All podcasts / All-In Podcast / Summary

The IPO Comeback: Why Tech Giants Are Finally Going Public | All-In Liquidity IPO Panel

2026-06-06 - 32 min - source - Read full transcript
Jason Calacanis (host)Brad GerstnerAndrew FeldmanWill Marshall

Key insights

Going public changes almost nothing about the day-to-day business.
Andrew Feldman said the IPO process itself is mostly overhead - endless review calls with no value added - and that the morning after the IPO, engineering has made no more progress than the day before, vendor relationships are unchanged, and the team just goes back to work with more cash in the bank.
ipo-comeback
Cerebras got its IPO timing right by getting it wrong for nearly a decade first.
Feldman framed roughly nine and a half years of a difficult path to going public (including scrutiny over a UAE investor under the prior administration) as the necessary precondition for the final twelve months, when market conditions flipped and demand for the offering became overwhelming.
ipo-comeback
For companies with institutional or government customers, going public is a credibility signal, not just a capital raise.
Will Marshall said Planet Labs's government, defense, and agricultural customers needed proof the company would still exist in the future before fully depending on it; being public and having capital access if needed reassures customers who cannot afford for a data supplier to disappear.
ipo-comeback
Most of a successful company's value gets created after its IPO, not before.
Feldman said every study he's seen shows more money is made post-IPO than pre-IPO, both in percentage and absolute terms, because pre-IPO capital deployment into most venture companies is comparatively small; Planet Labs illustrated this directly, going public via SPAC in 2021 at roughly $2B and then not creating 90% of its value until year three or four, later 10x-ing from $5 to $50 a share.
venture-liquidity-strategy
LPs pressuring funds to distribute IPO shares immediately can destroy huge amounts of value.
Gerstner described investing pre-IPO in a company at a $1B valuation, distributing shares at $3-4B under LP pressure, and watching the stock reach $50B within 24 months - a cautionary tale for LPs who demand immediate distribution instead of letting funds hold winners.
venture-liquidity-strategy
The 'dribble lockup' is a new mechanism for releasing IPO shares gradually instead of via a hard cliff.
Rather than a single lockup-expiry date that dumps a flood of shares on the market, the dribble lockup releases shares over about six months tied to performance hurdles - a structure Gerstner said Altimeter helped pioneer and expects SpaceX to adopt in a similar form.
venture-liquidity-strategy
The 'stay private forever' era is reversing; companies are targeting IPOs at much smaller valuations again.
Gerstner said a decade of venture orthodoxy (led by firms like Andreessen Horowitz) pushed companies to avoid public markets as long as possible, but portfolio companies are now increasingly aiming to go public at $1-5B rather than waiting for OpenAI/Anthropic/SpaceX-scale valuations, because public-market scrutiny sharpens focus and execution.
ipo-comeback
Space-based data centers become cost-competitive once launch costs fall to $200-300/kg, and that threshold is a few years away.
Marshall said a Planet Labs study with Google (done eight or nine years prior) found space data centers become cheaper than terrestrial ones at $200-300/kg launch cost; costs are around $1,000/kg today, have fallen roughly 10x in a decade, and Starship's trajectory suggests hitting the threshold in two to three years.
space-based-computing
Space-based compute is fundamentally a power-economics play, not just a hardware one.
Marshall explained that a solar panel in a sun-synchronous dawn-dusk orbit collects roughly 5x more energy than the same panel on the ground and never needs batteries, since it's in continuous sunlight - unlike terrestrial solar, which needs expensive battery, gas, or nuclear backup for intermittency.
space-based-computing
Satellite miniaturization mattered more to the space economy than falling launch costs.
Marshall said satellites that used to cost roughly $1B and weigh 20 tons now cost a few kilograms to a few hundred kilograms while doing as much or more - a mainframe-to-desktop-style shift that, combined with lower launch costs, is what is unlocking new space applications.
space-based-computing
Cerebras rejected the GPU form factor on purpose, because copying Nvidia's architecture could never yield a decisive advantage.
Feldman said that if you build a GPU, the odds of beating Nvidia at its own game are approximately zero because Nvidia has already captured the architecture's low-hanging fruit; Cerebras instead built a wafer-scale chip the size of a dinner plate with memory placed directly next to compute, betting that a fundamentally different architecture was the only path to a large (target: 20x) performance advantage.
ai-compute-infrastructure
AI inference speed is becoming a hard consumer requirement, not a nice-to-have.
Feldman argued the market for slow AI, like the market for slow search or dial-up internet, is effectively zero; users abandon a website that takes 3-5 seconds to load, and Cerebras built its wafer-scale architecture specifically to deliver inference 15-18x faster than GPU-based systems so users don't wait.
ai-compute-infrastructure

Companies

Techniques and frameworks

Summary

This is a live panel recorded at an All-In event, moderated primarily by Altimeter Capital's Brad Gerstner (with Jason Calacanis present and opening the segment), featuring two CEOs of newly public, AI-adjacent companies: Andrew Feldman of Cerebras Systems (the AI chip maker that had IPO'd roughly three weeks earlier) and Will Marshall of Planet Labs (the earth-imaging satellite company that went public via SPAC in 2021). The conversation opens with both founders describing the anticlimax of going public: enormous procedural overhead, a celebratory moment for employees and their families, and then a return to the exact same operational problems the business had the day before. Feldman is candid that Cerebras got its IPO timing right only after nearly a decade of getting it wrong, while a shift in market conditions over the final twelve months made the offering wildly oversubscribed.

The panel then turns to the economics of staying private too long. Gerstner argues, using Planet Labs's own trajectory (a ~$2B SPAC entry in 2021 followed by 90% of the company's value creation arriving in years three and four, and a later 10x stock move) as Exhibit A, that the venture industry's decade-long "stay private forever" doctrine cost LPs enormous upside by keeping high-growth companies out of public markets during their steepest value-creation years. He shares a personal story of an LP-driven decision to distribute shares early in a company that later ran from $3-4B to $50B, and frames the newer "dribble lockup" structure (gradual, performance-gated share releases instead of a hard lockup-expiry cliff) as one mechanism the market is building to make later, staged liquidity events more palatable to both founders and investors.

Marshall and Feldman then each explain the technical bets underlying their businesses. Marshall lays out the case for space-based data centers: launch costs have fallen roughly 10x in a decade and are still a few years from the $200-300/kg threshold where space-based compute becomes cheaper than terrestrial, largely because continuous, unobstructed sunlight in a dawn-dusk orbit delivers about five times the usable solar energy per panel with no batteries required. He also credits satellite miniaturization (from billion-dollar, 20-ton satellites to sub-hundred-kilogram units) as a bigger unlock than launch costs alone. Feldman explains Cerebras's core architectural bet: AI unlocked entirely new categories of compute problems (images, language, real understanding rather than storage), and rather than build a faster GPU, Cerebras built a wafer-scale chip with memory placed directly next to compute, reasoning that matching Nvidia's own architecture could never produce the 15-18x speed advantage Cerebras eventually delivered to customers like OpenAI.

The episode closes with both founders framing the next several years as an acceleration point where AI and space converge: Marshall's vision of "planetary intelligence," where large language models trained only on internet text gain access to real-world sensing data, and a shared view among the panelists that public-market scrutiny sharpens execution rather than distracting from it.

Notable Quotes

"It's really difficult to overestimate the amount of garbage that's involved in going public... your engineering projects have made no progress since the day you weren't public... not a damn thing changes in the important parts of your business." - Andrew Feldman

"AI is only as good as the data it's trained upon." - Will Marshall

"If you build a GPU, the odds that you're better than Nvidia in our view are approximately zero." - Andrew Feldman

"How big is the market for slow search today? It's zero... You will not wait for AI. We have to deliver it to you in real time." - Andrew Feldman