned by a model maker will face. The Information

6. Big rounds and VC moves

Warp Sciences: Former OpenAI science lead Kevin Weil launched a start-up that trains robots to make and test drug compounds, raising $215 million at an $865 million valuation, led by Lightspeed. TechCrunch Whop raised $205 million led by Sequoia at a $5.1 billion valuation; Xona, building a satellite rival to GPS, raised $250 million at $2.45 billion; and Warhound, from Colossal's Ben Lamm and Anduril's Palmer Luckey, emerged with $200 million. Midu · StartupReader a16z Frontier: Andreessen Horowitz renamed a16z Crypto as a16z Frontier and widened its $2.2 billion fund to cover AI, robotics and other new technology, while still backing crypto. It follows similar moves by Paradigm and Haun Ventures. Fortune · TechCrunch

7. China clears DeepSeek and MiniMax models for global use

What happened: A social media account linked to Chinese state broadcaster CCTV said DeepSeek and MiniMax have won state approval for global use of their AI models. China proposed export controls on 23 AI technologies a week ago, but said most would stay exportable, with simpler licensing for open-source models. This comes days after the US ordered federal agencies to remove DeepSeek. Why it matters: Beijing is signalling it wants its cheap open models used abroad, under its own rules, even as Washington pushes them out of government. Source: Reuters via Investing.com

Key hires

No new senior hires at the big coding or AI-agent companies were confirmed today. On the VC side, a16z general partner Ali Yahya, the first full-time member of its crypto team, has told some investors he plans to leave to start a company, according to The Information as cited by Fortune.

Coming up

OpenAI aims to sign its funding round by the end of next week. Microsoft, Alphabet and Amazon report earnings next week, and the US Federal Reserve meets too. OpenAI's model access for Cursor ends on 12 November.