The IPO Comeback: Why Tech Giants Are Finally Going Public | All-In Liquidity IPO Panel
A panel discussion at an IPO-focused event features Brad Gerstner of Altimeter Capital interviewing Andrew Feldman (CEO of Cerebrus Systems) and Will Marshall (CEO of Planet Labs) about their experiences going public, the future of AI silicon, and the convergence of space infrastructure with artificial intelligence. The conversation covers IPO timing, investor lockup innovations, and secular technology trends including space-based data centers and domain-specific AI chips.
Summary
The panel opens with host Jason Calacanis introducing Brad Gerstner, Andrew Feldman, and Will Marshall as representatives of two major technology trends: AI silicon and space-based data infrastructure. Gerstner frames the discussion around the IPO comeback, suggesting 2026 could be a record year for public offerings.
Andrew Feldman, whose company Cerebrus Systems recently IPO'd at $185/share (opening at $320, currently ~$230, valuing the company at $50-60 billion), describes the post-IPO experience as anticlimactic in operational terms — existing problems remain, no customers are magically added, but employees receive emotional validation and the company gains new constituents. He notes that Cerebrus faced a decade of challenges including CFIUS scrutiny due to a UAE investor before a relatively smooth path to public markets.
Will Marshall describes Planet Labs' more gradual market recognition after going public via SPAC in 2021 at a $2 billion valuation. The stock has risen approximately 10x over the past year as investors began understanding the space data market. Planet operates ~200 satellites imaging the entire Earth daily, with roughly 60% of revenue coming from defense and security customers. Marshall emphasizes that being public provides legitimacy with large institutional customers like governments who need assurance of long-term viability.
A significant portion of the discussion focuses on the future of space-based computing. Marshall argues that when launch costs drop to $200-300/kg (currently ~$1,000/kg, down 10x over 10 years), it will become cheaper to operate data centers in space than on Earth. Key advantages include access to continuous solar power in sun-synchronous orbits (5x more energy per panel than ground-based), eliminating the need for batteries or backup power. Planet is already partnering with Google to launch TPUs into space and has launched NVIDIA GPUs. Feldman offers a more cautious timeline, noting that inter-chip communication within space-based clusters remains an unsolved engineering problem.
Feldman explains Cerebrus's differentiated approach to AI silicon: rather than building GPU-like architectures (where beating NVIDIA would be nearly impossible), they built a wafer-scale chip the size of a dinner plate with memory placed directly adjacent to compute. This architecture makes Cerebrus 15-18x faster than GPUs for inference tasks like those used by OpenAI. He contextualizes AI silicon's rise by noting that AI fundamentally expanded the range of problems computers can address — previously, computers could store images and language but not meaningfully process them.
Gerstner raises the investor distribution question, using Planet Labs as a case study where most early VCs (Google, Capricorn, Founders Fund, DST) held their shares post-IPO and captured the majority of value creation. He introduces the concept of a 'dribble lockup' innovation where shares can be released gradually over six months based on performance hurdles — a structure he suggests SpaceX may adopt. Both Feldman and Gerstner argue that more value is historically created post-IPO than pre-IPO, and that staying private too long (as with Anthropic, OpenAI, and SpaceX) transfers enormous value gains to private market investors rather than public ones. Gerstner sees the pendulum swinging back toward earlier public listings.
Key Insights
- Andrew Feldman argues that going public changes almost nothing operationally — existing vendor relationships, engineering progress, and customer pipelines remain exactly as they were, and the primary tangible benefit is employee emotional validation and additional capital.
- Will Marshall states that Planet Labs derives approximately 60% of its revenue from defense and security customers, a larger fraction than originally anticipated, driven by geopolitical demand for advance threat warning capabilities.
- Marshall argues that space-based data centers will become economically cheaper than terrestrial ones once launch costs fall to $200-300/kg, primarily because sun-synchronous orbits allow solar panels to generate 5x more energy continuously without needing battery storage.
- Feldman contends that any new AI chip architecture must look fundamentally different from a GPU to have any chance of outperforming NVIDIA, since GPU designers have already exhausted all incremental optimizations — Cerebrus achieved 15-18x speed advantages over GPUs by using a dinner-plate-sized die with memory adjacent to compute.
- Gerstner argues that historically more absolute dollar value is created for investors after a company's IPO than before, and that the decade-long 'stay private forever' trend pushed by firms like Andreessen Horowitz transferred enormous wealth gains from public to private market investors.
- Feldman frames AI's significance as having expanded the addressable problem space for computers — previously computers could only store images and language, not meaningfully process them, and AI opened those domains to computational analysis for the first time.
- Gerstner introduced a 'dribble lockup' innovation where IPO lockup shares are released gradually over six months based on performance hurdles rather than all at once, which he believes SpaceX will adopt for its eventual public offering.
- Marshall claims that current large language models are effectively 'blind' to the real world because they are trained only on internet text, and that integrating real-time earth observation data into AI models will unlock what he calls 'planetary intelligence' — a vastly larger category of applications.
Topics
Full transcript available for MurmurCast members
Sign Up to Access