Nvidia’s $105B Bet on OpenAI’s Ohio Data Center

FEATURED INSIGHT💡

Nvidia Moves From Chip Supplier to Financial Backer

Nvidia is putting its balance sheet behind OpenAI’s biggest infrastructure bet yet. The chipmaker agreed to guarantee up to $105 billion in lease and power payments for a planned Ohio data center developed by SoftBank-owned SB Energy. OpenAI will lease the site for 20 years, Nvidia will be its exclusive chip supplier, and the campus could reach 8 gigawatts of computing capacity. Nvidia is also investing $1.5 billion in SB Energy, according to Reuters.

This is more than a hardware sale. Nvidia is using its financial strength to help a major customer secure the land, power, and long-term financing required to buy enormous quantities of Nvidia systems. The arrangement could accelerate construction while giving Nvidia unusually deep influence over the infrastructure surrounding its chips. It also exposes the company to a different kind of risk: if OpenAI cannot meet its obligations, Nvidia’s guarantee could turn projected demand into a costly liability. That tension is why the deal has renewed concerns about circular financing when a supplier helps finance the customer buying its products.

The headline numbers are ceilings, not installed capacity. Reuters reports that the first 800 megawatts are expected in 2028, while the broader financing structure remains unfinished. The practical takeaway is that frontier AI is becoming an infrastructure-finance business as much as a model business. Future advantages may depend less on one benchmark win and more on who can secure power, capital, and upgrade rights for decades. Nvidia’s bet could strengthen its platform—or concentrate its fortunes even further around a handful of customers and projects.

Stop making AI decisions in the dark.

Leadership is asking: are we getting value from AI? Which tools are worth the spend? Where are we exposed? Right now, most teams have no idea.

Harmonic Security Usage Explorer changes that.

You get a complete picture of how your organization uses AI, automatically categorized into custom tasks and use cases.

You’ll see the projects being worked on, who’s using what tools, where AI investments are driving value, and where employees are engaging in risky behavior.

CIOs can rationalize spending and cut wasted licenses. CISOs can pinpoint where risk exists and neutralize it. AI committees can show exactly how their efforts are paying off.

See it in action

ON THE HORIZON 🌅

Claude Adds a Hidden Signature to AI Content

Claude’s next generation will carry a signature users cannot see. Anthropic says Claude models launched on or after August 2 will add machine-readable watermarks to generated text at the model level, while supported image and file outputs will carry signed C2PA provenance metadata. The policy applies worldwide across Claude products, APIs, and participating cloud platforms—not only in Europe—according to Anthropic’s guidance.

The immediate trigger is regulation. The European Commission began enforcing new AI Act transparency rules on August 2, including requirements for AI-generated or altered content to carry machine-readable marks. More than 180 organizations have signed a voluntary code intended to make those obligations practical, the Commission says. Anthropic’s implementation shows how a regional rule can become a global product default when providers operate one model stack across markets.

Watermarking could give publishers, educators, platforms, and compliance teams a new signal for tracing content after it leaves a chatbot. It will not settle authorship disputes by itself. Anthropic warns that very short passages, substantial edits, stripped metadata, or unsupported platforms can weaken detection, and a mark may show only that Claude processed existing material. Expect the next phase of AI governance to focus less on visible labels inside apps and more on whether provenance signals survive copying, editing, and distribution in the wild.

LATEST IMPORTANT NEWS 📰

Google puts sign-language translation into a phone keyboard

Google DeepMind’s new SL2T model turns American Sign Language into English text inside Gboard and Live Transcribe on Pixel 11. Trained on more than 100,000 hours spanning over 50 sign languages, it tracks simultaneous hand, body, and facial movement rather than treating signing as word-for-word English. The initial release lets users sign searches, messages, documents, and replies; more devices and languages are planned. Google also acknowledges occasional errors and published an impact report with Deaf-community advisers, making this a meaningful accessibility launch rather than only a lab demo.

Qwen opens the weights of a Max-class model

Alibaba’s Qwen team released Qwen3.8-2.4T-A95B, its first open-weight model in the Qwen-Max performance class. The mixture-of-experts system contains 2.4 trillion parameters but activates 95 billion per token, offers a native 262,144-token context window, and can be served with vLLM or SGLang. The catch is substantial: the repository is nearly 4.9 terabytes, the license is custom, and the open model is text-only with mandatory reasoning mode. This is an important open release, but not a casual local download.

Cursor moves from code editor to code host

Cursor launched Origin, an early-beta Git forge that places repositories, pull requests, and coding agents in one system. Paid users can host code directly or sync selected GitHub repositories while keeping GitHub as the source of truth; comments and pull-request updates move both ways. Cloud agents can clone, branch, commit, and open pull requests against Origin remotes. The bigger signal is strategic: AI coding companies are expanding beyond assistance inside the editor toward owning the collaboration layer where software changes are reviewed and merged.

See how AI delivers fast, secure, personalized experiences

Watch Fin and Plaid now on demand to see how AI can resolve issues directly in customer conversations. This includes everything from reducing bank-linking friction, enabling proactive support, and helping customers complete real financial tasks.

Watch now

FOR THE TECHNICALLY INCLINED 🛠️

Linguistic Puzzles Reveal What AI Benchmarks Miss

A new benchmark asks AI to solve linguistic puzzles where the rules must be discovered before they can be applied. The IOL-AI Challenge used unseen problems from the 2026 International Linguistics Olympiad, drew 731 submissions from 46 teams under a one-T4, 30-minute compute limit, and separately tested 15 unconstrained frontier and open models. Claude Opus 4.8 earned a jury score comparable to a human gold medal, yet the two resource-constrained systems submitted for jury grading landed in the bottom 5% of contestants. Size was not destiny: some 14-billion-parameter entries beat models twice as large, with gains coming from decoding and output handling. The study also found that automatic scoring preserved the jury’s ranking but compressed differences, inflating weak systems by about 13 points and understating strong ones. It is a sharp reminder that benchmark design and grading can hide as much as they reveal.

AI TOOL OF THE DAY 🚀

Wispr Flow turns spoken thoughts into polished text in any app on Mac, Windows, iPhone, or Android, automatically removing filler words and formatting the result, with a free tier of 2,000 words per week.

No follow-up questions required

Every sales leader knows the feeling. You walk into a pipeline review with a number you believe in, and twenty minutes later, you're defending every line item to a CEO who just wants to know what's actually going to close.

HubSpot Sales Hub ends that conversation. Every deal, every rep's activity, and every buyer signal are all in one place and updated automatically. So your forecast is built on what's actually happening. And when you present that number, you can stand behind it.

Get a demo