Peter Singlehurst: Lessons from Turning Down Stripe, Coinbase and Losing Money on Northvalt

Summary

• Baillie Gifford’s private-company investor Peter Singlehurst uses a 10-questions framework (growth over 5 and 10+ years, enduring determinants of success, financial analysis, valuation) and enters at a median of $200M revenue, 70% growth, -14% EBITDA margins — product risk is de-risked, while business-model quality and scalability risk remain. The metric venture treats as “almost an anathema within the venture world” is return on equity: its absence produces the “foie gras-ing of startups” — overcapitalized companies force-fed cash.
• Every investment is modeled to a consistent 5x upside, then tested on probability: a random company has ~5% odds of 5x-ing (30 years of public-market data), so 30–50% probability is a bet “we’ll take every time” — and “if you think that there’s an 80% chance of making a five times return… you’re probably deluding yourself.”
• No LLM positions: offered OpenAI at 300, Grok at 50 or Anthropic at 60, “I would buy none of them” — not because the businesses are bad, but because “we still are trying to define what we think competitive advantage will look like at the large language model level” amid open-source and DeepSeek commoditization. They own the layers where the moat is legible: Databricks and Tenstorrent.
• The bull case for ByteDance — the position he’s considered mad for holding: “the most astonishing revenue and profit generation company in China,” #1 in Chinese online advertising, ~#3 in e-commerce via likely Toutiao and likely Douyin. “Our base case is that it does get banned and we still see a path to making at least five times our money” — TikTok US is not part of the base case for a 2019 position.
• Anduril rhymes with Tesla 2013 and SpaceX 2018: solved hardware problems that are software-enabled, proven product-market fit, decades-stagnant giant markets, and “clear water between them and their next nearest private competitor.” Its last round included only public-market crossover investors beyond insiders — evidence of the institutionalization of growth investing.
• The mistake taxonomy: Northvolt was a true error (“too enamored with the idea of a business… needing to exist” while misjudging execution), Intarcia’s FDA rejection was known uncertainty manifesting. The omissions sting more — skipping Stripe’s ~$50bn down round (“I think that was a mistake”) amid discussion of a possible ~$91.5bn round, while it is priced below Adyen on multiples and growing faster, and passing on Coinbase because “my very clever model… was just wildly off.”
• Discipline in practice: the 2021-vintage fund deployed very little in 2022–23 while everyone played convertible-note games “pretending that companies were still worth what they were in 2021,” then started deploying more in 2024 across six countries. On herding: eight sheep in a pen, one jumps out, how many left? None — “you don’t understand sheep.”
• “You can build a better business by staying private for longer” — liquidity will come from company-facilitated secondaries (Stripe, Databricks) and maybe private dividends, not private exchanges. His one 10-year hold: Bending Spoons, “an immune cell that kind of goes around gobbling up all of these slightly broken businesses” — its addressable market is “the broken parts of the venture capital ecosystem.”

Deep dive

1. Not all bad investments are mistakes — Northvolt was

• Singlehurst’s taxonomy separates losses into two camps. The first bankruptcy Baillie Gifford’s private team ever took was likely Intarcia, a biotech developing a GLP-1 — “imagine if that company had managed to stay solvent” — whose therapy was rejected by the FDA. That was known uncertainty manifesting: painful, “but part and parcel of investing.”
• Northvolt sits in the other camp. “We were too enamored with the idea of a business like Northvolt needing to exist — for all the reasons of energy sovereignty in Europe — but what we got wrong was the team’s ability to execute.” That one he kicks himself for.
• On warning signs: “there’s always signs with hindsight.” The team did make follow-on investments, but when execution flags appeared — and the team believed the financing-round structures would lead to “real alignment within the cap table” — they passed on further capital even as the company kept asking.

2. The entry point: $200M revenue, 70% growth, and the anathema metric

• The strategy has narrowed over a decade to “true growth stage” — no product risk, only business-model quality and scalability risk. Median entry company: $200M revenue, ~70% YoY growth, -14% EBITDA margins. Wise is the template: in at $50–60M revenue, now multi-billion revenue with a business-FX arm that didn’t exist at entry, and a high return on equity despite being loss-making when they bought.
• The load-bearing concept is return on equity — “almost an anathema within the venture world.” Not a criticism of VCs, whose stage makes ROE “somebody else’s problem down the road,” but it “distorts the quality of company formation” through overcapitalization: “you know how they make foie gras… we’ve got the foie gras-ing of startups — but with capital.”
• The growth-vs-profitability debate is a misnomer: “it should be about incremental return on invested capital, about long-run return on equity.” Amazon (firm position since 2004) and Tesla (2013) were held through years of losses while Baillie Gifford was considered mad for owning them — the scalability and competitive advantage eventually manifested as profit.

3. The 10-questions framework, and why AI can’t kill a real moat

• The framework — dating to the $50bn Long Term Global Growth strategy where Singlehurst worked with James Anderson, Mark Urquhart (likely) and Tom Slater before volunteering in 2014 to cover privates — breaks into four areas: growth over the next 5 and the next 10+ years; enduring determinants of success (product, competitive advantage and how it evolves with time and scale, plus organizational culture — not “good” culture but its alignment with the specific mission); financial analysis (can this earn a high return on equity, judged against industry precedents); and valuation, done public-market style, seeking intrinsic value “much much greater than the price we’re able to pay today.”
• Against Harry’s worry that AI’s cannibalization makes enduring advantage unpredictable: the most durable moats “don’t lie in a particular product — ‘my mug is better than your mug.’” Bending Spoons’ advantage is an M&A playbook, an integration machine, and founder culture — “can that erode over time? Yes. Is it something that can be destroyed by AI? I’m not sure.”
• Roughly nine of their ten biggest investments remain founder-led, partly selection (a founder who can’t reach $200M revenue has usually churned out before their stage). Vinted is the non-founder exception — which Harry reframes, and Peter accepts, as “sort of” a refounding.

4. Model everything to 5x — and the omissions that teach

• Every case is modeled to a uniform 5x upside so probabilities are comparable across deals. The base rate: “the probability of any given company going up five times, if you were just picking randomly, is something like 5%.” So a 30–50% probability of a 5x is “really good odds” — “we’ll take those bets every time” — while 80% confidence means “you’re probably deluding yourself.”
• The Coinbase pass is the over-intellectualization cautionary tale: “I created a very elaborate spreadsheet… estimating all the volume and liquidity you would need in Bitcoin to get to 5x returns. I felt like I was being very clever and I was just wildly off.” Sometimes the best investments are obvious — Tesla in 2013 had real-money Model S pre-orders and hired people who had scaled car factories, at a roughly $3bn market cap.
• The Stripe omission: first bought at ~$30bn, watched it mark to ~90 then back to ~50, and skipped the down round over doubts on the nascent software build-out beyond merchant acquiring. “I think that was a mistake. We should have put more in.”

5. No LLM bets until the moat is definable

• They’re shareholders in Databricks and chip/infrastructure player Tenstorrent, but “we haven’t taken the plunge into any of the big AI LLM companies… we still are trying to define what we think competitive advantage will look like at the large language model level” — they think they know what it looks like at infrastructure and distribution, but open-source models and DeepSeek are “forces of commoditization” in the middle.
• Quick-fire confirmation — OpenAI at 300, Grok at 50, Anthropic at 60: “I would buy none of them,” not from criticism but because “I don’t know what the answer is to enduring competitive advantage at the large language model level.” Harry’s counter is distribution — making Google “one of the most unappreciated companies right now”; Peter’s reply: “if you went down that route you’d say Microsoft, wouldn’t you?”
• On the application layer scaling “three, four, five million a week” (what was likely Manus, Midjourney, Lovable, Bolt): discipline doesn’t mean abstinence. “The trick is not paying high prices, it’s being judicious and selective about when you choose to pay a high price” — the danger is telling yourself every company is the special one.
• Harry’s darker worry — has the industry misled a generation of triple-triple-double-double enterprise companies about what revenue scaling now takes? Peter concedes the possibility but argues pre-AI companies with regulatory nuance, fintech especially, have “foundational problems in building those products that aren’t just going to be totally blown apart” by AI tools.

6. Patience, herding, and the sheep

• The fund closed in 2021 deployed very little in 2022–23: “all kinds of games being played with convertible notes and everybody sort of pretending that companies were still worth what they were in 2021.” Real deployment resumed in 2024 as pricing and structure games diminished.
• Harry pushes back — it doesn’t feel rational out there; the excess cash concentrates into known outliers, producing “unbelievably expensive 5, 10 billion rounds” at Series C/D. Peter agrees it’s herding into a small number of names: “the industry is still digesting the trauma of 2021… you look for safety by not being too different from what your peers are doing.”
• Harry’s parable, which Peter loved: eight sheep in a pen, one jumps out, how many are left? The boy says none. “No — you don’t understand sheep.” The antidote is breadth: a 2,000–3,000 company universe, six countries invested in last year — Bending Spoons in Milan, plus Portuguese, Brazilian, Indian and likely Israeli companies — with macro risk (Harry: Brazil has one Nubank in 20 years) handled by demanding a price that pays for it.

7. ByteDance: base case is the ban — and still a 5x

• Asked which holding he’s considered mad for owning that’s actually a killer: ByteDance. “The most astonishing revenue and profit generation company in China… it’s just off the charts.” Likely Toutiao (“a better version of Apple News”) and likely Douyin make it #1 in Chinese online advertising and roughly #3 in e-commerce.
• The kill-shot risk is priced in: “our base case is that it does get banned and we still see a path to making at least five times our money even with TikTok not being part of that investment case.” Position initiated in 2019; liquidity via an eventual US or Hong Kong listing (“one of those is probably a little bit more likely than the other”) and ongoing share buybacks funded by its own profitability.
• China generally remains a big opportunity: “everyone is fearful of China right now — if everybody is saying one thing, that China’s uninvestable, you would be mad not to be questioning that.” He’s planning a summer trip. The firm’s DNA is global — its first-ever investment was a Malaysian rubber plantation supplying tires for the Model T; its first private investment was in China — which is why deglobalization is his single biggest worry.

8. Anduril: the Tesla-2013 / SpaceX-2018 pattern

• The thesis is pattern recognition: products that are largely software-enabled but solve really hard hardware problems, proven demand (“no question about product-market fit”), and “very very large markets that have largely not changed in decades” — with “clear water between them and their next nearest private competitor.” True of Tesla in 2013, SpaceX in 2018, Anduril today.
• In the round, beyond insiders, the only new money was traditional public-market investors who can also do privates — the owners a company needs to transition to public markets. Peter’s history of the growth stage: natural public-market owners got pushed out as companies stayed private, “the opportunists came into that vacuum,” many were pushed out after 2021, leaving ~10–20 consistent institutions — a number he doesn’t expect to double.
• On Elon-adjacent political risk to the analogues: “I worry about it.” Comfort comes from SpaceX’s “amazing management team that is not Elon Musk,” kept deliberately low-profile — but pressed on state contract cancellations and SpaceX’s Canada/Mexico exposure, he concedes plainly: “it’s a big worry.”

9. Stay private longer — and where liquidity actually comes from

• “I think what people realize today is that you can build a better business by staying private for longer.” The reason is focus: public life means misaligned shareholders, telling competitors everything, doing it all “in the cold light of day.” Epic’s Tim Sweeney gave him the countervailing frame: at some point liquidity needs, M&A currency or regulators mean “the easier option is to be public.”
• The liquidity answer isn’t private exchanges — share-class complexity, likely ROFRs and company control kill those — but very large company-facilitated secondary rounds (Stripe, Databricks) “starting to become more of a feature,” and possibly, eventually, dividends paid by private companies.
• Databricks in practice: first invested around $30bn, took pro rata at $60bn — “we stood our hand… a little bit different from saying we’re doubling down” — and yes, he still sees a 5x from 60. A possible ~$91.5bn Stripe round is roughly double Adyen’s market cap, yet on multiples Stripe “is priced even less than Adyen — and it’s growing quicker.”
• On London: “I do share the sentiment that we’re in a dire state for the London Stock Exchange” — a supply problem (Harry can name 10 great £300M-revenue companies, “but you can’t name a hundred”) and a demand problem (UK investors’ risk appetite trails NASDAQ’s). His fix runs before the LSE itself — and yes, Europe probably should have a single public market.

10. The machine, the partnership, and the one stock to hold

• Last year’s funnel: 1,000 companies met → 600 rounds examined → 65 first cuts → 30 deep dives → 11 investments, by a team of 10 drawing on 170 public-market growth investors. Diligence culminates in a “10Q” note — “we don’t use PowerPoint, these look like essays” — debated by the whole team Thursday afternoons (90 minutes and growing), decided by a four-person Friday investment committee. Checks run $10M to $150M; his one wish is “three, five, ten times the amount of personal time” per decision.
• The firm’s structure is built around stewardship: a 115-year-old intergenerational unlimited liability partnership where “carry goes to the firm” and people are paid via synthetic-carry bonuses — stewardship over extraction. His biggest change of mind: growth-stage value-add. “I was just dead wrong on that” — companies do need help going public, being public, building independent boards.
• The one stock he’d hold for 10 years: Bending Spoons — “an immune cell that kind of goes around gobbling up all of these slightly broken businesses,” whose addressable market “is basically the broken parts of the venture capital ecosystem: good products but really bad businesses.” Core risks: acquisition-price sensitivity and whether it scales. It’s also his answer to Harry’s stranded cohort of $200M-revenue, mid-teens-growth companies — “this is their market.”
• The closing bull case for growth investing now: “there’s been a lot of spaghetti thrown at the wall and we get to see which bits stick,” human capital with scaling experience has never been better, and capital availability sits at “this kind of Aristotelian mean — enough to invest, but not so much that it detracts from the long-term quality of those businesses.”