Summary

AI has broken the annuity thesis that supported public SaaS valuations. Coding-model advances over the past six months have investors questioning terminal value, stock-based-compensation treatment, and even which products survive; because reported results look backward, Lucas Swisher expects another 3–9 months of uncertainty. Until then, sequential revenue, net-new ARR, and retention are the best guideposts—but “why own anything” when every company supports both a bull and bear case?
Public software may look statistically cheap, but private markets increasingly hold the investable future. Harry Stebbings points to Monday.com at roughly 1.5x revenue and Wix at 2.5x, versus private rounds near $10 billion; Swisher counters that cheap assets are often cheap for a reason. Of the roughly 20 private “platform companies,” he estimates 18 would already have been public a decade ago: “If you want to own the future, you kind of have to be in the privates.”
For exponential growers, Coatue considers valuation last—but only after proving the company attacks a gigantic, expandable market. Lovable rose from $3 million to $20 million of revenue while its Series A documents were completed, turning an apparent 70x multiple into 10x; similarly, $20 million of ARR at a $3 billion valuation looks cheap if revenue becomes $200 million, then $600 million, then $3 billion. Swisher says price still matters, but valuation should be considered last.
High entry prices require outcomes far beyond the old $10 billion-company test. For a roughly $50 million ARR company at a $5 billion post-money valuation, Swisher needs to believe it can eventually reach $5 billion of revenue, at least 30% margins, and continue growing—meaning there must be $50 billion of revenue to go get. The test is now whether it can become an enduring $50–100 billion public company whose next investor can still make 3x.
Concentration and repeated investment—not spraying early bets—make mega-growth-fund economics work. Swisher says 20 companies represent 80% of private-market enterprise value and four represent 65%; a $1 billion investment that returns 10x alone produces 2x on a $5 billion fund. The best round is often “the double-down round,” while the percentage of companies that 10x can counterintuitively rise with market-cap bands.
Margins matter at scale, but early gross margin can misprice architecture-shift winners. Snowflake had roughly 20% margins early, while Databricks also had very low early margins despite conventional SaaS expectations of 80%; AI applications may improve as token costs fall and workloads move among proprietary, frontier, and smaller models. They may retain lower gross margins because they pay both cloud and LLM providers, yet AI-driven reductions in engineering, sales, and legal expense could produce higher terminal operating margins.
OpenAI and Anthropic embody different forms of strategic durability as AI shifts from assistants toward agents. OpenAI combines a formidable consumer franchise, Codex-led enterprise expansion, and the “unknown unknown” of Jony Ive’s device work; Anthropic pairs its coding beachhead with support across clouds and Trainium, TPUs, and GPUs, giving it capacity and allies. Swisher now believes machine inputs can address labor at a scale he doubted 12 months ago, though enterprise integration means adoption will still take time.

Deep dive

1. AI has turned SaaS terminal value into the disputed variable

Swisher’s diagnosis of the public-software selloff starts with a broken promise: SaaS businesses were valued like insurance companies, with recurring revenue and profit pools extending “forever and ever and ever.” Recent coding models from Anthropic, OpenAI, and others now make investors question that terminal value—and, with it, generous treatment of stock-based compensation and GAAP versus non-GAAP earnings.
The second shock is indiscriminate uncertainty. A design platform can plausibly become more valuable by embedding AI throughout creation, yet the opposite thesis is equally coherent: “I just create all my designs in ChatGPT now, so why would I even need this design tool?” When almost every public SaaS company supports both cases, investors take their capital elsewhere.
The near-term evidence Swisher wants is sequential revenue growth, rising net-new ARR, retention, and customer behavior. His hedge matters: earnings are retrospective while products are changing almost continuously, so “for the next three months, six months, nine months, we’re not really going to know” which companies are genuinely being displaced.
Harry’s opportunity-cost challenge is sharp: Monday.com near 1.5x revenue and Wix around 2.5x—with a cited $4.5 billion market cap against $2 billion of revenue—look safer than private rounds near $10 billion. Swisher concedes public liquidity is valuable, but warns that apparently cheap securities “oftentimes look really cheap for a reason.”

2. The future has migrated into private platform companies

Swisher’s public-private distinction is less about headline multiples than exposure. Public investors can trade easily, but it is “very hard to own the future”; an investor seeking concentrated exposure to token production, frontier models, or the fastest-growing AI applications may need OpenAI, Anthropic, SpaceX/xAI, and private downstream companies.
Harry’s own desired “stocks”—Anthropic, Revolut, and OpenEvidence—cannot be bought in public markets. Coatue calls the leaders platform companies: huge, fast-growing, multiproduct businesses that could operate publicly but elect to remain private. Swisher estimates that 18 of today’s top 20 private companies would probably already have listed under the market structure of a decade ago.
That shift is both an access problem for ordinary investors and an opportunity for flexible private capital. Swisher does not want a mandate forcing him into a Series B every year; Coatue’s metaphor is a “rowboat that rows up and down the river,” deploying wherever the best risk-adjusted opportunity appears.

3. Exponential growth makes price the final question

Lovable is the cleanest specimen of valuation compression through execution. Harry said that during its Series A process, revenue rose from about $3 million to $20 million. What began as approximately 70x revenue had become 10x before closing—prompting his joke that founder Anton should have reopened negotiations.
Coatue therefore asks about valuation last when growth is 10x or 50x year on year. Swisher’s illustration: $20 million of ARR at a $3 billion post-money valuation looks absurd until revenue becomes $200 million in one year, $600 million the next, and eventually $3 billion. The real underwriting task is finding businesses capable of staying on that curve.
The old internal hurdle was the “$10 billion public company test.” Larger AI markets have raised it to whether the business can become an enduring public company—perhaps worth $50 billion or $100 billion, depending on stage. Market pull must be strong enough to make both the revenue curve and later earnings path believable.
Swisher’s concrete framework is that a $50 million ARR company at a $5 billion post-money valuation must plausibly reach $5 billion of revenue, with at least a 30% margin and continued fast growth. That means there must be $50 billion of revenue to go get; without the mega-market premise, paying a mega-market price is indefensible.

4. The best entry earns the right to keep buying

Harry presses the opportunity-cost problem: even if a company can grow from $50 million to $250 million and then $750 million, why choose the investment over ten simpler alternatives? Swisher’s answer is optionality—the initial round may not be best, but it can secure access to later rounds in a company whose “best days are ahead of it.”
Jeff Horing’s maxim inside Coatue’s thinking is that “the best round is the double-down round.” The litmus test is qualitative: if Coatue invests at $5 billion and execution is excellent, is the idea, founder, and market strong enough that it would eagerly invest again six months later at $10 billion?
Harry adds that investors underestimate “the ease of the next double”: moving Harvey from $6 billion to $12 billion may be much easier than creating a $6 billion company from zero. Swisher’s internal chart goes further—the percentage of companies that 10x rises across valuation bands, making a $10 billion-to-$100 billion 10x more probable than one in the preceding band.
This does not make price irrelevant. Swisher says there is always a point where entry valuation erodes returns enough to walk away, but “generational companies, it’s almost never too late for them.” When Coatue instigates or preempts a round, it can also help establish what it considers the appropriate current price.

5. Mega-fund math demands concentration in mega outcomes

Swisher says roughly 20 companies have generated 80% of private-market enterprise value, while four account for 65%. That distribution makes “spray and pray” dangerous: an investor can choose the wrong horse, commit attention to the wrong market, and invest time in the wrong opportunity.
He distinguishes a $3 billion venture fund from a $5 billion growth fund. The former must acquire meaningful early ownership in too many exceptional outcomes, a “tough putt”; the latter can exploit companies staying private longer. A $1 billion investment that returns 10x creates $10 billion—already a 2x gross return on a $5 billion fund.
Bigger AI outcomes complete the argument. In the SaaS wave, Salesforce, Workday, and ServiceNow represented only a few hundred billion dollars of market capitalization collectively, constraining fund-scale returns. If AI substitutes tokens for human inputs and addresses labor pools, Swisher expects materially larger companies. For a mega-fund, traditional vertical SaaS can remain an excellent business without being the best deployment of capital.
A 3x investment is not exciting enough by itself. To deliver roughly 3x net and about a 25% net IRR, a portfolio containing a 1x needs a corresponding 5x; a zero needs a 6x, and a 2x needs a 4x. Swisher must believe that after his 3x, another investor can rationally underwrite another 3x: “Somebody’s got to sit on the other side of that stock.”

6. Durability comes from crossing markets, not defending one product

Databricks illustrates what Swisher wants from a platform founder. Since Coatue invested in 2019, he has watched Ali Ghodsi repeatedly reinvent the company—from an ETL and data-transformation layer, to running inference and training models, to becoming the center of enterprise data. Each transition found another S-curve rather than merely extending the original one.
Market and founder are inseparable, but Swisher still puts market size first. A superb founder in a niche with no natural expansion can build an excellent first act yet struggle to produce acts two, three, and four. The platform company instead demonstrates an ability to “skip TAMs” and repeatedly widen the attainable outcome.
Canva passes that test despite Harry’s challenge that Figma is worth $11 billion and image generation sits directly in frontier-model providers’ path. Canva moved from yearbooks to online design, then SaaS, then a suite of roughly a dozen fast-growing products; it also began integrating AI before ChatGPT, after Cliff Obrecht contacted Coatue about the shift.
Swisher’s mistakes usually come from overestimating TAM or a company’s ability to launch multiple products—not from missing a metric, weak growth, or a bad team. That is why the respectable SaaS company moving from $10 million to $25 million can be “good” without fitting Coatue’s strategy or offering a clear terminal value.

7. Margin matters at scale; retention determines whether low margin survives

Swisher keeps the principle but changes its timing: “Margin matters at scale.” Hyperscalers were low-margin early, while Snowflake had roughly 20% margins and Databricks also had very low margins early, despite investors insisting SaaS required 80%. During an architecture shift, early gross margin can be actively misleading.
The AI bull case is a falling cost curve. An application with 10% inference margin today may have been negative one quarter ago and super negative two quarters ago; over time it can route workloads among its own models, frontier systems, and smaller cheap models. Swisher expects optimization, though he preserves the structural caveat that AI companies pay both cloud and LLM suppliers.
Lower gross margin need not mean lower operating margin. AI may reduce the required engineering, sales, and legal expense base, producing greater operating efficiency than the prior generation. The likely profile is larger revenue pools and somewhat lower gross margin, with terminal operating margin potentially higher because opex falls.
His data doctrine is equally qualified: “Data is a prerequisite. It is not the answer.” A low-margin AI company must show exceptionally sticky behavior and high retention because it has no room for error; yet investors living entirely in Excel can miss the forest, as Swisher once did when Databricks’ net-new ARR failed to accelerate dramatically in a particular quarter.

8. Capital can help proven product-market fit but can also distort seed economics

Coatue’s clearest lesson from 2021 is that pre-revenue companies with no product and very high valuations are not its business. Swisher argues that investors excluded from established platform companies sometimes move toward whatever part of the market their mandate permits; Coatue instead wants real businesses, rapid growth, durability, and credible liquidity.
Harry shows how mega-funds can distort seed economics: his firm offered $3 million on a $15 million valuation, while a larger investor offered $10 million on $100 million with no liquidation preference, no pro rata, and no other protections. Swisher agrees seed ownership is harder to obtain as AI businesses require more capital and emerge with larger rounds and valuations than SaaS startups did.
He rejects literal “kingmaking.” Tier-one investors and abundant capital can deter competitors and become a major advantage when product-market fit is already “insane,” funding sales capacity to capture an active market. But too much capital without fit can be a disadvantage; no syndicate can simply decree that competition is over.
The force-feeding risk depends on stage. Scarcity can sharpen an early company, while growth businesses with genuine traction and measurable return on invested capital can absorb rapid successive rounds. Danger appears when growth funds chase venture-stage companies, encouraging complacency and spending before the underlying engine exists.

9. Great judgment sees the inflection without worshipping the spreadsheet

Mary Meeker taught Swisher to express a complex company in a few Excel lines and tell stories through data. Early at Kleiner Perkins, he arrived stronger at founder conversations than modeling and was “absolutely destroyed” in an exercise; her ability to spot an error in a detailed model shaped his emphasis on analytical precision.
Mamoon Hamid supplied the complementary lesson: detect the moment a business “kinks up.” With Figma at roughly $500,000 of ARR and InVision viewed as the winner, Hamid studied retention and usage within major customers—Swisher recalls Google, Square, and Amazon—and decided within about 30 seconds: “We’re doing it.”
Swisher’s most memorable founder meeting was Winston from Harvey. Language models excel at text in and text out; law is intensely text-heavy; and Harvey’s document-generation and analysis thesis made founder-market fit immediately obvious. Coatue still lost the Series A, reinforcing the consolation that for truly great companies, “there’s always another round.”
His enduring miss was Anduril’s billion-dollar round. As a metrics-focused SaaS investor, he saw an ugly P&L and passed, missing the founding team, the importance of the trend, and where the world was heading. It remains his clearest example of mistaking a prerequisite—financial analysis—for the answer.

10. Public markets still offer feedback, legitimacy, and clean liquidity

Longer private lives create secondary liquidity, particularly for early funds, but Swisher does not expect every platform company to remain private forever. Public markets still provide capital at true scale and true liquidity, avoiding “layers and layers and layers of SPVs,” opaque ownership, and cap-table administration that companies themselves may dislike.
His second argument cuts both ways: public markets are an “incredible feedback mechanism.” Netflix’s transition from discs to streaming was identified and debated by analysts and public investors early; however imperfect any individual 25-year-old analyst may be, the aggregate market acts as a weighing machine for founders navigating another architecture shift.
Listing also makes a major company harder to “mess with.” Its stock is embedded in 401(k)s and indices, and its public status creates a protective institutional rigor. Swisher’s eventual test is therefore concrete: will his public-market colleagues want this stock more than every competing opportunity in their book?

11. Frontier-model winners need both product advantage and allies

Swisher declines Harry’s forced OpenAI-versus-Anthropic choice but gives distinct bull cases. OpenAI owns an exceptional consumer franchise, is gaining enterprise strength through Codex and large transformational deployments, and carries an “unknown unknown” through its acquisition of Jony Ive’s company—an option on a device category that may take 5–10 years to reveal itself.
Anthropic’s case begins with coding, the first AI use case Swisher believes truly took off. Because “everything in the digital world is code,” that beachhead expands into analytical enterprise work. Building for every cloud and for Trainium, TPUs, and GPUs requires infrastructure investment but improves cost, deployment flexibility, and access to scarce compute capacity.
Coatue also asks, in Philippe Laffont’s framing, “Who’s going to want to help you and who’s going to want to hurt you?” Anthropic’s architecture gives more counterparties an interest in its success. Harry notes that this sounds suspiciously like kingmaking; Swisher’s concession is precise: allies “certainly help,” even if they cannot guarantee the winner.
Swisher’s biggest 12-month change of mind is outcome size: using Claude Code convinced him the market is moving from assistants toward agents and from human inputs toward machine inputs. Anthropic, he says, reached $9 billion of ARR while growing 800%, versus roughly 60% average growth for the three hyperscalers at that scale—faster than SaaS, though integrations, deployment, and sticky human behavior mean enterprise transformation will still take time.

Inside Dan Sundheim's Bets on Anthropic, OpenAI, and SpaceX The End of SWE-Bench Verified — Mia Glaese & Olivia Watkins, OpenAI Frontier Evals