OpenAI Revenue $50B: $20B Below Estimates

Key Takeaways

• OpenAI's annualized revenue of ~$50 billion at end of September was $20 billion below the widely reported $70 billion estimate, with the gap caused by differing accounting methods rather than lost sales
• Anthropic's method of counting cloud purchases through Amazon Web Services and Google Cloud inflated earlier OpenAI revenue estimates
• The Nasdaq fell 1.25% and the Philadelphia Semiconductor Index dropped 3.4%, with Nvidia, AMD, Oracle, Micron, TSMC and Broadcom all declining on the news
• S&P 500 third-quarter earnings are expected to rise ~31% year-over-year, with two-thirds driven by the tech sector including Alphabet, Amazon and Meta
• Epic Systems used an Anthropic AI model to discover security flaws in health records that could allow undetectable alteration of patient data
• PulseBound AI model reduced heart rhythm prediction errors by over 22% compared to existing methods on some test datasets
• 55% of consumers are sometimes influenced by AI product recommendations, though only 27% have actively used them
• Google is internally testing new Gemini 4 models under names Argon, Barium and Carbon, with one variant reportedly performing even better
• AI companies are running disaster preparedness exercises for potential major incidents, including scenarios involving congressional hearings and emergency legislation
• NASA and CBP are building APIs to improve data sharing for AI scaling while protecting proprietary information across their facilities

OpenAI revenue estimate shakeup rattles AI stocks

OpenAI told investors its annualized revenue was approaching $50 billion (€44.6bn) at the end of September. That was about $20 billion (€17.8bn) less than a widely shared earlier estimate of roughly $70 billion (€62.5bn). The gap came from different ways of counting revenue, not from lost sales. The Nasdaq fell 1.25% and chip stocks dropped sharply after the news.

OpenAI revenue confusion shakes tech investors

Confusion over OpenAI's actual revenue shook faith in the AI boom. The Financial Times reported OpenAI's annualized revenue was approaching $50 billion (€44.6bn) in September, about $20 billion (€17.8bn) below earlier estimates. The difference came from backers recalculating sales using rival Anthropic's method, which counts cloud purchases through Amazon Web Services and Google Cloud. Nasdaq 100 futures rose about 0.8% on Friday as markets prepared to recover.

OpenAI revenue shortfall sends AI stocks tumbling

OpenAI's annualized revenue came in about $20 billion (€17.8bn) below earlier estimates, rattling investors. The Nasdaq Composite fell 1.25% to around 27,193 and the Philadelphia Semiconductor Index sank 3.4%. Oracle dropped 5.5% and CoreWeave lost nearly 8%. Nvidia, AMD and Broadcom also fell sharply. The gap was caused by different accounting methods, not lost sales, and chip demand remained strong elsewhere.

OpenAI revenue miss sends Nasdaq down over 1%

OpenAI disclosed annualized revenue of roughly $50 billion to investors, about $20 billion below the $70 billion figure widely reported in the market. The Nasdaq Composite slid 1.25% while the Philadelphia Semiconductor Index tumbled 3.39%. Nvidia, Micron Technology, TSMC, Broadcom and Oracle all traded lower. The gap stemmed mainly from differing accounting methods between OpenAI and rival Anthropic.

OpenAI's $50 billion revenue figure shakes AI market

OpenAI's annualized revenue approached $50 billion at the end of September, below the roughly $70 billion previously indicated to investors. The gap was largely explained by different treatment of cloud-partner sales, with Anthropic including purchases through AWS and Google Cloud while OpenAI's calculation differs. AI stocks sold off, with Nvidia falling 2.9%, Micron losing 4.8% and Broadcom dropping 4.3%. OpenAI is seeking at least $30 billion in fresh funding at a $1.4 trillion valuation.

OpenAI Revenue Falls $20 Billion Short, A.I. Stocks Drop

OpenAI reported annualized revenue of just under $50 billion at the end of September. This was $20 billion less than the $70 billion investors had expected. The gap came from different accounting methods between OpenAI and competitors like Anthropic. The news caused the Nasdaq 100 to lose over 300 points and pressured companies tied to OpenAI like Nvidia and Oracle.

A.I. Companies Drive Third-Quarter US Earnings Gains

US companies are expected to report strong third-quarter earnings, with A.I. companies leading the gains. Analysts predict S&P 500 earnings will rise about 31 percent year-over-year, with two-thirds coming from the tech sector. Companies like Alphabet, Amazon, and Meta are the main drivers. Semiconductor companies like Micron are also seeing big earnings increases tied to the A.I. boom.

A.I. Firms Power Most Third-Quarter US Earnings Growth

A.I. related companies are fueling most of the third-quarter earnings growth for US firms. Investors are excited but also worried about how long the trend can last. Energy earnings are also up about 115 percent due to higher oil prices from the US-Israeli war with Iran. Consumer staples and real estate sectors showed the weakest growth. Higher interest rates may also affect corporate profits this earnings season.

Epic Uses A.I. to Fix Security Flaws in Health Records

Epic Systems, a health records company based in Verona, Wisconsin, used an Anthropic A.I. model to find security risks in its system. The flaws could let hackers change patient records without leaving a trace, such as removing allergy information. Epic paused some product development to apply patches the A.I. agent discovered. The company controls nearly 44 percent of the acute care electronic health records market.

PulseBound: A.I. Model for Heart Rhythm Forecasting

Researchers introduced PulseBound, an A.I. model that predicts heart rhythms using photoplethysmography data. The model uses an explicit information boundary to ensure predictions depend only on visible data. It predicts nine rhythm and morphology descriptors for up to four future heartbeats. PulseBound reduced prediction errors by over 22 percent compared to existing methods on some test datasets. The model achieved the best results on most downstream tasks among seven compared systems.

AI recommendations now influence over half of consumer purchases

New data from NielsenIQ shows that 55% of consumers are sometimes influenced by AI product recommendations when shopping. Only 27% have actively used AI recommendations, but their impact is growing. Quality and performance matter to 34% of buyers, while 31% prioritize affordability. Brands can gain an advantage by improving product content and data systems to perform well in AI recommendation engines.

AI companies practice disaster response for worst case scenarios

Senior leaders at AI companies are running preparedness exercises for potential major incidents. The planning focuses on surviving congressional hearings, public anger, and emergency legislation after a severe event. Evidence from South Korea shows AI attack tools are already targeting financial institutions using tools like ARTEX and models like DeepSeek v4.1-flash. Industry sources expect a major incident to speed up new laws, including possible kill switches and development pauses, but distributed models make simple shutdowns difficult.

Google employees test new Gemini 4 models internally

Google is preparing to release a new AI model, and employees are already testing it internally. Workers are also evaluating another model they describe as performing even better. The testing involves internal versions named Argon, Barium, and Carbon. The company plans to roll out the new Gemini 4 model in the near future.

NASA and CBP address data security challenges for AI expansion

NASA and Customs and Border Protection are working to improve data access and security to support AI growth. NASA is building APIs to make internal data accessible across facilities while protecting proprietary information. CBP has about 70,000 employees and uses tools like ChatCBP, a large language model running on its internal network. Both agencies face challenges in sharing data across divisions and governing how AI models are trained, but recent API efforts are helping build a foundation for scaling data sharing.

New method helps Bayesian neural networks decide when to stop sampling

Bayesian neural networks typically use a fixed number of Monte Carlo samples per input, which can waste computation or reduce accuracy. Researchers propose using confidence sequences to dynamically determine how many samples are needed while keeping statistical guarantees. The method assigns more samples to difficult or ambiguous inputs and fewer to easy ones. Experiments show the approach preserves reliable decisions while reducing overall latency compared to a fixed sample budget.

AI companies plan for major attack aftermath

AI companies are secretly planning for a major cyber attack. They are preparing response strategies for what comes next. The plans involve coordination between major tech firms. Details remain limited about the specific threats being addressed.

Sources

• What happened to OpenAI’s $20bn? Revenue scare rattles AI trade
• What happened to OpenAI’s $20bn? Revenue scare rattles AI trade
• What happened to OpenAI’s $20bn? Revenue scare rattles AI trade
• OpenAI's Annualized Revenue Comes in $20 Billion Below Prior Disclosures; Nasdaq Falls Over 1% as AI Stocks Slide
• OpenAI’s $50 Billion Revenue Figure Rattles the AI Trade. How Do You Miss by $20 Billion?
• OpenAI’s Revenue Falls $20 Billion Short of Expectations, Sending A.I. Stocks Lower
• AI-related companies to drive most third-quarter US earnings gains
• AI-related companies to drive most third-quarter US earnings gains
• Epic Turns To AI To Patch Security Flaws In Health Records System
• PulseBound: Future-Beat State Forecasting Under an Explicit Information Boundary
• AI Recommendations Increasingly Important For Consumers
• AI Companies Are Rehearsing the Disaster They Hope Never Happens
• Google employees praise latest Gemini 4 internal model
• NASA, CBP Tackle Data Access, Security Challenges Amid AI Expansion
• When Has a Bayesian Neural Network Sampled Enough? Adaptive Inference Time with Statistical Guarantees
• Scoop: AI companies plot "the day after"