|
Tuesday, August 18, 2026
|
|
|
|
| |
|
Today’s top story
INDUSTRY
Nvidia guarantees $105 billion in financing for OpenAI Ohio data center
TechCrunch
A securities filing disclosed on August 17 shows Nvidia will guarantee up to $105 billion in credit for the Ports-Pike data center near Cincinnati, where OpenAI is anchor tenant and Nvidia is the sole compute supplier. Nvidia is also putting $1.5 billion of equity into SB Energy, the SoftBank-backed developer building the site, which already counts OpenAI as an investor. The campus is designed to scale from 4.25 gigawatts to 8 gigawatts, powered by a $33 billion natural gas plant on former Department of Energy land.
|
|
Why it matters
Nvidia is now underwriting the debt that buys its own chips, a circular financing structure that makes its balance sheet the load-bearing wall under OpenAI's build-out. It also narrows the earlier $250 billion backstop talks into a concrete, filed commitment tied to one site.
|
|
|
More headlines
|
|
FUNDING
Groq raises $350 million at a cut valuation for neocloud pivot
Groq closed a $350 million round led by Disruptive at a $3.5 billion valuation, roughly half its $6.9 billion mark from September 2025, with Nvidia expected to participate. Having abandoned its own LPU silicon after Nvidia hired founder Jonathan Ross in a $20 billion licensing deal, Groq now runs 13 Nvidia-powered inference data centers and plans to grow from 54 megawatts to over 200 megawatts by 2027.
TechCrunch
|
|
POLICY
Justice Department probes a16z over board seats at rival AI firms
A civil antitrust investigation nearly a year old is examining whether Andreessen Horowitz partners violated Section 8 of the Clayton Act by sitting on the boards of competing data companies, specifically Ben Horowitz at Databricks and Martin Casado at Fivetran. Enforcers have historically applied interlocking-directorate rules to private equity, making this the first significant push against a venture firm.
Fortune
|
|
POLICY
OpenAI disbands preparedness team ahead of expected IPO
OpenAI dissolved the group that assessed whether frontier models could enable cyberattacks, biological and chemical threats, mass persuasion or autonomous replication, folding its work into specialized cybersecurity and biosecurity units in July. It is the third safety-focused team the company has wound down in about two years, and OpenAI described the move as streamlining before a planned listing.
iTechPost
|
|
FUNDING
Wispr raises $280 million at $2 billion valuation, previews Canto model
Menlo Ventures led a $280 million Series B for the voice-to-text startup, bringing total funding to $361 million and quadrupling its valuation, with Acrew, Forerunner, Goodwater and Peak XV joining. Wispr also previewed Canto, its first proprietary speech model, which it says cuts word error rates from over 30% to 5-10% in noisy conditions and heavy accents.
TechCrunch
|
|
INDUSTRY
Amazon is cutting up rare books to scan them for AI training
Investigators traced rare and out-of-print titles bought at auction to an Amazon facility in Las Vegas, where staff reportedly slice off bindings to feed pages through high-speed scanners. Amazon confirmed it acquires books through commercial channels but would not say how many it has destroyed, echoing a destructive-scanning practice Anthropic previously used.
TechCrunch
|
|
This week in benchmarks
|
|
vs. Tuesday, August 11, 2026
|
|
Coding
Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) fell from #7 to #9
|
|
Intelligence
Kimi K3 (max) fell from #6 to #7
|
|
Intelligence
GPT-5.6 Sol (xhigh) fell from #7 to #8
|
|
Intelligence
Claude Opus 5 (Adaptive Reasoning, Medium Effort) fell from #8 to #9
|
|
Intelligence
Qwen3.8 Max fell from #9 to #10
|
|
Coding
GPT-5.6 Terra (max) fell from #6 to #7
|
|
Read past briefs and the live benchmarks at
ai-newsletter-eta-lemon.vercel.app.
Forward this to a friend who’d appreciate the brief.
|
