August 18, 2026 —
OpenAI locks in an 8-gigawatt Ohio AI campus as Nvidia backs the build
Today's North America AI brief is led by an unusually large commitment to physical infrastructure: OpenAI has secured capacity at an Ohio campus that could reach roughly 8 gigawatts of IT load, with Nvidia supporting the initial buildout. The rest of the issue tracks the capital, products, safety controls, and workflow infrastructure moving alongside that expansion.
OpenAI locks in an 8-gigawatt Ohio AI campus as Nvidia backs the build
OpenAI said it has agreed to secure approximately 8 gigawatts of IT capacity at the PORTS-Pike Technology Campus in southern Ohio. SB Energy will build, own, and operate the site under a 20-year lease; the first 800 megawatts are expected in 2028. OpenAI says the campus will exclusively host Nvidia compute.
Nvidia will invest $1.5 billion in SB Energy and provide credit support for the land, power, and shell tied to the initial 4.25 gigawatts. OpenAI projects 35,000 construction jobs through 2032 and 2,500 long-term operating roles, while acknowledging that later phases depend on permits, financing, generation, and transmission buildout.
Anthropic's reported revenue run rate clears $65 billion
Anthropic's annualized revenue run rate surpassed $65 billion at the end of July, according to reporting by Axios and TechCrunch citing figures first reported by Bloomberg. Axios said preliminary second-quarter revenue exceeded $11.5 billion, while TechCrunch noted that the run rate had climbed from $47 billion in May and $9 billion at the end of 2025. Anthropic did not publicly confirm the latest figure.
Groq raises $350 million for its post-chip inference cloud
Groq announced a $350 million Series A led by Disruptive, with planned participation from Nvidia, valuing the reconfigured company at $3.5 billion. The company now operates 13 data centers and says the financing will help expand from 54 megawatts today to more than 200 megawatts in 2027. Groq is positioning itself as a Nvidia-powered inference cloud after its earlier chip-focused business changed course.
OpenAI launches a dedicated ChatGPT experience for teens
OpenAI introduced ChatGPT for Teens for users who report being 13 to 17 or whom its systems estimate are under 18. The experience turns on stronger safeguards around self-harm, eating disorders, violence, sexual content, and emotional dependence, while adding responsible-homework reminders, Study Hours, break prompts, sensitive-image warnings, and parent controls. The launch consolidates earlier protections with several new learning and safety features.
Wispr raises $280 million and pushes beyond dictation
Voice-interface startup Wispr raised a $280 million Series B led by Menlo Ventures at a $2 billion valuation, bringing total funding to $361 million. Alongside expansion into meeting notes, the company introduced a speech model called Canto and said it can reduce error rates from about 30% to below 10%. TechCrunch reported the round and product plans; Axios separately listed the financing and investor group.
Independent researchers expose the blind spots in AI usage reports
A new AI Observatory assembled 24,521 consented conversations spanning 85,633 turns, seven public datasets, 5,000 users, and 52 models. MIT Technology Review reports that usage varied sharply by model and that applying Anthropic's work-focused filtering would have excluded 48% of the Observatory's sample, including disproportionate shares of health, relationship, sexual, harassment, and hate-related exchanges. The researchers caution that voluntary datasets still underrepresent sensitive use.
AWS adds Nvidia's Nemotron 3.5 Lightning to SageMaker JumpStart
The open 30-billion-parameter mixture-of-experts model activates 3 billion parameters per pass, supports a one-million-token context window, and targets high-volume agent tasks. AWS cites Nvidia tests claiming up to four times higher throughput and 30% faster task completion.
AWS gives OpenClaw agents bounded payment authority
An AWS and OpenClaw integration lets agents settle approved x402 payment requests through Bedrock AgentCore. Humans provision wallets, recipients, budgets, networks, and expiry limits; the model-facing runtime cannot create or expand payment sessions.
GitHub frames canvases as durable control surfaces for agents
GitHub described how Copilot canvases keep plans, state, approvals, and validation visible outside a scrolling chat. The examples span Java modernization and site production, with emphasis on steering and auditing multi-step agent work.
OpenAI and CodeAI launch a year-long AI literacy push
The partnership will create an expert advisory council, an Hour of AI program, a student Builders Challenge, support for a free high-school course, and career programming aimed at teaching students to question and shape AI systems.