In 2015, the median valuation of listed software publishers was 4.6 times their revenue. Slack had just grown from $1.2 billion to $2.8 billion in six months, Zenefits had just raised $500 million at a $4.5 billion valuation, and Brad Feld had just formulated the Rule of 40. A year later, the founder of Zenefits resigned, LinkedIn lost 43.6% in a single session, and Square went public at less than half of its last private valuation.

In 2026, the median valuation of listed publishers is once again 4.6 times their revenue, after peaking at 19.3 times in December 2020. At the same time, OpenAI raised $110 billion in a single round, Anthropic surpassed a $965 billion valuation, Mistral raised €3 billion in Paris, and Cursor, at $100 million in annualized revenue in January 2025, was acquired for $60 billion in August 2026.

The temptation is strong to overlay 2015 onto 2026 and conclude that a correction is coming. Perhaps. But the analogy is useful in another way: it reveals what happens inside a company, on its board, and to its leader when the valuation precedes the income statement. This is what this article examines, without predicting a crash and without denying the risk.

Key Figures to Know

Why This Topic Is Essential Now

Three facts from the past four weeks point in opposite directions.

On September 8, Mistral raised €3 billion, the largest ever round for a European AI company, at a valuation that has nearly doubled in a year. On the same day, Cognition was valued at $48 billion, before announcing an annualized revenue of $1 billion on the 25th. On September 29, Bain published its annual technology report: to fund $1.5 trillion in annual infrastructure spending, an AI market of $6 trillion per year would be needed by 2031, which is $4.2 trillion more than current projections. The report states it bluntly: infrastructure is being built ahead of the demand curve.

That same month, in France, AI captured 82% and then 79% of the amounts raised by startups over two consecutive weeks. The question for a fund, a director, or a CFO is concrete: at what price to enter, remain, or sell.

Two incorrect interpretations are circulating. The first says that 2026 is 2000 and that everything will end like the Nasdaq, at -78%; it forgets that AI company revenues are multiplying tenfold each year. The second says that growth justifies everything; it forgets that the Rule of 40 was written a few months before the first SaaS correction. The 2015 precedent lies between these two extremes.

2015, the Precedent

The SaaS era of 2014-2016 was the last time an entire category of software was valued on a promise of growth, and then called back to its accounts.

Slack went from $1.2 billion in October 2014 to $3.8 billion in April 2016, for a revenue that would not reach 400 million until 2019. Zenefits raised 500 million at 4.5 billion in May 2015; nine months later, its founder resigned following regulatory bypass issues. Square went public in November 2015 at $2.9 billion, compared to $6 billion privately a year earlier. In February 2016, LinkedIn lost 43.6% in a single session due to a downward revision of its outlook.

The table of multiples tells the rest of the story over ten years.

Three lessons help us interpret 2026. Multiples do not return to the average on a gentle slope; they correct in steps, and each step has a name. Private valuation and public price are two different things, and the gap is settled at the IPO, never before. And in every case where the correction took the form of a crisis, governance had failed before the accounts: Zenefits did not fall because its market turned around, but because its growth had been achieved by bypassing rules that the board had failed to notice.

2026, the Hatching

The 2026 sequence has no equivalent in scale. In 2015, Slack was raising $120 million. In February 2026, OpenAI raised $110 billion at $840 billion post-money, with Amazon contributing $50 billion, and SoftBank and Nvidia $30 billion each. In May, Anthropic raised $65 billion at $965 billion, after growing from about $1 billion to $47 billion in annualized revenue in sixteen months. Both filed confidential IPO registration statements in June and captured, by themselves, 43% of global venture capital in the first half of the year.

The phenomenon is not limited to research labs. Cursor multiplied its annualized revenue by thirty in sixteen months before being acquired for $60 billion. Cognition grew from 10 to 48 billion in one year, Harvey from 3 to 15.5 billion in eighteen months on 190 million in 2025 revenue, and Databricks added 56 billion in six months. In Europe, Mistral is worth over €21 billion and Lovable, founded two years ago, is valued at $13.3 billion.

The table below compares valuations to known revenues. Methodological note: these ratios are rough estimates calculated from heterogeneous public figures (post-money on one side, annualized or annual revenue on the other), not published multiples.

Two things become clear. The leaders that publish their revenue—Anthropic, Cursor, Databricks—are trading at between 20 and 27 times, which is not absurd for revenues multiplying tenfold each year, and remains below the multiple of the top ten listed publishers. As soon as the revenue is thinner or older, the ratio climbs to 50, 65, or 80 times and becomes a promise of future growth rather than a measure of the present. This is where the risks lie.

Where the Parallel with 2015 Holds

The decoupling of private and public pricing. Square went public at less than half of its private valuation; Box waited ten months. In 2026, the two largest AI companies have filed their IPO documents, and the secondary market is already signaling: according to EquityZen, transactions are occurring at 38% below the last round for pre-AI software, with a premium for AI. The question of the actual price has not yet been put to the labs by a public market. It will be.

The correction in steps. LinkedIn lost 43.6% in a day. On February 3, 2026, $300 billion was wiped off listed software publishers in a single session when the market realized what AI agents do to per-user licensing. SaaS went from 19 times to 4.6 times in six years, and every step carries a date.

Governance failing before the accounts. Zenefits in 2016; WeWork in 2019, whose IPO was withdrawn for reasons explicitly tied to governance, valuation, and outlook; Builder.ai in 2025, which fell into insolvency following disclosures about inflated revenues. In each case, valuation had outpaced control structures.

Concentration. The number of deals is at a ten-year low, even as funding amounts break records, and two companies captured 43% of global venture capital in six months. The rest are raising capital on terms they would not have accepted two years earlier.

The return to efficiency metrics. The Rule of 40 was born in February 2015. In June 2026, a Navigate Ventures investor explained that investors are returning to capital efficiency, the burn multiple, and this very same rule. The cycle begins again in the same place.

Ignored warnings. Sam Altman himself stated in August 2025 that he believed a bubble existed, and the Bank of England, the IMF, and the Bank for International Settlements have pointed to the risk. Back in 2000, warnings were not lacking either.

Where the Parallel Breaks Down

Four differences change the nature of the problem.

Orders of magnitude. The funding rounds of 2026 are roughly a thousand times larger than those of 2015. A 50% correction in the valuation of Anthropic alone would wipe out over $480 billion, equivalent to all global venture capital for a record-breaking half-year. A mistake in 2015 resulted in layoffs and a disappointing IPO; a mistake in 2026 impacts the balance sheets of the world's largest market capitalizations.

Cost structure. Listed software publishers boast a median gross margin of 76%: serving one more customer costs almost nothing. Conversely, Bessemer documents fast-growing AI companies reaching $40 million in first-year revenue with a gross margin of around 25%, often negative, because compute cost is proportional to usage. In this model, growth costs money instead of making it.

The nature of capital. SaaS in 2015 was equity-funded by venture capital firms. AI in 2026 is funded by its suppliers and clients (Amazon and Nvidia at OpenAI, Amazon and Google at Anthropic), by debt ($35 billion at CoreWeave, €830 million at Mistral for its GPUs), and by reciprocal contracts ($100 billion announced by Nvidia for OpenAI, $300 billion between Oracle and OpenAI). FrenchWeb summarized this mechanism in January: capital is temporarily playing the role of the customer, waiting for the market to take over. The top ten market capitalizations account for 38.9% of the S&P 500; a market downturn would not just affect venture funds.

Speed. No 2015 SaaS multiplied its revenue by ten for three consecutive years. Anthropic did, Cursor went from 100 million to 3 billion in sixteen months, and Lovable reached 100 million eight months after launch. This speed makes 20x multiples sustainable if maintained, but untenable once growth slows to a normal SaaS pace. This is a binary bet that did not exist in 2015.

One final difference: 2026 SaaS is not just the precedent, it is also the victim. AI is not the SaaS of ten years ago; it is what has caused it to plummet.

The Risks of High Valuation

Whether or not a bubble exists, an excessive valuation produces identifiable effects, which can be seen today.

The gap between capital expenditures and available revenues. Sequoia raised the question as early as June 2024 regarding the $600 billion in missing revenue needed to justify compute expenditures; Bain raises that figure to $4.2 trillion by 2031. On the client side, MIT estimated in August 2025 that 95% of enterprise AI pilots failed to reach their revenue goals, and Deloitte reports that while 20% of organizations derive revenue growth from AI, 74% merely hope to do so. Final demand exists, but it is moving slower than the infrastructure waiting for it.

Capital circularity. When a chip vendor invests in its customer who buys its chips, and when a cloud provider signs a $300 billion contract with a company in which it holds a stake, value circulates in a closed loop. While not fraudulent, this makes actual demand harder for a board to read and links the fate of multiple balance sheets to a single outcome.

Debt. CoreWeave carries $35 billion in debt with a single customer accounting for over 60% of its revenues. A business funded by debt to purchase GPUs whose rental prices are falling must grow faster than the depreciation of its asset. SaaS companies never faced this constraint.

Governance and people: the risk boards notice least. H Company raised $220 million in seed funding in May 2024; within eighteen months, its five co-founders had left, and the CEO was replaced by an execution-focused executive from Palantir France. At Scale AI, the founder left for Meta a month before layoffs. At Thinking Machines, two co-founders returned to OpenAI six months after a $2 billion round, at a time when Meta is offering up to $100 million to researchers. In a company whose value relies on a handful of individuals, the investment thesis can walk out the door with any one of them.

An early investor in Groq summarized the mindset on September 28: "We fully expect about half of them to fail." This is a description of venture capital, not a prophecy; the difference with 2015 is that this half is measured in hundreds of billions.

What This Demands of Executive Roles

A valuation at 20 times revenue demands growth of 100% or more for several years. At 65 or 80 times, it demands never disappointing. It changes the nature of every executive role.

Two lines warrant special mention. At Anthropic, it is the CFO who publicly presents the last three funding rounds, in a company whose annualized revenue was one billion eighteen months ago: a profile capable of communicating with investors committing tens of billions and preparing for listing, which is exceptionally rare. As for the founder, Zenefits, WeWork, Builder.ai, and H Company show that the execution leader arrived after the crisis in every instance. The lesson is to hire them during the round that makes the company too large to be run like a research project.

The Case of French and European AI Startups

This is not just an American issue. Mistral is worth over €21 billion after raising €3 billion from Samsung, EQT, and PSG, alongside €830 million in debt for its GPUs. Lovable, in Stockholm, is valued at $13.3 billion. Dust, in Paris, raised $40 million from Sequoia with over $20 million in annualized revenue. Over two weeks in September, AI captured 82% and then 79% of the amounts raised by French Tech, and Europe experienced its best quarter in four years in Q2, at $24 billion, with more than $10 billion going to AI.

Two specificities apply. Revenues are further behind valuations than in the US: Mistral is valued on a target revenue, and FrenchWeb pointed out that infrastructure is only sustainably sovereign if clients agree to fund its operations. Additionally, European capital is thinner, leaving less room for the half that will fail. For a European fund, the question is not whether to have exposure to AI, but what portion of that exposure relies on companies whose revenue matches their valuation. H Company, with $220 million in seed funding and five departed co-founders, is a textbook case of a valuation established ahead of the organization.

Frequent Mistakes

Reading 2026 using 2015 multiples. A 20x multiple on a revenue that is growing tenfold each year and a 20x multiple on a SaaS with 30% growth do not mean the same thing. The board must compare the multiple to both speed and gross margin.

Confusing funded demand with real demand. When a portion of revenue comes from a shareholder, a supplier, or a reciprocal contract, the board must know what would remain if capital stopped acting as the customer.

Waiting for the first slowdown to hire the execution leader. In all precedents, they arrived after the crisis. The right time is the round that detaches valuation from current revenues.

Treating gross margin as a detail that growth will resolve. At a 25% gross margin, each additional customer costs money, and scale does not mechanically improve a model where compute cost remains proportional to usage.

Ignoring key-person risk. A company whose value rests on five researchers does not have the risk profile of an 800-person SaaS. Retention is a board-level issue.

Neglecting governance transparency. A foundation appointing the entire board, a trust electing the majority, or a founder controlling without owning: these structures are legitimate, provided they are explained to those contributing tens of billions without voting rights.

How to Assess a Leader for a High-Valuation AI Company

1. Can they relate their company's valuation to its gross margin and compute cost, and not just to its growth? A leader who only talks about annualized revenue has not yet faced the question the public market will ask.

2. Have they ever delivered news of a slowdown to a board or to investors, and how? The response distinguishes someone managing a trajectory from someone managing a narrative.

3. What portion of their revenue comes from players who are also shareholders or suppliers? If they cannot answer precisely, the board does not know either.

4. Have they ever successfully retained a team when faced with offers they could not match? What they describe reveals whether they understand that their company's value rests on people as much as on a product.

5. How do they describe the role of the execution leader alongside them? A founder who describes it as a constraint imposed by investors is not ready. One who describes it as the condition for remaining a founder is.

Frequently Asked Questions

Is there an AI bubble in 2026? The facts support both interpretations. Sam Altman stated in August 2025 that he believed in a bubble, and Bain estimated on September 29, 2026, that $4.2 trillion in annual revenue is missing by 2031 to fund infrastructure. Conversely, JPMorgan concluded in December 2025 that AI does not meet the classic criteria of a bubble, and market leaders are publishing revenues that multiply tenfold each year.

What are the valuation multiples of SaaS in 2026? The median of listed publishers is valued at 4.6 times their revenue in August 2026 according to Aventis Advisors, below the ten-year average of about 5 times, compared to a peak of 19.3 times in December 2020. The top ten publishers trade at 31.6 times.

How much is OpenAI worth? $840 billion post-money after the $110 billion funding round on February 27, 2026, the largest venture capital round in history, compared to $157 billion in October 2024. The company filed a confidential IPO registration statement in June 2026, according to Crunchbase.

How much is Anthropic worth? $965 billion post-money after the $65 billion Series H on May 28, 2026, with annualized revenue of $47 billion, compared to $61.5 billion in March 2025. The company filed a confidential IPO registration statement on June 1, 2026.

How much is Mistral AI worth? Over €21 billion post-money after the €3 billion Series D on September 8, 2026, led by Samsung Electronics alongside EQT and PSG Equity, the largest ever round for a European AI company. The annualized revenue target is around $1 billion by late 2026.

What happens to a company when its valuation corrects? Precedents show a fairly consistent sequence: layoffs, replacement of the founder with an execution leader, delayed IPO or going public below the last private valuation, sometimes an acquisition. Zenefits in 2016, Square at its IPO, WeWork in 2019, Builder.ai in 2025, and H Company in 2025-2026 illustrate this, and in every case, human issues preceded the financial correction.

Key Takeaways

• Listed SaaS is valued at 4.6 times revenue in 2026, as in 2015, down from a peak of 19.3 times. AI companies are trading at between 20 and 80 times, as rough estimates.

• The parallel holds on mechanics: private/public decoupling, correction in steps, governance failing before the accounts, concentration, ignored warnings.

• It breaks down on structure: funding rounds a thousand times larger, gross margins of 25% vs 76%, capital provided by suppliers, debt, and reciprocal contracts, unprecedented growth speed.

• The risks are visible today: $4.2 trillion in missing revenue according to Bain, capital circularity, debt secured by depreciating assets, founder departures.

• What this demands: a CFO capable of preparing for a listing two years after creation, a board addressing pricing before the market does, an execution leader hired during the mega-round.

The Laroze Partners Perspective

Most debates about AI valuations focus on a single question: will there be a correction, and of what scale. No one knows the answer. What we observe is more concrete: a valuation that outpaces the income statement changes what the company expects from its leaders long before a market delivers its verdict.

In the precedents of 2015 as in the cases of 2026, the breaking point was never financial first. It was human: a founder alone at the helm of a company that had grown too large for their decision-making style, a board without the means to distinguish real demand from funded demand, a team whose value could walk out the door with a few individuals. The Laroze Pattern®, our method for strategic analysis of trajectories, leadership behaviors, and performance dynamics, looks for exactly this in a leader's background: not whether they have grown revenue, but whether they have already managed an organization whose value had outpaced its foundations, and what they did when that gap began to close.

The AI companies that survive the decade will not necessarily be those that raised the most. They will be those that, when valuation detached from revenues, put in place the people capable of bringing the two back together before the market did it for them.

Sources

Aventis Advisors, SaaS valuation multiples, updated August 31, 2026 · Clouded Judgement (Jamin Ball, Redpoint), August 21 and September 11, 2026 · Bessemer Venture Partners, State of AI 2025 (August 13, 2025) and State of the Cloud 2024 · Crunchbase News, February 27, May 28, July 2, July 7, July 9, August 25, September 3, and September 28, 2026 · CB Insights, Venture Trends Q2 2026 · Mistral AI, press release, September 8, 2026 · FrenchWeb, January 5, September 8, and September 20, 2026 · Maddyness, June 11, 2025, May 18, September 7, and September 25, 2026 · Bain, Technology Report 2026 (September 29, 2026) · Sequoia Capital, "AI's $600B question" (June 20, 2024) · Fortune, August 18, 2025 (MIT NANDA) · Bank for International Settlements, Annual Economic Report (June 28, 2026) · Deloitte, State of AI in the Enterprise 2026 · BCG, AI Radar (January 15, 2025) · Cursor, Cognition, Harvey, Lovable, Anthropic (official press releases and blogs) · Wikipedia (Slack, Zenefits, Box, Block, LinkedIn, WeWork, Builder.ai, Scale AI, Thinking Machines Lab, Meta Superintelligence Labs, CoreWeave, S&P 500, AI bubble), accessed September 29, 2026.