OpenAI’s revenue machine keeps accelerating, Apple quietly built a China-specific AI model with Alibaba’s help, and Nvidia just slashed its Ohio data center guarantee by more than half. The throughline this week: companies are turning AI hype into financial commitments, and the numbers are getting serious enough that investors are pushing back.
OpenAI Crosses $40B Annualized Revenue as IPO Looms
Bloomberg reported last Thursday that OpenAI’s annual revenue run rate has climbed above $40 billion, nearly double the $20 billion figure CFO Sarah Friar shared at the end of 2025. The growth comes from AI coding tools (Codex), ChatGPT Work subscriptions, and a nascent advertising business that just expanded to five more countries.
President Greg Brockman said in an internal announcement that annual revenue run rate increased more than 20% month over month in July alone. The company also appointed Dali Rajic as its new Chief Revenue Officer, poaching him from Wiz (recently acquired by Google). He’s the second person in that role in under a year. OpenAI now reaches over one billion weekly active users and two million businesses.
Here’s what this tells us: the IPO race between OpenAI and Anthropic is real, and both are padding their financials. Anthropic reported a $47 billion run rate in May. Both companies have filed confidential IPO paperwork. OpenAI’s $852 billion valuation looks less insane when you’re doing $40 billion a year, but they’re still unprofitable, and 42 state attorneys general have already issued subpoenas ahead of a potential public offering. The revenue is impressive. The legal exposure is staggering.
Apple Trained Its Own China AI Model With Alibaba’s Help
Apple built a China-specific large language model with training support from Alibaba, according to Reuters. The Cyberspace Administration of China registered Apple’s generative AI service last month, making Apple the first foreign company Beijing has cleared to run a proprietary AI model inside the country. Apple Intelligence is expected to reach Chinese iPhones through an iOS update in the coming months.
This is a notable break from Apple’s earlier strategy, which leaned on local partners to supply the models. ChatGPT and Claude aren’t available in China, so Apple had no outside partners to fall back on. Huawei has been shipping AI-equipped handsets for two years while Apple sold none. The gap mattered: Apple still trails Huawei’s 22.6% market share with its own 18.1%, though iPhone shipments in China grew 24.4% in Q2 thanks to steady pricing.
This is an infrastructure play, not a feature play. Apple isn’t trying to win the model race. It’s trying to not lose the China market. Running a proprietary model that passes Chinese content checks, while also supporting Qwen and Baidu alongside it, tells you how complicated that market has become for Western tech companies. Alibaba chairman Joe Tsai confirmed the partnership back in February 2025, saying “in the end they chose to do business with us.” Eighteen months later, the model surfaced.
Nvidia Cuts OpenAI Ohio Guarantee From $250B to Under $120B
Nvidia slashed its planned financial guarantee for OpenAI’s Ohio data center project from a previously discussed $250 billion to less than $120 billion, according to The Wall Street Journal. The chipmaker will back only the first phase of the 10-gigawatt campus, which SB Energy (a SoftBank subsidiary) is developing. A deal could be signed as early as this weekend.
The revision followed investor concerns about Nvidia’s exposure to large AI financing commitments. Seaport Research analyst Jay Goldberg warned in late July that vendor-financing arrangements can backfire, citing Lucent Technologies as a cautionary example. Nvidia’s August 10 financing partnership with six major financial institutions aims to raise more than $500 billion in third-party capital, shifting financing responsibility to outside investors rather than carrying it on Nvidia’s balance sheet.
The original $250 billion guarantee surfaced in July. Cutting it by more than half in three weeks tells you the pushback from investors was serious. Nvidia is trying to support AI infrastructure expansion without becoming the bank for the entire industry. OpenAI, still unprofitable at an $852 billion valuation, faces its own funding pressure as it negotiates a binding lease for the full site. The Ohio campus could become the largest announced data center project if completed. Right now, nobody wants to be the one holding the bag.
OpenAI Launches Ultrafast Mode: GPT-5.6 Sol at 14X Speed
OpenAI introduced Ultrafast on August 13, a preview service tier that runs GPT-5.6 Sol up to 14 times faster than standard processing, generating up to 750 output tokens per second. The mode is powered by Cerebras hardware, building on a partnership announced in January that added 750 megawatts of low-latency AI compute to OpenAI’s platform.
Early customers include Jane Street, where John Crepezzi called the speed increase “impressive” for focused model work, and Podium, where product lead Courtland Lykins said it “completely changes the call experience” for voice AI. OpenAI listed incident response, financial research, customer support, commerce, and live research as primary use cases. The preview is limited to a small group of API customers, with broader access planned as capacity grows.
This matters because it breaks the speed-intelligence tradeoff. Until now, real-time performance meant choosing a smaller or less capable model. Ultrafast lets you run frontier intelligence at speeds that make interactive workflows viable. The Cerebras partnership is paying off: GPT-5.3-Codex-Spark already exceeded 1,000 tokens per second on Cerebras hardware in February. The pattern here is clear. OpenAI is building speed tiers the way cloud providers built latency tiers. Different prices for different speeds, same intelligence underneath.
OpenAI’s Enterprise Data Shows AI Moving From Assistance to Execution
OpenAI published two studies on August 12 examining enterprise AI adoption. The headline finding: frontier firms (top 10% of AI usage) now generate 8.3x as many output tokens per active user as typical firms, up from 2.6x in January. The gap is widening. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers, indicating a shift from Q&A to delegated multi-step work.
The sector breakdown is telling. Weekly active enterprise Codex users grew 108x in legal, 41x in sales, 41x in recruiting, and 26x in marketing since February, compared to 5x in engineering. Agents are spreading beyond engineering into knowledge work. Usage is highest among early-career employees and falls among senior staff, suggesting younger workers have a comparative advantage in adopting AI tools.
Quick Hits
Anthropic – Claude Fable 5 and Mythos 5 redeployed globally on July 1 after US export controls were lifted. Anthropic also proposed an industry-wide framework for scoring jailbreak severity, developed with Amazon, Microsoft, Google, and other Glasswing partners. The framework comes after Amazon researchers found a safeguard bypass in Fable 5 that triggered the original government restrictions.
Google DeepMind – Introduced Gemini 3.7 Flash in August, the latest in a rapid cadence of model releases (3.5, 3.6, and 3.7 Flash all shipped since July). Also announced sign language AI for users and WeatherNext, an AI model achieving breakthrough forecasting for cyclones.
Hugging Face – Published a new post on August 13 about combining Strands Agents, LeRobot, and Hugging Face Storage Buckets into a single workflow for recording, training, and deploying robotics models from one place.
Mistral AI – Announced in-region inference, open models, and new European infrastructure for sovereign AI. The company is positioning itself as Europe’s domestic AI champion with infrastructure, open models, and long-term commitments for the EU to control its AI future. Also teased Shieldstral and a system of record for prompts and skills.
OpenAI – ChatGPT ads expanded to the UK, Mexico, Brazil, Japan, and South Korea as of August 11. The ad pilot launched in the US in February and has shown “no impact on consumer trust metrics” with low dismissal rates. Plus, Pro and Business tiers remain ad-free. Ads fund free and Go tiers. The commercial machine keeps growing.
Rundown for August 17, 2026. Sources: Yellow, OpenAI, Anthropic, Google DeepMind, Hugging Face, Mistral AI.