✨ The Age of AI is starting to bloom

Why AGI isn’t imminent, why an AI crash isn’t either, and why the economy is already showing signs of genuine AI-driven lift

My fellow pro-growth/progress/abundance Up Wingers in America and around the world:

Two of the questions I’m most frequently asked by reporters or podcast hosts: “What is your AGI or superintelligence timeline?” and “Are we in an AI bubble?”

Which is fine. When I’m the interviewer, I often ask the same things. They’re important questions, after all — with big implications. Let me briefly offer what are probably contrarian answers:

First, “superintelligence super soon” isn’t my baseline. That’s an opinion based partly on a) my natural caution from having experienced more than a few tech hype cycles, b) my conversations with tech folks, and c) predictions markets. On that final point:

• Kalshi: “When will OpenAI achieve AGI?” — 41 percent (before 2030)

• Metaculus: “Date of ‘weakly’ general AI” — Nov. 20, 2027

• Metaculus: “Date of ‘general’ AI” - July 2033

• Manifold Markets: “When will AGI arrive?” - 2034

• Manifold Markets: “By 2028, will AI produce a clearly noticeable shift in U.S. GDP, GDP per capita, unemployment, or productivity trends?” - 30 percent

All that said, if in 10 years AI has evolved to something we can generally agree is AGI or beyond, that hardly seems like a distant date to me.

Bursting bubble speculation

Second, we’re not in an AI bubble as of right now. Annual AI investment has risen by roughly $200–300 billion since 2023. Which is a lot, definitely. Yet Goldman Sachs in estimates that generative AI could create about $20 trillion in present-discounted economic value for the US, including around $8 trillion in capital income for firms. Their earlier work implies roughly a 15 percent eventual boost in US labor productivity from generative AI—equivalent to roughly $4–5 trillion at today’s output—and micro-studies often find task-level productivity gains on the order of 25–30 percent. This is hardly the stuff of crazy-got-nuts speculative froth.

Nor is the AI infrastructure build-out obviously running ahead of reality. Goldman also estimates that demand for AI training queries is growing at roughly 350 percent a year and demand for frontier models at about 125 percent, while compute efficiency improves by only about 40 percent annually. Finally, the Financial Times’s Richard Waters keenly notes that the much-touted $1.4 trillion data-center pipeline has only about 10 percent actually committed. The world faces a shortage of capacity, not a glut. Meanwhile, the giants funding the boom mint cash, trade at tolerable multiples, and already book real AI revenues. Painful corrections may come, but the classic conditions for a bubble simply are not present.

Of course, both my baselines could be badly wrong. Other outcomes are plausible. Maybe we’re in a bubble that is starting to pop even now. Or maybe we’re 18 months from living in a different world. Feel free to speculate.

Green shoots

That said, what interests me most is this question: Is there any good evidence that the past three years of AI advances are now spreading throughout the US economy and have begun to be used productively and profitably by American business?