Pondero AI logo

Pondero AI

Archives
Log in
Subscribe
August 18, 2026

Pondero Brief: Anthropic's run rate hits $65B and passes OpenAI

Pondero Brief - AUGUST 18TH, 2026

7x in seven months resets the API-vendor bet before either lab files to go public.
pondero. BRIEF · AUG 18, 2026

Anthropic's run rate hits $65B and passes OpenAI

Today: Nvidia locks OpenAI's Ohio campus, OpenAI drops its safety team, and Amazon confirms shredding rare books for AI data.

Anthropic hit a $65B annualized revenue run rate at the end of July, up from $9B at the close of 2025, per TechCrunch citing Bloomberg and the Financial Times. That is 7x in seven months, past a May checkpoint of $47B, and ahead of OpenAI's roughly $40B.

Also in today's brief

  • Higgsfield's $400M round quadruples its valuation
  • Nvidia finances and locks supply at OpenAI's Ohio campus
  • OpenAI disbanded its catastrophic-risk safety team
  • Amazon tracked shredding rare books for AI training
  • ChatGPT Mac logs activity - and flags its own risks
 
Money & Moves
Higgsfield Series B announcement illustration

Higgsfield raised $400M at a $5.4B valuation

DST Global led the Series B, quadrupling the January mark in eight months on $700M ARR and 390 Fortune 500 customers, per TechCrunch. Co-founder Alex Mashrabov, ex-Snap, put the cost plainly: "Just one minute of video is like processing 60,000 words." Why it matters: AI video is now an enterprise line item with real ARR behind it, not a demo reel, and compute is the moat. Our take →

 
Nvidia and SB Energy PORTS-Pike campus illustration

Nvidia backstopped $105B for OpenAI's Ohio campus

Nvidia guaranteed construction financing for the PORTS-Pike campus in Pike County, took a $1.5B equity stake in builder SB Energy (disclosed in an Aug 17 filing), and locked exclusive rights to supply every chip on site. OpenAI signs a 20-year lease on the grounds of a former uranium enrichment plant. Why it matters: Nvidia is now financier, shareholder, and sole supplier for one buildout, tying its GPU sales directly to OpenAI's infrastructure. Our take →

 
Policy & Legal
OpenAI Preparedness team dissolution illustration

OpenAI dissolved its Preparedness team

The catastrophic-risk group was disbanded at the end of July and its work scattered across product units; ethics lead Chloe Bakalar and safety head Johannes Heidecke both departed, per Engadget citing the FT. OpenAI framed it as streamlining ahead of its IPO. Weeks later, preview models autonomously reached Hugging Face during red-teaming. Why it matters: no single team now owns the clearance decision for frontier releases. Our take →

 
Amazon rare books AI training facility illustration

Amazon is destroying rare books to train AI

A 404 Media reporter hid a GPS tracker in a rare book; it ended at Amazon's VGT3 facility in Las Vegas, where workers cut spines off books, flatbed-scan the pages, and discard the copies. Amazon confirmed it "purchases books through commercial channels to improve products and services." Labs want out-of-print titles because they predate AI-contaminated web text. Why it matters: the first documented trace of a physical scanning pipeline authors suspected but could not prove. Our take →

 
Tools & How-To
ChatGPT Computer History Mac activity tracker illustration

ChatGPT shipped a Mac activity tracker, then flagged its own risks

The opt-in Computer History tool logs clicks, typing, and app switching (not screenshots or audio) for Pro, Business, and Enterprise on macOS, per TechRepublic. In its launch-day docs OpenAI disclosed two problems: memory files are stored as unencrypted plain text any same-user process can read, and tracked content is a prompt-injection vector. Why it matters: enable it for staff and you get an activity log plus a new local attack surface in one switch. Our take →

 
Quick Hits
• Automate this week's agentic workflows. Visual builder, generous free tier: Make connects models to the apps you already run. Try Make →
From the Pondero Stack
Proton Lumo AI assistant review illustration

Proton Lumo is the pick when zero-access encryption is non-negotiable

4.0 out of 5. Buy it for HIPAA, GDPR, or confidentiality-bound work where the provider must not be able to read your prompts; for general use, ChatGPT or Claude Pro are the stronger picks. Read the review →

 
Prompt injection defense production guide illustration

Six prompt-injection defenses, rated against three real 2026 attacks

We scored six production defenses against GitLost, the Claude Code GitHub Action CVE at CVSS 7.8, and the Hugging Face agent breach. One held on its own; five needed a second layer. If you run agents in production, this is the layering order. Read the guide →

 

How was today's brief?

★★★★★ Nailed it  |  ★★★ Solid  |  ★ Missed

Jonathan Hildebrandt Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.

Affiliate disclosure  ·  Unsubscribe  ·  Manage preferences

X  ·  LinkedIn  ·  Bluesky

Pondero earns commissions on some links. This does not affect our editorial picks.

Don't miss what's next. Subscribe to Pondero AI:
← Newer Pondero Brief: OpenAI paused Astra training. Anthropic raised its misalignment alarm. Older → Pondero Brief: Anthropic clears $11.5B and turns a profit before its IPO
pondero.ai
Bluesky
LinkedIn
Twitter
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.