The 'Dribble Lockup' and the New IPO Architecture
On 7 June 2026, three of the most-watched operators in American tech sat down on the all-in podcast and spent two hours describing the same phenomenon from three different angles. Andrew Feldman, chief executive of Cerebras Systems, the AI-silicon challenger priced at $185 and opening at $320 on its first trading day. Will Marshall, chief executive of Planet Labs, whose Earth-imaging constellation has compounded roughly 10x since its 2021 SPAC debut. Brad Gerstner, the Altimeter Capital founder and Cerebras board member who has now seen more IPOs than most regulators.
The phenomenon: the IPO is back. But the structure around the listing has changed, and the change is the story.
The thesis is straightforward. After a decade in which the smartest venture firms preached stay private forever — delay listing, raise jumbo private rounds, milk the late-stage premium — a cluster of high-profile 2025–26 listings has flipped the script. Cerebras, valued between $50 billion and $60 billion at the time of recording roughly three weeks after its debut, is the headline example. But the more durable innovation is not the price. It is the lockup.
For most of the modern era, a tech IPO has come with a 180-day insider lockup, after which venture and growth funds dump stock into the market, often regardless of whether the business is performing. The pattern kills post-IPO performance, frustrates long-only public investors, and forces general partners to immediately distribute shares to their own limited partners — locking in whatever the secondary market will pay on day 181.
Cerebras, by Gerstner's account on the all-in Liquidity IPO Panel, used a different structure. He called it a "dribble lockup." Instead of releasing all insider shares on a single date, the company is releasing them over six months, gated by performance hurdles — meaning the float expands only if the business hits milestones that retail and institutional buyers can actually verify. "It's the most innovative lockup I've seen in a long time," Gerstner said. He added that SpaceX, the Elon Musk-led private space company, is "planning to use a similar approach" when it eventually lists.
If the structure holds up, the implications ripple well beyond Cerebras.
The LPs Who Don't Want to Sell
The default assumption inside venture is that LPs — the pension funds, endowments, and family offices that sit at the top of the fund stack — are desperate for liquidity after a decade of mark-ups they cannot realise. When a portfolio company IPOs and the lockup expires, the conventional wisdom says, pressure the GP to distribute shares immediately.
Gerstner framed this as a mistake. Quoting on the panel, he argued that "most of the big tech companies went public at a few billion, not a few trillion. Like there's a lot of zeros in between those" — meaning the bulk of the wealth creation historically comes after listing, not before. Feldman agreed, arguing on the same panel that "more money is made after IPO than before, in both percentage and absolute terms," though he acknowledged that venture investors can only deploy "modest capital" pre-IPO while public markets offer much larger pools.
Planet Labs is the case study Marshall held up. The company went public via SPAC in 2021 at a roughly $2 billion valuation, a moment the market quickly wrote off. Marshall described a 10x move since, putting the equity near $50 per share at the time of recording. Google, Planet's largest single investor, "has not sold a share," Gerstner noted on the panel. Capricorn, the early-stage vehicle, held until very recently. "Early VCs captured nearly all of the 10x move post-IPO," he said. The LPs who pressured their funds to distribute on day 181 got the bottom of that move. The ones who let managers hold got most of it.
The dribble lockup is, in effect, a structural answer to that problem. If insiders cannot dump regardless of price, the GP has a defensible reason to hold through volatility, and the LP is told, in advance, that distribution will track milestones rather than the calendar.
Architecture as a Bet
The other thread running through the panel is that this is not just a better-listing-structure story. It is a story about two businesses built on architectural bets that incumbents cannot copy without breaking what makes them money.
Cerebras's wager is direct. "If you want to be 20 times better than somebody, your architecture can't look like them," Feldman said on the panel. The company's chip is roughly the size of a dinner plate, with memory placed adjacent to compute rather than separated by a bus that creates latency. The result, by Feldman's account: 15–18x faster than GPUs on OpenAI workloads, with a single device doing the work of dozens of conventional accelerators. That is not a normal chip race. It is a refusal to compete inside Nvidia's design envelope.
Nvidia's response, broadly, has been to argue that integrated stack and software ecosystem matter more than raw silicon throughput — a defensible position when your software is the de facto standard. Cerebras's bet is that workloads are moving faster than software can adapt, and that raw architectural advantage on inference will eventually be self-justifying. The OpenAI workload data, if it holds, is the first real evidence.
Planet's architectural bet is different. Marshall described a fleet of roughly 200 satellites — the largest Earth-imaging constellation in operation — imaging the planet daily and building a multi-year time series. The defensibility is not the satellite. It is the archive, the revisit rate, and the data pipeline that turns pixels into answers. He framed the total addressable market for Earth observation data alone at $75–100 billion, before factoring in AI applications built on top.
The kicker: about 60% of Planet's revenue now comes from security and defense customers, "higher than initially expected," Marshall said, driven by what he described as geopolitical demand for "seeing threats around the corner" weeks or months in advance. This is not the environmental-monitoring pitch the company started with. It is a surveillance-and-intelligence business that also monitors farms.
The Space-Compute Bet
The most speculative claim on the panel came from Marshall, and it is the one most likely to define the next decade if it lands. Quoting on the all-in podcast, he described a study Planet did with Google "eight or nine years ago" on non-terrestrial data centers. The conclusion: space-based compute becomes cheaper than terrestrial data centers when launch costs fall to $200–300 per kilogram. Current cost is roughly $1,000 per kilogram, down 10x over the last decade, and Marshall said the threshold should be hit within two to three years via SpaceX's Starship.
The supporting argument is that solar panels in sun-synchronous dawn-dusk orbit collect roughly 5x more energy per panel than ground installations, with no need for batteries. The implication is not that every data center moves to orbit. It is that the marginal compute workload — training runs, batch inference, anything that is not latency-sensitive — eventually orbits the economics.
Marshall noted that Planet has already launched Nvidia GPUs and Google TPUs in test missions. "We're not gonna do it on the ground," he said, half-joking about the orbital data center concept. The point was that the experiments are no longer PowerPoint. They are payloads.
The CFIUS Drag
One cautionary note from Gerstner on the panel: Cerebras's path to listing was slowed by a CFIUS review triggered by a UAE investor, a process that, in his account, ran into friction under the Biden administration and contributed to the unusually long gap between the company's filing and its debut. The "9.5 years of difficulty followed by 12 months of extreme ease" framing was his, not a market consensus. But the underlying point — that a single foreign investment in an AI-silicon company can now trigger multi-year national-security review and materially delay a listing — is the kind of friction that does not show up in pitch decks.
It also explains, in part, why the dribble lockup matters. If the regulatory path to public markets is longer and more uncertain than it used to be, founders and boards have an incentive to negotiate every other term harder. Lockup structure is one of the few terms left where innovation is still possible.
What to Watch
Three things will determine whether the dribble lockup is a one-off or a template.
First, Cerebras's price action through the second half of 2026. If the stock holds above the offering price as the performance-gated tranches release, the structure will be studied and copied. If it does not, the experiment dies with one deal.
Second, SpaceX. The Musk-led company has not filed, and any timing claim is speculative, but Gerstner's on-the-record comment that SpaceX is "planning to use a similar approach" is the first credible signal that a dribble-style lockup may scale to a deal an order of magnitude larger than Cerebras.
Third, LP behaviour. If pension funds and endowments accept the structure, and stop demanding immediate distribution at lockup expiry, the secondary-market overhang that has defined post-IPO performance for a generation gets structurally smaller. The investor who benefits is the one who can wait. The investor who loses is the one who cannot.
Feldman closed his segment with a line that applies to more than just chip architecture: "The way to get timing right is by getting it wrong for a decade." The companies that just went public waited. The lockups are being rewritten because of what the wait cost the people who could not.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://www.youtube.com/watch?v=jLICvWE7w2Q