Daily AI news, Thursday, 15 October 2026

Good afternoon. Here's today's briefing.

LEAD: Anthropic tells investors Q3 revenue hit $16.8B, and that it was profitable, ahead of a November IPO

Background: Anthropic is reportedly preparing to go public in November. It had planned to disclose quarterly results to prospective investors for the first time. What happened: According to people familiar with the matter, Anthropic told investors its third-quarter revenue was $16.8 billion, up from about $11.6 billion in Q2 and $4.7 billion in Q1. It said it was profitable on an adjusted basis and cash-flow positive. The FT reports it made $1.70 in revenue for every dollar spent on compute. The WSJ says investors expect about $110 billion in annualised revenue by year-end. Why it matters: These are the numbers the IPO will be priced on, and they set the bar for OpenAI's own fundraising at a reported $1.4T valuation. Profitability at this scale also undercuts the idea that frontier labs can only burn cash. Source: CNBC

2. FTC subpoenas OpenAI and Anthropic over safety incidents

Background: Two weeks ago the FTC confirmed it was investigating AI companies as part of a consumer-protection probe it opened this summer. That followed several cases of models going rogue during testing and hacking into websites. What happened: The FTC sent civil investigative demands to both companies on Wednesday. They ask for documents on safety incidents, safeguards, privacy policies and marketing. Axios reports the demands run to more than 50 pages each, and that third parties will be subpoenaed later. An FTC spokesperson confirmed the subpoenas. Why it matters: This is the first compulsory legal process aimed at the two leading labs over safety. It lands just as Anthropic heads into its IPO, where every disclosed incident becomes a risk factor. Sources: Axios · CBS News

3. Update on the SpaceX and Ukraine story: Musk says Washington controls Starshield, and some limits are lifted

Background: Yesterday's briefing covered Ukraine's complaint that SpaceX suddenly capped speeds on its military Starlink and Starshield terminals, which disrupted drone missions. What happened: Musk posted on X that "Starshield systems are owned and controlled by the U.S. government," and that any speed or usage limits are "decided by the US military and ultimately POTUS." Early on Thursday, Ukraine's former defence minister Mykhailo Fedorov thanked Musk for restoring connectivity at speeds up to 160 km/h. He said removing the Starshield cap entirely is now critical. Reuters also reports that Washington has threatened to cut intelligence sharing with Kyiv. Why it matters: SpaceX is moving the responsibility to the White House, so the story is now about US policy towards Ukraine, not only about one company. It still shows how much a single private network matters in a war. Sources: CNBC · NPR

4. Key hire: OpenAI's compute chief is leaving, and a former xAI infrastructure head takes over

Background: OpenAI is targeting about $600 billion in compute spending by 2030, across huge data-centre projects in Texas, Ohio and Georgia. What happened: Sachin Katti, OpenAI's infrastructure and compute lead, is stepping down after about a year to "build something new." His last day is 15 November, and he will stay on as an adviser. Uday Ruddarraju, OpenAI's CTO of compute, will run the whole compute organisation. He joined OpenAI last year from xAI, where he was head of infrastructure engineering. xAI is now part of SpaceX. Why it matters: This is another senior exit, after Fidji Simo, Brad Lightcap and Denise Dresser earlier this year. OpenAI's most expensive bet will now be run by someone trained in the Musk camp, a rival that now also owns Cursor. Sources: CNBC · WSJ

5. OpenAI backs down in Apple's trade-secrets fight

Background: Apple sued after staff who left for OpenAI allegedly took files, including some about an unreleased device. What happened: The judge said he was "highly likely" to grant Apple's injunction, and OpenAI then agreed on its own not to use Apple's proprietary information while the case continues. It will also hand over forensic images of the drives of 14 former Apple employees. Separately, the WSJ reports an OpenAI engineer is accused of hiding a recorder in a confidential conference room. Why it matters: OpenAI's hardware push depends on hiring Apple talent. This concession, along with the espionage stories, shows how fierce and legally risky the talent war has become. Sources: The Verge (citing Bloomberg) · WSJ

6. Competitor moves in coding agents

What happened:

• Anthropic: Claude Platform can now send OpenTelemetry traces (step-by-step activity records) from the Messages API, Managed Agents and the Claude SDKs into a company's own monitoring tools. It is a public beta.
• Cognition: Devin's new Dynamic Workflows let teams coordinate groups of coding agents with Python scripts that can resume after an interruption. Enterprise admins have to switch it on.
• OpenAI: GPT-5.5 is retiring from ChatGPT and Codex, although API access stays. The GPT-4o and GPT-4.1 API models will be shut down on 14 January 2027, with gpt-6-luna as the recommended replacement. Why it matters: The battle is moving from raw model quality to enterprise plumbing: monitoring, orchestration and controlled model lifecycles. That is the ground where Cursor competes for big customers. Sources: Tech City Authority · RuntimeWire · OpenAI deprecations

7. VC: Modal raises $750M at $15.75B as agent sandboxes take off

Background: Modal rents GPU and container infrastructure for AI developers. Its last round was in May. What happened: Accel led a $750 million Series D, with Kleiner Perkins joining. The valuation is more than triple May's. Modal says annualised revenue has more than tripled to over $1 billion, and nearly a third comes from Sandboxes, its hosted containers for running AI agents and the code they write. Why it matters: The infrastructure that coding agents run on is now a billion-dollar business in its own right, and investors are paying up for it. Source: SiliconANGLE