The AI Bubble Crowd Has It Exactly Backwards

Lisa Tamati reporting on Jordi Visser — macro investor with 30+ years on Wall Street, head of AI-macro research at 22V Research, and author of the VisserLabs Substack — on why a brutal month for the hyperscalers is a midcycle slowdown, not the top.

June has been ugly for the hyperscalers and the Mag 7, and half the macro world is forwarding around the same take: the bubble is finally cracking. The cleaner read is the opposite. This is not the top. It is a midcycle slowdown, the pause inside an expansion where growth stays positive and the rate of change simply cools. That one distinction is the entire story, and almost everyone shouting "crash" is missing it.

Quick word on whose read this is, because it matters. Jordi Visser is a macro investor with more than 30 years on Wall Street. He started at Morgan Stanley, spent the better part of two decades as President and Chief Investment Officer of Weiss Multi-Strategy Advisers, and now runs AI-macro research at 22V Research while writing the VisserLabs Substack and a weekly AI-Macro-Crypto show that has become one of the sharpest sources on how AI is rewriting market structure itself. What makes him worth your time is rare: most macro people can talk Fed and rates, most AI people can talk models and tokens, and almost nobody holds both in one head and tells you what it means for profit margins, valuations and where the money actually flows. He has been right at the turns that counted, from the Liberation Day bottom to Bitcoin's distribution phase to the semiconductor supercycle. So when he says ignore the crash talk, it pays to understand why.

---

A rotation, not a top

The S&P was off 2% on the week and the Nasdaq dropped nearly 5%, but the rest of the tape tells you what is really happening. Healthcare ripped 8%, real estate and utilities were green, and small caps kept outperforming, with the Russell up another one and a half percent for its third up week in a row. If the doomers were right and three rate hikes were coming, small caps would not be leading. They are. Inflation expectations in the swap and Treasury markets keep falling, two-year yields have handed back the entire hawkish move, and the economy is still running hot. Every "this is the hawkish pivot" podcast amounted to nothing. The index weakness is not macro fear. It is a rotation, and the cleanest way to understand it is the benchmark arbitrage.

---

Short the spenders, own the receivers

Hyperscalers are down roughly 14% month to date, but not because the AI story broke. Their margins are still climbing and they are still not hiring. They are down because they are the ones writing the cheques, and capital is rotating away from high-multiple mega caps toward the businesses receiving the spend. This is the trade Visser has been on for months: long the thematic AI names, short the hyperscalers on a relative basis. The spenders carry multiple-compression risk, the market's growing doubt about whether they can pull revenue in the door fast enough to justify what they are laying out. The receivers, meaning memory, semis, industrials and power, are where the earnings actually land.

The bubble crowd should sit with one number. The annualised return of the S&P since the financial-crisis bottom is about 14.7%, and we are running at that exact pace through the halfway mark of this year. Fifteen years of normal, and somehow this is the bubble. Meanwhile the Nikkei is up 38%, the KOSPI has doubled and the TAIEX is up 53%, not because those markets are wild but because they carry a higher benchmark weighting toward AI than the US does. The receivers of the money are driving everything.

---

This is earnings, not speculation

Here is the part the bears keep missing. The S&P is up roughly 7% year to date, and that is almost entirely the hyperscalers' earnings, not a speculative melt-up. Forward EPS has gone parabolic. The small caps that flatlined from the day ChatGPT launched until late last year are now inflecting higher. Global EPS estimates for 2027 and long-term IT earnings growth near 38% tell the same story. A speculation-driven bull market needs retail energy and breaks when sentiment turns. An earnings-driven bull market only breaks when earnings turn, and to get bearish here you have to forecast when earnings roll over. That is genuinely hard, because the engine underneath is a physical shortage that is not resolving.

---

The memory wall

This is the heart of it. Micron printed a blowout, numbers that look like ten years of earnings compressed into three quarters, signed a supply deal with Anthropic, and the stock still finished the week down. That is the slowdown in the rate of change. But the substance is what matters: Micron told the market it has no line of sight to memory supply catching demand, and does not expect supply to even improve until 2028. We are in 2026.

This is not a normal cycle. Micron has now signed 16 structurally strategic customer agreements, take-or-pay deals with binding volume commitments over multi-year terms, that already represent 20% of volume and are heading toward half of company revenue. Read that as a subscription business, not a commodity one. Demand is being pulled forward by recursive self-improvement: the models are now improving themselves, which makes them better, which makes more people want them, while the hardware to run them does not yet exist.

The tell is Apple. Go looking for a maxed-out Mac Studio and you cannot buy one. The largest configuration got pulled. If Apple, one of the biggest and most important buyers on the planet, is stuck in the queue for memory, what does that say about everyone else trying to build sovereign AI? And the demand still coming has barely started: high-end PCs, autos moving from 20% to over 40% L2-and-above, and humanoid robots that carry roughly ten times the memory of a car. The food the system runs on is in short supply, and the supply chain to make more of it takes years. When supply finally catches up the move lower will be violent, so this is later innings for memory specifically, but at a 7.5x forward multiple on a name still beating every quarter, the risk-reward looks nothing like it did in January.

---

Loops and tags change the margin math

If you are bearish on AI, this is what you are really betting against. The week's big tell was the conversation around loops. The head of Claude Code called them an emphatic yes, the next abstraction layer after agents. Peter Steinberger of OpenClaw fame put out a post that crossed 8 million views arguing you should no longer be prompting coding agents, you should be designing the loops that prompt them. Andrej Karpathy is saying the same thing. A loop is an agent running a workflow on repeat, reviewing, debugging, opening PRs every few minutes, with the human taken out of the middle.

This changes the margin story. Until now, the profit-margin gains came from not hiring, not from firing, which is why there has been almost no job creation outside healthcare for 16 months. Loops change that. Picture a 100-person marketing team where two people are real AI power users. Loops let those two teach an agent their entire workflow, and if the other 98 do roughly the same work, the agents do the other 98. That is the next phase, and it is arriving now. Critically for the infrastructure thesis, token demand stops scaling with the number of users and starts scaling with the number of recurring processes automated. Every workflow that becomes a loop runs around the clock. Add the new agent-in-your-Slack capabilities, agents that read the conversation and just go build the thing by morning, and token usage does not grow, it explodes. Do not get bearish on AI while token usage is going vertical, unless you also believe rates are about to rip higher. They are not.

---

Agentic commerce and the case for Bitcoin

Watch the Stripe Sessions keynote when you get the chance. New business creation on Stripe has gone parabolic, and the standout is the solo operator running a global business with an agentic stack handling the CFO, COO and infrastructure. AI-native businesses are moving from vibe coding to vibe deploying: it is one thing to build an app, another to ship it and get money in the door.

The sequence is the thing to internalise. Agents first write code. Then they operate infrastructure. Then they deploy real products. Then they transact. And to transact autonomously, you need guardrails, meaning programmable money, settlement and identity. That is where crypto stops being a casino and becomes the financial rail of the agentic economy. This is why Bitcoin is the collateral asset on the other side of all this. Tokenization is coming for the roughly 700 trillion dollars of dormant assets sitting outside the global economy, the largest monetary expansion in history, happening on a rolling basis, and the collateral underpinning that world is Bitcoin. It is also why, in a debasement-trade unwind, everything without a valuation, gold, silver, Bitcoin, the high-beta names, capitulates together while the earnings-driven trade keeps working.

---

Crypto: a bear market you do not need to be a hero in

Bitcoin looks ugly. The 200-day is rolling over, it keeps making lower lows, and it printed new lows on the week. The disciplined call is the right one: this is a bear market, you do not need to step in and be a hero, and you do not need to pick the bottom. The patience that worked on Micron works here too, getting stopped out on small adds until momentum actually turns. The interesting tell is the broader ecosystem basket of agentic-economy names, which did not make new lows while Bitcoin did. The ecosystem tends to lead the way out and Bitcoin follows. For anyone not paying attention to crypto right now, the warning is blunt: good luck playing catch-up a year from now.

---

The rotation into healthcare

The capital leaving tech this week did not vanish, a chunk of it rotated into healthcare and biotech, up 8% on the week. This is a genuine renaissance driven by AI, with drug discovery accelerating as labs partner up and apply real compute. The standout equity call is Eli Lilly as a candidate to become the largest company in the world within five years, on the back of GLP-1s and the next generation beyond them, plus a balance sheet buying up IP, biotech and its own compute. As a market theme, the rotation is real and the leadership change is worth respecting.

---

The take

I am with this read on essentially all of it, and the reason is that none of it rests on vibes. It rests on facts you can check: parabolic forward earnings, take-or-pay memory contracts, token usage going vertical, and a hardware wall with no line of sight to relief until 2028. The bubble crowd is staring at hyperscaler capex and ignoring the receivers actually printing the money. That is the mistake.

The honest part, said plainly because it makes the case stronger, not weaker. Frontier-lab projections of when the hyperscalers turn cash-flow positive are speculative and will be revised. Multiple compression on the spenders is real and may have further to run. And the memory trade is later innings: when supply finally catches demand, the move down will be fast, and the market will not have the right multiple on at that moment. The single thing that would change this view is a real rollover in forward earnings revisions, and we do not have it yet. Until we do, the midcycle slowdown is the story, not the crash.

Lisa Tamati reports on AI, markets, and the signals that matter at PTLsignal.com