Evidence since April 2023 clearly shows (1) a large and growing overhang of overvalued late‑stage unicorns/zombiecorns and (2) a major talent and capital shift into AI startups—but available data do not let us say that a large share of those zombiecorns have already undergone recapitalizations that wipe out common equity, nor that such recaps are the primary driver of founders/employees moving into AI.

1. Zombiecorn overhang and valuation stress

• Analysts estimate that roughly 1,000+ unicorns were created in the 2020–21 boom and that hundreds are now “zombie unicorns”: overvalued, growth‑stalled, and stuck with high preference stacks and little prospect of exiting at their last private valuations. Multiple venture analyses describe a large accumulation of such zombie unicorns, with companies “neither dead nor alive” and LP capital trapped. (www-tc.aicoin.com)
• One investment‑bank analysis notes that of the 787 companies that became unicorns in 2021, an estimated 517 have not raised any capital since then—consistent with Friedberg’s point that many late‑stage companies are functionally stuck below their prior valuation marks. (business-money.com)
• CB Insights/PwC data on the global top 100 unicorns in 2023 show only 25 down rounds among all unicorns that year, with 79% of top‑100 unicorn valuations unchanged, and only a handful seeing large valuation increases (mostly AI). This indicates stress and stagnation, but not yet a broad wave of formal recaps. (pwc.co.uk)
• A PitchBook/EquityZen summary reports that 44% of late‑stage funding rounds in 2023 were flat or down, again confirming widespread valuation pressure but not specifying that common equity is being wiped out en masse. (blog.equityzen.com)

2. Recaps and cram‑downs wiping out common: present but not clearly “a large share”

• There are documented late‑stage “cram‑down” or recap deals where existing shareholders—often including common—are effectively wiped out or drastically diluted, such as Tonal’s proposed financing at ~90% below its prior valuation and Plenty’s recap round reportedly valuing legacy shares at under $15 million vs. a prior $1.9 billion valuation. (theinformation.com)
• Legal and VC‑practice commentary notes increased use of pay‑to‑play and cram‑down terms in 2023–24, particularly in distressed later‑stage financings. (axios.com)
• However, broad quantitative data show that many unicorns have avoided such painful recapitalizations so far by cutting burn, doing small insider extensions/bridges, or simply not raising, rather than fully resetting their cap tables. Several venture overviews explicitly state that incentive misalignment has “delayed the necessary market rebalancing,” with many zombie unicorns still living off past mega‑rounds instead of taking down‑round recaps. (www-tc.aicoin.com)
• Because of the opacity of private markets, there is no comprehensive dataset showing that a large share of late‑stage zombies have already been recapped in ways that zero the common; the best available sector‑level data (on down/flat rounds and lack of follow‑on funding) are consistent with some harsh recaps but also with a substantial number of companies still in limbo.

3. Founder/employee exodus into AI startups

• There is strong evidence of a big shift of founders, operators, and researchers into AI during 2023–25. High‑profile examples include:
  • Ilya Sutskever leaving OpenAI to found Safe Superintelligence Inc. (SSI), which rapidly reached multi‑billion‑dollar valuations. (en.wikipedia.org)
  • Mira Murati leaving OpenAI and launching Thinking Machines Lab in 2025, quickly hiring ~30 top researchers from Meta, Mistral, and OpenAI and raising multi‑billion‑dollar funding. (en.wikipedia.org)
  • Mustafa Suleyman leaving Google/DeepMind to co‑found Inflection AI. (en.wikipedia.org)
  • Other senior alumni from Big Tech or successful startups founding AI firms like Sierra (Bret Taylor, Clay Bavor) and AI21 Labs. (fortune.com)
• Funding flows reinforce this: one analysis reports that in early 2024, 75% of $2 billion raised by Kruze Consulting’s startup clients went to AI startups, with non‑AI sectors heavily capital‑starved. (ft.com) Carta’s data show overall startup headcount shrinking in 2023, with departures outpacing hires and new hiring concentrated in areas—like AI—where innovation and funding are strongest. (wsj.com)
• However, media coverage of AI founders and senior hires overwhelmingly features people coming from Big Tech (Google, Meta, OpenAI, Apple), successful or at least non‑recapped startups, and elite research labs—not specifically from recapitalized, underwater zombiecorns whose common equity was just wiped. The causal link Friedberg emphasized (equity wipeouts at zombiecorns → founders won’t “go work for the VCs” → they instead start AI companies) is plausible but not well‑documented at scale.

4. Why the outcome is ambiguous

• Directionally, Friedberg captured real macro trends: there is a large cohort of overvalued late‑stage private companies with preference overhangs; down rounds, cram‑downs, and failures have increased; and founders/employees are disproportionately flocking to AI, where the funding window is open.
• But to declare the prediction right would require evidence that a large share of those zombiecorns have already undergone recapitalizations that effectively wipe out common, and that these events are clearly driving a major exodus of their founders and employees into AI startups. Public data do not establish either point quantitatively.
• To declare it wrong would require evidence that such recap/wipeout events are rare or that talent is not moving into AI; instead, we see a growing (but only partly measured) incidence of harsh recaps and a very clear shift of talent and capital into AI.
• The time window is also still in progress: “over the next few years” from April 2023 reasonably extends beyond late 2025, and several venture commentators explicitly say the real re‑pricing and cleansing of the zombie unicorn overhang has been delayed and may still lie ahead. (www-tc.aicoin.com)

Because (a) private recap data are opaque, (b) aggregate statistics show valuation stress and some recaps but not a clearly documented broad wipeout of common across zombiecorns, and (c) the AI‑talent boom is evident but not clearly traceable specifically to those recap events, the prediction’s accuracy cannot be determined in a data‑driven way at this point. It is directionally on‑trend but not verifiable or falsifiable at the level of specificity Friedberg claimed, so the fairest classification is ambiguous.