- AI Fire
- Posts
- 💰 OpenAI’s $850B Public Play
In partnership with
The AI boom is starting to look like national infrastructure. OpenAI and the Trump administration are reportedly discussing a public stake before a possible IPO.
What's on FIRE 🔥
IN PARTNERSHIP WITH PLANABLE
BreakingSilos 2026 is a Free online conference on June 9th. No vendor pitches. No generic panels. Just people who've built AI visibility ecosystems.
There's no shortage of AI "hacks" and generic webinars promising to fix your visibility.
BreakingSilos 2026 brings together practitioners who've actually built visibility ecosystems where SEO, social, PR, content, and brand reinforce each other. And AI is citing them for it. Find out exactly how.
AI INSIGHTS
OpenAI and the Trump administration are reportedly discussing a possible government stake in the AI giant.
The idea is simple but big: OpenAI could donate equity to help seed a Public Wealth Fund, giving regular Americans a way to share in the upside of the AI economy.
According to CNBC, Sam Altman first raised the idea with the Trump administration in 2025. The talks continued this week as he met lawmakers and officials in Washington, D.C.
The timing matters. OpenAI is now valued at more than $850 billion, is preparing for a possible IPO, and recently closed a record funding round backed by MGX, a fund linked to Abu Dhabi’s sovereign wealth fund.
This also fits a larger pattern. The Trump administration has already taken stakes in strategic companies tied to Intel, IBM, quantum computing, and critical minerals.
There are no final terms yet. But the bigger signal is clear: Washington now sees AI as a national asset, not just a private tech race. If AI becomes the next trillion-dollar industry, the government wants the public to own at least a small piece of the upside.
PRESENTED BY HUBSPOT
2026 State of AEO Report
A year ago, most marketers weren't thinking about AI search. Now it's one of the fastest moving channels in the industry and nobody has a playbook yet.
So we built one. We surveyed hundreds of marketers to find out how they're approaching answer engine optimization, where they're investing, what's actually working, and what isn't.
The result is the 2026 State of AEO Report. Real data. Real strategies. A clear picture of where AI search is headed and how to get ahead of it.
AI SOURCES FROM AI FIRE
1. Step-by-Step Tutorial: Hermes AI Agent Got a NEW Research Brain with NotebookLM. Build an AI research workflow that can collect knowledge, reuse notebooks, turn research into scripts, emails, reminders, and daily briefings.
2. Claude Code Prompt Caching Guide: Save Tokens And Speed Up Long Sessions. Learn how Cache Write, Cache Read, /clear, session handoffs, and /usage help Claude Code reuse context, cut wasted tokens, and keep big projects running smoother.
3. [AI Personal Brand Playbook] Define Your Unique Edge with Claude. Use Claude to clarify your audience, core values, client personas, and brand direction so every post, proposal, and visual feels more focused, consistent, and aligned with your brand.
FIRE RECAP: BIGGEST AI NEWS THIS WEEK
👨💻 Anthropic says Claude now writes over 80% of the code merged into its own codebase. Even crazier: the typical engineer is shipping around 8x more code than before.
📱 OpenAI is reportedly planning ChatGPT’s biggest redesign yet, turning it into a superapp with Codex, AI agents, image tools, and partner services like Canva and Booking.com.
🛡️ Anthropic is expanding Claude Mythos Preview access from about 50 to 200 Glasswing partners. These partners already found 10,000+ serious security flaws.
🚀 OpenAI says Codex now has 5M weekly users, and non-developers are growing fast. New plugins help analysts, marketers, sales teams, designers, and investors use Codex too.
🧠 OpenAI launched Dreaming, a better memory system for ChatGPT. It helps ChatGPT remember fresher context, follow your preferences, and avoid using old details that no longer apply.
TODAY IN AI
AI HIGHLIGHTS
🇺🇸 The U.S. and Japan just announced a $1B AI research partnership under the Genesis Mission. Japan is now the first country joining America’s push to double science output with AI.
🛡️ Anthropic is reportedly helping the NSA deploy Claude Mythos for offensive cyber work. The model can detect and exploit software flaws, which makes this a huge national security story.
🤖 Cloudflare co-founder Matthew Prince says bots have now passed human traffic online for the first time. He expected it later, but agentic traffic is growing faster than predicted.
🛒 Gopuff launched Go, an AI shopping assistant built with SpaceXAI. It uses Grok, real-time signals from X, and order data to predict what users may add to their carts.
🔒 OpenAI is rolling out Lockdown Mode in ChatGPT. It blocks risky web actions, disables Deep Research and Agent Mode, and helps reduce prompt injection data leaks.
💰 AI Enterprise Deals: Snowflake and Anthropic deepened their enterprise AI partnership as demand for Claude inside Snowflake Cortex AI grows. Customers including Block, Carvana, Indeed, Notion, and eSentire are using it to run trusted AI agents on governed business data.
NEW EMPOWERED AI TOOLS
🎙️ Wave turns your voice into text anywhere on macOS, with local or cloud AI transcription that works across all apps.
🛒 Manus Shopify Connector lets paid Manus users build Shopify stores, manage catalogs, and create campaigns from one chat.
🎨 MAI-Image-2.5 generates and edits images with precise scene control, identity preservation, localized edits, and text rendering.
🌙 Dreambeans by Google Labs creates daily AI stories from your Gmail, Calendar, Photos, YouTube, and Search for AI Ultra users.
AI BREAKTHROUGH
Datacurve just dropped DeepSWE, a new long-horizon software engineering test designed to show where leading AI models truly separate. The results are a massive reality check.
On this new benchmark, GPT-5.5 hits just 70%. GPT-5.4 comes in at 56%, and Claude Opus 4.7 sits at 54%. Older benchmarks made these models look nearly identical, but DeepSWE rips the band-aid off to reveal a massive performance gap.
Why are the scores suddenly so much lower? It comes down to 3 major upgrades in how they test these models:
No more memorized answers: DeepSWE uses completely original tasks. This guarantees models aren't just regurgitating code they saw during their training runs.
Zero hand-holding: On DeepSWE, the prompts are actually half the length of older tests like SWE-bench Pro. Instead of telling the AI exactly which files to edit, the agent has to search the repo itself, understand the architecture, modify multiple files, and avoid breaking existing features.
Behavior-based grading: DeepSWE actually checks if the requested feature works, letting the model solve the problem in any valid way.
DeepSWE forces models to navigate ambiguity, figure out the repository's context, and write comprehensive, multi-file solutions.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply