• AI Valley
  • Posts
  • OpenAI is becoming too big to fail

OpenAI is becoming too big to fail

PLUS: New report reveals AI still struggles with real work

Together with

Howdy, it’s Barsee.

Happy Tuesday, AI family, and welcome to another AI Valley edition. This issue takes just 5 minutes to read.

Today’s climb through the Valley reveals:

  • OpenAI is trying to make itself too big to fail

  • New report reveals AI still struggles with real work

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

ROCKET.NEW

Image Credit: Rocket.new

Rocket, the $15M-funded startup behind Vibe Solutioning is redefining how teams build software. Turn your ideas into production-ready apps using intuitive / and @ Commands, shortcuts that let you execute complex actions precisely and efficiently. No need to craft perfect prompts or remember workflows; Commands make discovery effortless and execution error-free.

With Context, Rocket understands your goals across every stage, Day 0 (solutioning), Day 1 (getting started), and Day 2 (iterations and deployments). Build full-stack apps from prompts or Figma designs, integrate with GitHub, Stripe, or OpenAI, and ship faster than ever.

*This is sponsored

THROUGH THE VALLEY

Image Credit: Morning Brew

OpenAI has struck a massive $38 billion multi-year deal with Amazon Web Services, one of the biggest cloud partnerships in AI history. The agreement gives OpenAI access to AWS’s new EC2 UltraServers powered by NVIDIA GB200 and GB300 chips, with room to scale to millions of CPUs and GPUs over the next seven years.

The deal cements AWS as one of OpenAI’s main compute partners, alongside its other large-scale semiconductor and cloud agreements. OpenAI will get dedicated compute clusters to train its next-generation models, power trillion-parameter inference, and support products like ChatGPT, Sora, and upcoming multimodal systems.

OpenAI’s global infrastructure play now looks like this:

  1. $500 billion – Stargate deal

  2. $100 billion – NVIDIA

  3. $100 billion – AMD

  4. $38 billion – Amazon Web Services

  5. $25 billion – Intel

  6. $20 billion – TSMC

  7. $13 billion – Microsoft

  8. $10 billion – Oracle

  9. Multi-billion – Broadcom

  10. Launched Atlas, its new browser to compete with Chrome

  11. Became the world’s most valuable private company

  12. Considering a $1 trillion IPO by 2027

Why does it matter?

If OpenAI succeeds, it could spark a new industrial revolution. If it fails, the damage could reach far beyond Silicon Valley. The company’s supporters call it the next Apple, Google, and Tesla combined. Its skeptics call it the next bubble.

Just a few years ago, numbers like these would have seemed impossible. OpenAI has grown from a research lab into what insiders call a “nation-state of AI infrastructure.”

Its goals are clear:

  • Control the world’s supply of advanced compute so rivals can’t easily access training power.

  • Monetize its massive surplus by reselling compute to other companies that build on top of its models.

By partnering directly with every major chipmaker and cloud provider, OpenAI is turning itself into the central hub of the AI industry, the company that connects the world’s most powerful chips, data centers, and models into one unified network for frontier AI.

Remote Labor Index from arxiv

A new benchmark called the Remote Labor Index (created by Scale AI and the Center for AI Safety) tested top AI models on real freelance projects. The result: even the best systems completed less than 3% of tasks at professional human quality.

Researchers collected 240 verified Upwork projects across 23 job categories, including writing, design, marketing, and data analysis. Six AI models attempted the same tasks, and their results were compared to the original freelancers’ work.

Manus led with a 2.5% success rate, while Grok 4 and Claude Sonnet 4.5 followed at around 2.1%. That means roughly 97% of outputs failed to meet even basic client expectations.

Most AI work fell apart because of incomplete results, formatting issues, or low quality, performing well only in narrow areas like logo design, audio editing, or chart creation.

Why does it matter?

The findings reveal how far AI still is from handling complex, real-world jobs. Models can summarize or reason well in isolation, but they struggle with multi-step projects that need structure, feedback, and judgment.

For now, AI agents work best with humans in the loop, automating repetitive tasks while people handle context and decision-making. As one researcher summed it up: “AI is great at doing parts of a job, but not the job itself.”

The real progress will come from workflows where humans and AI collaborate, not compete.

TRENDING TOOLS

  • Averi > Where complete marketing workflows happen. AI-powered strategy and creation, expert collaboration, and institutional memory… all in one workspace *

  • Komos > Turn a 5-minute screen demo into a repeatable workflow

  • Liminary > AI superpowered memory that surfaces your saved knowledge in context with the work you're doing

  • Perplexity Flight Tracker > Track commercial flights and live status changes directly inside Perplexity

  • Usage4Claude > A macOS menu bar app that shows how much you actually depend on Anthropic’s Claude

  • FormAI > Your personal AI gym partner that gives real-time form corrections and exercise recommendations

(*) signifies sponsored tool

THINK PIECES / BRAIN BOOST

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.