There's one number Elon Musk put on a slide on Saturday night that most people missed.
2%.
That's how much AI compute every semiconductor company on Earth, like TSMC, Samsung, Intel, all of them together, actually produces relative to what Tesla, SpaceX, and xAI say they'll need.
His conclusion was short: "We either build the Terafab, or we don't have the chips."
So he's building the Terafab.
It's a joint venture between Tesla, SpaceX, and xAI, located on Tesla's Austin campus, and it costs $20–25 billion. The target is 1 terawatt of AI compute per year, i.e., roughly 100 billion chips at 2-nanometer, the most advanced process on the planet. For context, Terafab at full scale would produce 70% of TSMC's current global output. Except it's one building that didn't exist a week ago.
Why did this happen?
Because compute is the new oil, and Musk got tired of waiting at someone else's pump.
Every AI company in the world is fighting over the same finite supply of NVIDIA chips. You can't win that race by buying faster. You win it by owning the factory. Amazon did it with Trainium, Google did it with TPUs, and Apple did it with their Neural Engine. Musk watched all of them build their independence from NVIDIA, watched his own AI roadmap get bottlenecked by supply, and made the move.
Well, here’s a note on credibility: Musk has never manufactured semiconductors. His timelines has a history - FSD in 2020, the 4680 cell, the Hyperloop. The skepticism is fair.
But the logic isn't wrong. And the $20+ billion isn't a press release. Tesla's CFO confirmed it sits atop their existing $20 billion capex plan for the year.
Whether Terafab ships on time or five years late, what it tells you right now is that the AI race is no longer just about models. It's about who controls the machine that builds the machines.
The CEO of Salesforce stopped hiring engineers. Then made $46 billion.
Marc Benioff hired zero new engineers in FY2026. AI coding agents gave his engineering team 30% more productivity. He's now hiring 2,000 salespeople instead. Because the work that grew was selling AI to everyone else.
His message to every CEO: "We are the last generation to manage only humans."
Zuckerberg is building an AI version of himself to run Meta.
Meta has already cut down to one chief of staff company-wide. He's also telling all 78,000 employees to build their own personal AI agent using an internal tool called Second Brain, powered by Claude.
The CEO of the most valuable social company in the world is testing his own replacement in real time.
Agile Robots Partners with Google DeepMind to Power Next-Gen Industrial Robots.
The collaboration will use real-world robot data to continuously improve Gemini, enhancing autonomy and intelligence. With over 20,000 robots already deployed, Agile Robots aims to scale intelligent automation and accelerate fully autonomous production systems.
The long-term deal signals a major push toward AI-powered industrial transformation, though financial details remain undisclosed.
FAST BREAK
Salesforce's AI agents now resolve 85% of customer service conversations without human involvement.
Customer service is the single largest workforce category in the world. Millions of people across every country and every industry spend their careers answering questions that have already been answered a thousand times.
Benioff isn't predicting this future. He's running it, at scale, inside one of the biggest software companies alive.
The question isn't whether AI will replace support jobs, because it already has. The question is which job is next.