Note to the team
1. Key Themes
Theme: The AI trade is tightly coupled to Nvidia and the macroeconomy, so an unwind would be broad and painful
Nvidia as the system's linchpin
Rechtman argues the AI capex ecosystem is so interconnected that trouble at Nvidia propagates everywhere.
"The AI capex trade is too interconnected and too much of that interconnectivity runs through Nvidia for this to not matter."
The economy is now the AI trade
The macro exposure means an AI unwind isn't contained to tech.
"And it's been clear for a long time that the unwinding of the AI trade would be painful for the macroeconomy; the economy IS the AI trade at this point."
Nvidia as a "bank" dependent on Jensen
The recursive financing structure depends on one person's dealmaking capacity, which raises succession risk.
"Nvidia has functionally become a bank running on his singular charisma, relationships, dealmaking, and recursive access to capital. Some CFO running the company as interim CEO will not have the authority and charisma to muster $100B in credit agreements and guarantees to prop up the neoclouds, model labs, etc."
Theme: Historical downturns offer three playbooks for investors, and each worked for the right type of fund
Playbook 1: Be greedy when others are fearful
Funds that provided liquidity quickly, at stale or depressed valuations, made outsized returns in prior crises.
"During FTX/SVB Greenoaks and others made great money stepping into the crisis to provide liquidity to companies like Rippling. They acted extremely quickly, put big dollars to work, and bought in at a stale/depressed valuation." "In early covid, everyone who stepped in to provide a backstop to their companies crushed it when interest rates went to 0% and the market ripped. Silverlake ripped Airbnb's face off when their revenue was collapsing. a16z and others did Stripe at $36B at the same time and knocked it out of the park."
Playbook 2: Dollar-cost average and play the game on the field
Consistent deployment through cycles works for early-stage, diversified books that are relatively insulated from macro.
"This is basically Y Combinator. Their valuations and terms are roughly fixed and they only adjust up and down by changing the size of the classes / aggregate dollars out the door." "Similarly, the seed funds that keep going through the GFC wind up killing it with some of the best vintages of all time (eg FRC's Uber and Square fund)."
Playbook 3: Retrench and reposition
Some of the best franchises sat out the bubble's end and built a thesis for the next phase.
"Fred Wilson launched USV as the successor to Flatiron Partners with a web 2.0 thesis in the shadow of the dotcom boom. Essentially he was out of market entirely for two years while working on the thesis for what become of the best franchises of all time."
Theme: From "installation" to "deployment": the application layer is where the next gains are
The installation/deployment pattern
The dotcom analogy suggests the infrastructure phase gets blown up while the application phase creates the durable winners.
"And they all made their huge gains backing the deployment phase (internet companies) after watching the installation phase (fiber and networking companies) get entirely blown up."
Rechtman's conclusion for AI
"It is my strong belief that we are about to move from the frenzy and financial bubble to the synergy and the deployment phase of AI." "The application layer and 'neoeconomy' (new businesses made possible by AI) are the closest parallel to the web 2.0 thesis, and the overall arc of history clearly bends in the direction of more power, diffused intelligence."
Theme: Frontier data and training infrastructure remain areas of active investor interest
Specific areas of focus
Despite a cautious macro view, Rechtman is ramping up on specific niches.
"I'm also starting to ramp up on couple areas of particular interest: 1. Robotics data and model training 2. Long running simulation 3. Continual learning and self play"
Network gaps
"My network today is weakest in the robotics data value chain so that's of particular interest."
2. Contrarian Perspectives
"Beta neutral" is an illusion in a crisis, so staying the course isn't a safe option Many early-stage funds assume they're insulated from public-market swings. Rechtman pushes back on this.
"But in a crisis correlations go to 1 and it would be a mistake to believe that we're truly neutral beta just because we're not super levered long. Our companies need to raise capital, our valuations move in correlation with the market (even if they're lower across the board), and our capital base is indexed / levered long even if we're not. So blindly 'staying the course' does not seem like a viable option."
Evidence: Portfolio companies' fundraising needs, valuation correlation with markets, and LPs' long-biased capital bases all transmit macro risk to a seed fund.
The catalyst for the AI unwind may be trivial, but the consequences are not Rechtman expects the trigger to be something seemingly minor (he calls it "profoundly stupid"), yet argues it still matters, even as markets partly rebounded that day.
"It just never seemed possible that the catalyst would be so profoundly stupid as this." "Irrespective of the short term swings in the market (Nvidia has reversed some of its pre-market losses and S&P is up on the day), this will be the beginning of the end the period of financial frenzy."
Evidence: He points to the interlocking financing structure, not the news itself, as the source of fragility, and to the CEO transition as an accelerant.
An AI bubble popping doesn't invalidate the long-term AI thesis, and may be bullish for application-layer investors
"And they all made their huge gains backing the deployment phase (internet companies) after watching the installation phase (fiber and networking companies) get entirely blown up."
Evidence: Historical precedent (USV, Founders Fund, Accel, a16z in the GFC) shows the best returns came after the infrastructure bust, not before it.
3. Companies Identified
Nvidia
- Description: AI chip leader and central node in AI capex financing.
- Why mentioned: The expected starting point of the AI unwind; functions as a financier to the ecosystem.
- Quote: "It was always obvious or likely that the unwinding of the AI boom would start with Nvidia."
- Description: HR/workforce management platform.
- Why mentioned: Case study of investors buying discounted shares during FTX/SVB and the 2022 downturn.
- Quote: "Kleiner Perkins and others aggressively started buying discounted shares in names like Rippling and SpaceX as the markets ground to a halt and insiders panicked."
SpaceX
- Description: Private space and launch company.
- Why mentioned: Example of discounted secondary purchases during a market freeze.
- Quote: (see above)
- Description: Corporate card and spend management fintech.
- Why mentioned: Thrive, GC et al invested in a down round.
- Quote: "Soon after Thrive, GC, et al bought into Ramp in a substantial down round."
- Description: Travel marketplace.
- Why mentioned: Silver Lake provided capital when revenue was collapsing in early COVID.
- Quote: "Silverlake ripped Airbnb's face off when their revenue was collapsing."
- Description: Payments infrastructure company.
- Why mentioned: a16z and others invested at $36B during early COVID.
- Quote: "a16z and others did Stripe at $36B at the same time and knocked it out of the park."
- Description: Startup accelerator.
- Why mentioned: Exemplar of the "dollar cost average" strategy.
- Quote: "Their valuations and terms are roughly fixed and they only adjust up and down by changing the size of the classes / aggregate dollars out the door."
- Description: Ride-hailing and payments companies.
- Why mentioned: Backed by a GFC-era seed fund (First Round Capital) as examples of vintage-defining returns.
- Quote: "the seed funds that keep going through the GFC wind up killing it with some of the best vintages of all time (eg FRC's Uber and Square fund)."
Union Square Ventures, Founders Fund, Accel, a16z
- Description: Venture firms.
- Why mentioned: Franchises built by repositioning around the next phase after a bust.
- Quote: "Founders Fund, Accel, etc. all look roughly similar in this regard."
- Description: AI-focused venture fund founded by Sarah Guo.
- Why mentioned: Example of a successful repositioning from crypto to AI.
- Quote: "Sarah Guo started Conviction as a crypto fund and completely repositioned to the AI trade as one of the first people there with an extremely correct shot call."
Slow Ventures
- Description: Venture firm (~$325M fund) where Rechtman is a partner.
- Why mentioned: The author's firm; context for the note.
- Quote: "I'm a partner at Slow Ventures, where I lead pre/seed rounds from a ≈$325M fund."
4. People Identified
Jensen Huang
- Description: Nvidia CEO (referred to as "Jensen").
- Why mentioned: His departure is seen as accelerating the unwind, given his central role in financing the ecosystem.
- Quote: "That Jensen is stepping down makes it all the more likely that the effects will be swift."
Fred Wilson
- Description: Founder of Union Square Ventures.
- Why mentioned: Example of sitting out the market and building a thesis before launching a great franchise.
- Quote: "Fred Wilson launched USV as the successor to Flatiron Partners with a web 2.0 thesis in the shadow of the dotcom boom."
Sarah Guo
- Description: Founder of Conviction.
- Why mentioned: Pivoted from crypto to AI early and correctly.
- Quote: "Sarah Guo started Conviction as a crypto fund and completely repositioned to the AI trade as one of the first people there with an extremely correct shot call."
Yoni Rechtman
- Description: Partner at Slow Ventures; author of 99D.
- Why mentioned: The author, seeking connections in robotics data and related areas.
- Quote: "I'm a generalist investor looking for weird takes on important stories: N-of-1 companies taking non-obvious approaches to markets that matter."
5. Operating Insights
Use certainty and speed as currency in a downturn. When capital is scarce, offering fast, certain execution can justify aggressive terms.
"Start being much more aggressive on terms either with existing names in our portfolio (offer extensions to the last round, eg) and/or new names ('i know you were expecting $50M post but we'll do it at $25M today or not at all'). Give certainty and execution in exchange for CoC."
Prepare a "shopping list" before the dislocation. Having target names and a thesis ready lets you act decisively.
"Put together a shopping list of names that we want outside the portfolio and find/create special situations."
Adopt a defensive-but-opportunistic operating posture. Rechtman lays out a five-point plan: slow pace, don't blindly follow on, be valuation-sensitive, be opportunistic, and invest in thesis work.
"Slow down our pace. We can always speed up later. Do not blindly follow on. Treat every dollar as precious. Be valuation sensitive. Assume that the market will correct down." "Do more thesis work. We need to be ready to act in the next act of the story."
6. Overlooked Insights
Compute and capital could both become constrained at once. A brief caveat suggests a harsher scenario than a simple valuation correction.
"But the next few years may be really bumpy, especially if compute AND capital become severely constrained."
Neocloud and model-lab solvency depends on Nvidia-backed credit. The mention that Nvidia may need to "muster $100B in credit agreements and guarantees to prop up the neoclouds, model labs, etc." hints at how much of the AI ecosystem's financing is circular and dependent on a single backstop.
"Some CFO running the company as interim CEO will not have the authority and charisma to muster $100B in credit agreements and guarantees to prop up the neoclouds, model labs, etc."