Pondero AI logo

Pondero AI

Archives
Log in
Subscribe
June 16, 2026

Pondero Brief: 2026-06-16: OpenAI burned $34B in 2025 for a $38.5B net loss

Pondero Brief - JUNE 16TH, 2026

Audited, FT-verified, now in 42 AGs' hands. Plus the Brave lever for Claude citations.
pondero. BRIEF

OpenAI's audited 2025 financials are out. The numbers are worse than the estimates.

Audited financials published by Ed Zitron, independently confirmed by the FT. The 42 state AGs who opened their probe now have numbers. A $1T+ S-1 is expected late July.

JUNE 16TH, 2026 · BY JONATHAN HILDEBRANDT

Ed Zitron published audited OpenAI financials on June 15, with the Financial Times independently confirming the figures. OpenAI booked $13.07B in revenue against $34B in total costs, a $20.92B operating loss. Net loss landed at $38.53B, inflated by a $41.55B fair-value charge from the nonprofit-to-for-profit conversion. R&D alone ($19.18B) outran the entire year’s revenue. Microsoft collected $17.2B across cost-of-revenue and R&D, per Zitron’s analysis.

Why it matters. The 42 state AGs who opened their probe earlier this month now have audited numbers to work from, and a $1T-plus S-1 is reportedly due late July or August. The pitch and the books are about to meet in public.

See all our coverage →
 

Anthropic’s DC talks stalled with no deal on the Fable 5 shutdown.

JUNE 16TH · PONDERO NEWSDESK

Co-founder Tom Brown and policy chief Sarah Heck led June 15 sessions with Trump officials, who sources said were “speaking different languages.” Fable 5 hit day five of suspension with no path back, and CIA plus science-adviser meetings are next. Read our take.

 

The Anthropic export controls were planned weeks before June 12, per the Washington Post.

JUNE 16TH · PONDERO NEWSDESK

A four-byline WaPo investigation traced the takedown to a foreign firm’s request to access Mythos, not a jailbreak. Anthropic got 90 minutes to comply. This was a slow trust collapse, not an emergency, which makes a quiet reversal far less likely. Read the investigation.

 

Meta’s CTO called the Applied AI reorg “atrocious” in an internal memo.

JUNE 16TH · PONDERO NEWSDESK

Andrew Bosworth used that word in a June 15 memo (first reported by Wired) after the reorg moved ~6,500 engineers into Applied AI in May. He pledged ~20-report manager caps, an opt-out for drafted engineers, and a July hackathon. The retention bill on a forced AI sprint is coming due. Read the memo report.

 

GitHub leaned on AWS to absorb nine May outages driven by AI coding agents.

JUNE 16TH · PONDERO NEWSDESK

Microsoft quietly added AWS capacity after Copilot Workspace and third-party agents drove nine outages in May, per Business Insider. If you ship on agentic CI, the platform layer is now a real reliability variable, not a given. Read the report.

 
Claude vs ChatGPT source overlap: 8%. Brave rank is the lever

Brave rank is the cleanest lever for getting Claude to cite you.

JUNE 14TH · PONDERO GUIDE

Brave rank is the cleanest lever we found for getting Claude to cite you. Per Profound data we analyzed: Claude/ChatGPT source overlap is 8%, and Claude pulls from Brave’s top 10 at an 86.7% overlap rate. Here is the four-fix playbook. Read it · try Firecrawl.

 
CustomGPT vs Chatbase vs Zendesk AI comparison

CustomGPT vs Chatbase vs Zendesk AI for support.

JUNE 14TH · PONDERO COMPARISON

If your best answers live in closed tickets, CustomGPT wins. If you want zero new vendors, Zendesk native is the path. We name the pick per use case. Read it · try CustomGPT · try Chatbase.

 
Kit MCP setup guide. 76 endpoints, Creator plan

Kit MCP left beta June 11 with 76 endpoints.

JUNE 13TH · PONDERO GUIDE

Wire Claude or ChatGPT to your email list in five steps. The Creator plan ($33/mo, per Kit pricing) is the minimum that unlocks MCP, so factor that before you start. Read it · try Kit · try beehiiv.

 
GitHub Copilot model picker and agentic security setup guide

Claude Fable 5 is in Copilot’s model picker, with auto security validation.

JUNE 12TH · PONDERO GUIDE

Third-party agent code now gets automatic security checks, and Fable 5 is selectable. Here is how to configure the two-layer setup. Read it.

 
GitHub Copilot AI Credits pricing guide. Which plan covers your usage

GitHub Copilot moved to AI Credits on June 1.

JUNE 11TH · PONDERO GUIDE

Tab-complete stays unlimited; credits only burn on chat and agent sessions. We map which plan actually covers your usage so you do not overpay. Read it.

 

Firecrawl

JUNE 16TH · TOOL TO TRY

LLM-ready scraping that turns any site into clean markdown for RAG and citation pipelines — the same plumbing behind our Brave Search guide above. Try Firecrawl.

 

beehiiv

JUNE 16TH · TOOL TO TRY

Newsletter platform built for growth, with referral programs and ad network baked in. Worth a look if you are wiring an email list to Claude or ChatGPT via Kit MCP. Try beehiiv.

 

CustomGPT

JUNE 16TH · TOOL TO TRY

No-code GPT trained on your own docs and closed tickets — the support-bot path we compare against Zendesk AI in this week’s slate. Try CustomGPT.

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

Affiliate disclosure  ·  Unsubscribe  ·  Manage preferences

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

Don't miss what's next. Subscribe to Pondero AI:
pondero.ai
Bluesky
LinkedIn
Twitter
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.