Closing Time: The $18 billion gap that sank AI stocks
OpenAI told investors it hit roughly $50 billion in annualized revenue at the end of September, CNBC confirmed after the Financial Times first reported the figure.
The most important number in AI got smaller today, and the market noticed. OpenAI told investors it hit roughly $50 billion in annualized revenue at the end of September, CNBC confirmed after the Financial Times first reported the figure. That is well below the $68 billion number that circulated widely late last month. A person familiar with the matter said the bigger figure counted gross revenue from OpenAI’s partners, a method that makes for a more direct comparison with Anthropic, which told investors in August it had reached a $65 billion run rate.
AI stocks sold off as the report spread through the afternoon. By the close, Nvidia was down 2.9%, Oracle 5.5%, Broadcom 4.3%, Intel 5.3% and CoreWeave 7.8%. OpenAI’s investor presentation still had good news in it: run-rate revenue grew 77% in the third quarter, and its enterprise business grew 107%. But the company is valued at $852 billion and is widely expected to go public in 2027, so every revenue figure is now being read as a test of that price tag.
The bigger issue is how much of the AI build-out leans on one company’s growth. Chipmakers, cloud providers and data center lenders have signed huge deals on the assumption that OpenAI’s sales keep climbing fast. A $50 billion run rate is still extraordinary for a company this young. Today showed how little room investors are leaving for a number that merely looks smaller than expected.
Elsewhere in AI today:
TechCrunch reports Google launched a Gemini agent for businesses that can be handed goals rather than step-by-step instructions, then plan the work and connect to tools like Google Workspace, Microsoft 365, Slack, Jira and Snowflake. The agent gets its own Workspace account and email address, like a new co-worker, and users can switch it to other models, starting with Anthropic’s Claude. Google says Gemini has more than 1 billion monthly users and plans to bring the agent to consumers later.
CyberScoop reports OpenAI shut down two influence operations, one from Russia and one from Iran, that used ChatGPT and other AI tools to invent fake journalists and think tanks. Both managed to place stories in legitimate news outlets; the Iranian campaign alone used at least seven fake journalist personas. OpenAI rated them 5 and 4 on its 6-point severity scale, the first time it has reported campaigns it considers high-impact.
Semafor says the FTC is close to sending investigative demands to frontier AI companies, according to a senior agency official. The demands, which carry subpoena-like force, run dozens of questions long and cover what the companies have said about the dangers of their products and other consumer harms. Critics question how aggressive the probe will be, since it surfaced a day after the FTC chair joined AI CEOs at the White House to announce a voluntary industry accord.
The Verge reports USA Today Co. and several of its local papers, including The Tennessean and The Columbus Dispatch, sued OpenAI for more than $250 million, alleging it copied “hundreds of thousands” of articles to train its models. The suit joins a long list of copyright cases against OpenAI, from The New York Times to a coalition of nearly 400 local newspapers.
Anthropic launched what it calls the Cyber Mission, starting with a program that sends its models, engineers and threat research to the people who defend power grids, water systems and transportation networks. It also opened OSS Scanner, which gives critical open-source projects free, regular security scans from its strongest models. The reports will be fully model-generated with no human review, so Anthropic warns some may be wrong.
The Verge notes Claude can now add motion to what it makes, producing things like a 30-second explainer video, an animated chart or a product walkthrough that downloads as an MP4. Anthropic is also rolling out live dashboards that update from a company’s own data and taking Docs, Slides and Design out of beta.
TechCrunch reports Arena, the popular AI model leaderboard formerly called LMArena, raised $200 million at a $3.1 billion valuation, nearly doubling its value in 10 months. Lightspeed Venture Partners and Khosla Ventures co-led the round. The company also previewed an Arena Alignment Index, a new ranking meant to measure whether AI models can be trusted to behave safely.
Markets
Software held up while AI hardware took the hit on Thursday. Accenture rose 6.0% and Atlassian 4.0%, while Coherent and Astera Labs, which make the optical and connectivity parts that link AI chips together, each fell more than 9% and Nebius dropped more than 7% among the names we track.
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| ACN | Accenture | 208.36 | +5.96% |
| TEAM | Atlassian | 203.57 | +3.97% |
| ADBE | Adobe | 241.05 | +3.56% |
| SNOW | Snowflake | 343.40 | +3.17% |
| IBM | IBM | 226.61 | +2.77% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| COHR | Coherent | 302.35 | -9.63% |
| ALAB | Astera Labs | 347.05 | -9.21% |
| CRWV | CoreWeave | 81.58 | -7.77% |
| NBIS | Nebius | 219.71 | -7.35% |
| ARM | Arm | 275.29 | -6.48% |
Prices are Thursday’s U.S. cash close, October 8, 2026 (vs. Wednesday’s close). Market data is a snapshot, not live.
Deals Desk
Two compute and power deals from today worth knowing.
Sesterce × Anchor customer (not disclosed) — Oct 8, 2026
AI data center campus on the site of the former Kaipola paper mill · Jämsä, Finland
200 MW first phase, rising to 600 MW in phase two; about $10B investment; Sesterce targets more than 1 GW in Finland long term
Status · Announced · Source: DCD
SK Securities, Metheus Asset Management and SeedH × City of Sejong — Oct 8, 2026
Investment agreement to build an AI data center · Sejong, South Korea
360 MW; 7 trillion won (about $4.9B) through 2029
Status · MOU signed · Source: Seoul Economic Daily
Full tracker → https://www.theaigentic.com/deals
Aigentic Exclusive
FEATURED DEEP DIVE
Local AI, Big Tech Wants AI on Your Computer
Microsoft and Nvidia are bringing more AI work onto personal computers. The promise of greater control comes with a new calculation about cost, privacy and who keeps the system running.
Aigentic Exclusive
FEATURED HANDS ON
Claude Moves Into Google Docs, Sheets and Slides
A practical guide to using Anthropic’s new add-on to sharpen a memo, make sense of a spreadsheet and repair a crowded presentation.
Watch
Nate B Jones — Why would Google pay $10 million for old work emails?
Nate looks at a reported $10 million Google bid for Spirit Airlines’ old work emails and Teams records to train AI agents, and argues that much of the value people create at work never shows up in those records.
The AI Daily Brief — The best way to test new AI models
In this Operator’s Cut, Nufar Gaspar walks through a repeatable way to test new models on your own tasks, comparing output quality, speed and cost to decide which ones earn a place in your work.