AI Industry News Brief — August 19, 2026

Coverage window: the last 24 hours in absolute time — 2026-08-17 22:13 UTC to 2026-08-18 22:13 UTC.

Nvidia Guarantees Up to $105 Billion for an Ohio Data Center With OpenAI as Anchor Tenant

Nvidia disclosed an “aggregate payment obligation” capped at $105 billion tied to an AI data center being built by SB Energy on the PORTS-Pike Technology Campus in Pike County, Ohio. The figure lands $145 billion below what had earlier been reported, which has fed concerns that chip demand is being propped up by circular financing rather than end-user pull. The deal includes a $1.5 billion Nvidia investment in SB Energy, a SoftBank affiliate, with OpenAI signing on as anchor tenant under a 20-year lease. It follows Monday’s announcement that Nvidia is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing platforms meant to mobilize more than $500 billion for AI infrastructure — evidence, analysts argue, that Nvidia’s durable advantage is migrating from silicon to capital formation.

Reference: Fortune · CNBC

OpenAI Launches ChatGPT for Teens With Age Prediction, Quiet Hours and Study Mode

OpenAI is rolling out ChatGPT for Teens, a separate experience for users aged 13 to 17. Anyone who identifies as a teenager — and anyone OpenAI’s age-prediction system estimates to be under 18 — is placed into it automatically. The teen environment applies tighter limits on sexual or romantic roleplay, graphic violence and self-harm content, and adds safeguards intended to discourage emotional dependency on the chatbot. On the education side, the product steers students toward Study Mode rather than simply completing assignments, generating quizzes, homework reminders and learning visualizations. Parents who link accounts can set “Quiet Hours” during which the chatbot is entirely unavailable, and receive notifications in certain high-risk situations.

Reference: The Hill · Tech Startups

Meta’s Trial Against 29 States Opens, Putting Algorithmic Design Itself on the Stand

Opening arguments began in federal court in Oakland, California, in a case brought by a bipartisan coalition of 29 states alleging that Facebook and Instagram were designed in ways that harmed children and teenagers. Colorado, California, New Jersey and Kentucky are leading the case, which claims Meta built addictive product features, misrepresented platform safety and improperly collected children’s data. The states are seeking more than financial penalties: they want operational changes including age restrictions and the removal or modification of features such as infinite scroll. Mark Zuckerberg and Instagram head Adam Mosseri are expected to testify. A ruling that forces nationwide product changes would ripple through TikTok, YouTube, Snap and every other service built around engagement-optimized feeds.

Reference: Tech Startups (via Reuters)

Google Buys Bankrupt Spirit Airlines’ Internal Data for $10 Million to Train AI

Google won a bankruptcy auction for the entirety of Spirit Airlines’ internal business data, paying $10 million. The package includes roughly 100 million employee emails, 500 million Microsoft Teams chats, plus spreadsheets, calendars, marketing materials, HR documents, financial databases and audits. Customer data and personally identifiable information are excluded, and a third party will scrub the set before transfer. Google outbid AI data firm Mercor’s $7.5 million offer, and a U.S. bankruptcy judge is scheduled to sign off this week. Corporate operational records at this scale capture real decision-making, coordination and problem-solving that public web scrapes cannot — and the sale marks bankruptcy auctions as a new supply channel for workplace-agent training data.

Reference: CNN Business · Business Insider

Inference Chip Startup Etched Hits a $21 Billion Valuation While Poaching Nvidia Engineers

Etched, the semiconductor startup founded by three Harvard dropouts, has reached a reported valuation of roughly $21 billion. It has raised nearly $2 billion, shipped its first chips and signed quantitative trading firm Jane Street as an early customer, according to The Wall Street Journal. Etched is aimed squarely at inference — the stage where an already-trained model must answer prompts at enormous scale and low latency — rather than training. As AI economics shift from one-off training runs toward continuous production usage, specialized architectures can plausibly beat general-purpose accelerators on cost or speed for specific workloads. The company is actively recruiting engineers from Nvidia and other established chipmakers.

Reference: Tech Startups (via The Wall Street Journal)

ByteDance Signs Hollywood’s First AI Copyright Pact, Covering Seedance and Seedream

ByteDance and the Motion Picture Association signed a memorandum of understanding to protect film and TV intellectual property across ByteDance’s generative AI products — the first agreement the MPA has struck with an AI company. It covers the Seedance video model and Seedream image generator, along with every ByteDance AI model used in TikTok, CapCut and Dreamina. The deal follows a February cease-and-desist letter sent after Seedance 2.0’s release prompted a wave of clips riffing on actors including Brad Pitt and Tom Cruise. Crucially, the pact formalizes output-layer protections only — content filters, face blocking and C2PA Content Credentials — and explicitly does not resolve whether training on copyrighted material was itself infringement. No payments flow to studios; it is a truce, not a licensing deal.

Reference: Variety · NBC News

Velaura AI Raises a $110 Million Series A on the Bet That AI’s Next Bottleneck Is Electricity

Silicon Valley AI infrastructure startup Velaura AI raised $110 million in Series A funding at a valuation above $1 billion. Seligman Ventures led, with Capricorn Investment Group and Prosperity7 Ventures joining as new investors and Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund and StepStone Group participating. The company’s thesis is that the industry’s binding constraint is shifting from access to compute toward access to enough power to run it. Its Titan Core digital chip IP and design platform claims a 2x to 4x improvement in performance per watt on the mathematical operations AI accelerators depend on, without sacrificing throughput — and the underlying technology has already shipped across more than 30 million ASICs. Velaura says it is working with leading hyperscalers to fold the technology into future XPU roadmaps, and is targeting physical AI (robots, drones, autonomous machines) alongside the data center.

Reference: Tech Startups

Baidu’s Q2: AI Cloud Growth Still Cannot Cover the Hole Left by Advertising

Baidu reported second-quarter revenue that missed expectations as weakness in its traditional advertising business outweighed gains from AI cloud and newer AI products. Revenue from AI applications reached RMB 2.5 billion in the quarter, up 3% year over year. The results capture a structural bind facing older internet companies: generative AI creates new revenue lines while simultaneously eroding the search-advertising business that funded their rise, since assistants increasingly answer questions directly instead of routing users to ad-supported results. Baidu has responded by pushing its Ernie ecosystem, AI cloud services, enterprise tooling and robotaxis — bets that require heavy capital and must compound fast enough to outrun the decline in the legacy model.

Reference: Tech Startups (via Baidu)

Today’s Takeaway

The center of gravity in AI infrastructure has moved from “how many chips can you buy” to “who can raise the capital and secure the power.” Nvidia effectively recast itself as a lender with a $105 billion guarantee, while Etched and Velaura were awarded billion-dollar valuations for owning narrow slices — inference efficiency and performance per watt.

At the same time, regulation and copyright started arriving as line items. Meta is in court with 29 states over algorithmic design itself, OpenAI moved preemptively with a teen-specific product, and ByteDance cut Hollywood’s first AI copyright pact. The race for training data, meanwhile, has descended all the way to the bankruptcy auction floor.