Agents will compress bank multiples as AI revenue nears $170 billion

1. AI is the only macro — ▶ 1:46

   Rates, the Fed, oil and tariffs haven't helped people make money; AI has. Semiconductors are up 60%+ year to date, the fourth straight year of at least 40%, yet Visser hears daily from people who want to short them. Earnings and margins are expanding while rate-dependent businesses suffer, producing a K-shaped economy. The root is 2007: the smartphone built an economy on phones, and tech including Amazon, Tesla and Google is now 56% of the index versus about 13% then. Homebuilders, banks and cars no longer dominate, so the economy is far less rate-sensitive.

2. Coding agents to personal agents — ▶ 6:46

   Anthropic and OpenAI combined had $30 billion in ARR at the end of last year and will finish around $170 billion, more than the entire software stack, driven by coding agents. Those got good enough to spawn personal agents that connect to apps and use a computer, which arrived in September (Grokbot arguably in August). Visser says Muse is growing as fast as ChatGPT did at launch. He is "certain" Siri and Alexa will be able to order food within 6 months, and says Google already has a version with Spark.

3. Multiple compression, not collapse — ▶ 10:24

   Agentic commerce's real effect is deflation among the middlemen who take a toll. Visser splits the world into an enterprise economy and an AI-native one. Anthropic and OpenAI are worth $3.5 trillion before IPO, having been worth almost nothing 15 months ago, and AI natives use Stripe and open-source models, not banks. Public companies are valued on where they'll be in 3-5 years. If it's unclear a business exists by year three, terminal value hits a cliff and multiples drop. SaaS got re-rated this way, and the S&P 500 saw multiple compression this year.

4. Bank apocalypse, slowly — ▶ 12:28

   Salesforce and Adobe survived the SaaS apocalypse, but their stocks fell despite earnings. Banks peaked relative to the S&P around September 15 and now sit below their 200-day moving average, coinciding with Muse's success. Smart investors tell Visser they can't overweight banks. He cites Torsten Slok's viral piece on an agentic run on a bank: banks take deposits at zero and park them in Treasuries at 4.5%, a spread that shouldn't survive if cash sits in Stripe. His call is not that banks fall but that their multiples compress over the next 3 years.

5. Agents end inertia — ▶ 14:36

   Crypto and Stripe already offer cheaper, safer alternatives to bank services. Agents add decision-making that starts from "save me money": subscriptions first, then whether your cash could earn a higher yield elsewhere, answered instantly. Visser used Muse to cancel Fios and move to YouTube TV, and it reads his email and flags savings. Bank inertia, the reason people keep one bank since college, dissolves once it works for one task. "It will happen. The question is when." And once it starts, it never reverses, which hits the growth rate investors price.

6. Skeptics don't use it — ▶ 19:54

   At 5-6 dinners in 6 weeks, Visser found peers his age mostly don't use AI yet hold strong opinions, like reviewing a restaurant they've never visited. Many guests with "this will end badly" views have written books on it or are habitual bears. His counter: in 2008 no one could forecast Uber or Airbnb, so no one knows the economy in 5 years. He works from trailing facts such as GPU rental prices and company revenues over 6-12 months, and won't worry until a data point turns. He was himself called a permabear for most of his life.

7. Why he turned — ▶ 25:18

   In May 2008, at a China coal conference, Visser heard the room describe his own trade and closed most of his portfolio three months before Lehman, finishing his best year. After the crisis he decided recessions were over, then in 2012 concluded China had stopped growing. In 2013 he went to Silicon Valley and Singularity University and stopped worrying about debt, assuming the printing press would keep running. He now listens to under-35s over economists: Alexander Wang at 29, and Dylan Patel at 30, whose SemiAnalysis went from $20 million to $100 million revenue using AI.

8. Bitcoin as pure belief — ▶ 32:41

   Earnings are growing near 30% and revenue 14-15%, yet the S&P isn't up 30%. Visser reads that mismatch as a signal. Humanoids, about 3 years out, are held back only by intelligence. Because "all moats are disappearing" in a world of fast AI competition, Bitcoin, which has no fundamentals and is simply a belief like gold, is a store of value in the digital economy. At $2 trillion it's a top-10 asset, and since 2012 it beat everything on Sharpe except the Mag 7. He expects those to become debt-laden data-center utilities, with Bitcoin the S&P 500 of crypto.

9. Jobs and victim status — ▶ 41:08

   With S&P earnings up 30%, job creation is zero, he says, and negative outside healthcare. Graduates with debt live paycheck to paycheck, housing is out of reach, and the corporate ladder has been destroyed, which he thinks drives much of the public's depression about AI. His answer is "AI victim status": you only get hurt if you won't use it. He accuses wealthy doom-sayers on podcasts of having no need to adapt, and says he put his youngest through Python boot camp in high school and isn't sure he'll finish college.

10. Compute bull case, with risks — ▶ 49:51

    Safest trade for 3 years is "agentic compute", which he built an index for: semiconductors, plus Caterpillar and Siemens for turbines and batteries, plus chemicals. Supply won't meet demand in that window. Compute splits like a Hermès scarf and tap water: curing cancer needs frontier compute open source can't supply, and China keeps up by distilling US models. He sees Anthropic around $300-400 billion and OpenAI $200-300 billion next year, and hyperscaler equity is at all-time highs. Biggest risk is hacking (Japan's cyber emergency, a hijacked supertanker) and government slowdowns, with 20-30% corrections likely.

11. Crypto at 100 to 1 — ▶ 58:48

    Visser's consensus-in-a-year call is crypto. Money and stocks settle slowly, and banks profit from the float. Agents choosing the best outcome will avoid it, since crypto settles in seconds. He dismisses FDIC insurance because there's no money in it; the government would print more. He expects the S&P unchanged in a year, with bonds, VC and private equity not working. Wall Street doesn't cover crypto, which fits Peter Lynch's find-what-isn't-covered premise. It's a $3.5 trillion asset on the tote board at "100 to 1", with agents, not humans, its true users.

Personal AI agents will do to banks what coding agents did to software. They will not empty them. They will cut what investors are willing to pay for them. Jordy Visser is careful on this point. The owner of a hardware store is not wiring his cash to Stripe next month. But public companies are priced on where they will be in 3-5 years, and a business that might not exist in year three faces a cliff in terminal value. SaaS showed the pattern. Salesforce and Adobe survived, their earnings held, and their stocks fell anyway. Banks have been sliding since around September 15 and now sit below their 200-day moving average. Visser expects the multiple, not the franchise, to take the damage over about 3 years.

The mechanism is a spread that exists because customers are passive. Banks take deposits at zero and park them in short Treasuries at 4.5%. That works while people do nothing. An agent told to save money does not do nothing. Visser used his assistant to cancel Fios and move to YouTube TV, and he expects the same logic to reach idle cash chasing a higher yield, with the answer coming instantly. He points to Torsten Slok at Apollo, whose viral piece described an agentic run on a bank. Inertia was the moat. Agents remove it. His line on timing is the confident core of the thesis: "It will happen. The question is when."

Why believe agents are arriving now? Because the revenue is already there. Anthropic and OpenAI combined ended last year at about $30 billion in ARR and are heading for roughly $170 billion. Visser says that exceeds the entire software stack. Coding agents did that. They became good enough to produce a second wave: agents that connect to your apps and use a computer for you. He dates that to September and says Muse is growing as fast as ChatGPT did at launch. He expects Siri and Alexa to order food within 6 months. The two labs are worth $3.5 trillion before IPO, and 15 months ago they were worth almost nothing. An AI-native economy is growing on Stripe and open-source models, outside the banking system. That is the deflation he means: pressure on every middleman who took a toll.

All of this only makes sense if the old macro frameworks are broken, and Visser argues they have been since 2007. After the financial crisis the Fed could create facilities at will, which removed the recession tail. Corporate debt shrank relative to the economy and the equity market. Then the smartphone built an economy that lives on phones. Technology, counting Amazon, Tesla and Google, is now 56% of the index against about 13% in 2007. The economy is less rate-sensitive because homebuilders, banks and cars no longer dominate it. Hence the K shape: anything sensitive to rates is in recession, and everything else is booming. Semiconductors are up 60% this year, the fourth year in a row above 40%. People trying to short them are betting on history books. Visser's view is that "all moats are disappearing", because AI accelerates competition.

That leads to his most striking epistemic claim: nobody can forecast this, so forecasts are noise. In 2008 no one could have predicted Uber, Airbnb or what Amazon became. So he works from trailing facts such as GPU rental prices, company revenues and the trend of the last 6 to 12 months. Until a data point turns, he sees no problem. He is scathing about peers who warn of disaster without using the tools. He compares them to someone reviewing a restaurant they have never visited, and notes that many are habitual bears. He has been a bear himself. He was called a permabear for most of his life. He has also flipped before: after 2008 he decided there would never be another recession, then reversed in 2012.

His trade follows from the thesis. The safest place for the next 3 years is what he calls agentic compute: semiconductors, but also Caterpillar and Siemens for turbines and batteries, plus chemicals. Supply will not meet demand in that window. He expects the market to split like a Hermès scarf and tap water. Open-source models handle ordinary work, while scarce frontier compute does jobs like curing cancer. Chinese models keep pace, he says, by distilling US ones. He sees Anthropic at roughly $300-400 billion and OpenAI at $200-300 billion next year, $600-700 billion combined. He also notes that hyperscaler equity, which sits below the debt, just made new highs, so the equity market is not confirming the credit worry.

Now the strongest objection, which he mostly concedes. His case is extrapolation, and extrapolation is all he says anyone has. He expects 20-30% corrections and points to the July unwind in AI names, when nothing happened except crowding. He names hacking as the main risk: Japan declaring a cyber emergency, a supertanker off Texas taken over. He also expects governments to try to slow things down. Then there is the cost no market metric captures. Earnings are growing about 30%, yet job creation is zero, and negative outside healthcare. Graduates with debt live paycheck to paycheck and housing is out of reach. "The corporate ladder has been destroyed", he says, and he thinks that loss of hope drives much of the public hostility to AI. His answer is "AI victim status": you are only hurt if you refuse to use it. That is an answer for individuals. It is not an answer for a labor market with no entry rung.

The last move is the boldest. Agents, he argues, are the true users of crypto, not humans. Money and stocks settle in days. Crypto settles in seconds. The delay benefits only those holding the float, and agents choosing the best outcome will route around it. Bitcoin, which he calls a belief like gold with nothing but a moat, is the S&P 500 of that world. Wall Street does not cover it. He puts it on the tote board at 100 to 1 and expects it to be the best-performing asset of the next year, in a year when he thinks stocks, bonds, VC and private equity do nothing. The logic is internally consistent. It is also asserted rather than shown. He offers no evidence that agents choose crypto rails today. The thesis rests on what they will do.

What remains unresolved is timing, and he knows it. He says it will happen and cannot say when. His own clock is short: Siri and Alexa ordering food within 6 months. If that lands, the bank-multiple thesis gets its first real test. If it does not, a call built on a faster curve than anyone has seen loses its anchor.