Larry Aschebrook

Larry Aschebrook built G Squared around a five-to-seven-year liquidity strategy: convert late-stage private-company access into cash DPI. The portfolio “lands” with small checks, expands through repeated transactions, and concentrates 80%—hopefully 90%—of risk in about 10 companies. He calls TVPI and MOIC “fake numbers”; DPI “is the only thing you can use to buy food.”

Spotify validated the model and changed G Squared’s scale permanently. After six straight days of being denied a meeting in Stockholm, Aschebrook saw 25% Swedish penetration and learned the record labels were shareholders. Spotify then offered $150 million of stock, forcing him to assemble the money in 60 days and borrow the final $9 million from an early investor. G Squared ultimately put 40% of its $380 million third fund plus roughly $700 million of co-investment into Spotify, producing about $1 billion for LPs.

The realized record came from selling into demand, not perfectly forecasting eventual winners. G Squared owned 16% of Coursera, returned $800 million to LPs after selling around $36, and watched the stock later trade near $8; Lyft returned roughly 3x while Uber lost about 20 cents on the dollar, or approximately $50 million. “We made our multiple and went home and distributed the cash.”

The 2020 vintage broke when G Squared mistook a booming market and prior liquidity for proof that its judgment could not miss. It deployed about $900 million from early COVID through 2021, treated 12x LTM ARR-to-enterprise-value pricing for SaaS as conservative against public multiples near 25x, and later saw the comparable multiple fall toward 4x. Toast at $76 became Aschebrook’s “canary in the coal mine”: “We’ve overpaid for all of it.”

The rescue required admitting the error, raising another $300 million and buying protection while markets burned. G Squared sold inflated positions, lowered cost bases through secondaries, and negotiated structured equity paying a 25% IRR or 2.5x, whichever was greater; about 70% of the vintage became primary exposure, with structure on 40% of that. “As the house is on fire, we’re running in the front door with cash.”

His worst mistakes separate bad underwriting from bad behavior after underwriting. Theranos cost him personally a few million dollars to escape a binding agreement to buy roughly $50 million of stock; 23andMe could have produced about a 2x return when he began selling, but he chased a larger multiple and later said the position lost about $70 million. At Getir, the damaging decision was investing another $100 million rather than accepting the first loss: “You were dead man walking without knowing it.”

Co-investment now amplifies only positions the fund itself has already designated as core. Earlier vintages used co-investment at up to four times fund capital, and the 2020 vehicle even accommodated LP-requested thematic one-offs; Aschebrook said those experiences taught him that investors may blame the manager when a single-shot position fails. The current model restricts co-investment to conviction names such as Anthropic, Fanatics, Wiz, Databricks, Turo and Monzo.

On AI, Aschebrook would pay for leaders that have reached “escape velocity” rather than hunt for another foundation-model entrant. He sees little room beyond OpenAI and Anthropic, would buy Anthropic around the stated $61 billion valuation “all day long and twice on Sunday,” and agreed that it is an extraordinary business after Harry described a $350 billion-to-$1.5 trillion OpenAI scenario over five years. The hedge is picks-and-shovels exposure such as Lambda and Scale AI—and avoiding legacy companies that cannot rebuild AI into their DNA.

🔗 Original source & video: Larry Aschebrook, Founder & MP @GSquared: How We Lost Money on Uber and Made Millions on Lyft