Legal AI Valuations: When the Numbers Stop Making Conventional Sense
Harvey trades at 41-80x revenue. The median SaaS company trades at 7.4x. Either something extraordinary is happening, or something familiar is.
I have lived through enough technology cycles to recognise the pattern. The early excitement. The breathless valuations. The inevitable question: is this real, or is this 1999?
With legal AI, the honest answer is: it is both. The technology is real and genuinely useful. The valuations, however, are telling a story that deserves careful reading.
The Arithmetic
Let me walk through the numbers, because they are worth understanding precisely.
Harvey, the most visible legal AI company, was valued at three billion dollars in February 2025. Five billion in June. Eight billion in December. In ten months, the company's valuation grew by 167 percent. Revenue, by the most generous external estimates, grew by roughly 3.9 times during the same period, from about fifty million to perhaps 195 million in annual recurring revenue.
At eight billion against the confirmed one hundred million ARR from August, that is an eighty-times revenue multiple. Against the higher Sacra estimate, it is about forty-one times. The current median SaaS company trades at 7.4 times revenue. Even during the monetary stimulus frenzy of 2020-2021, when everything with a subscription model was being valued aggressively, the median peaked at about 18.6 times.
Harvey is not a median company, of course. Exceptional growth commands exceptional multiples. But "exceptional" and "sustainable" are different words, and the distance between them is where fortunes are made and lost.
What the Multiples Are Pricing In
Working backward from an eight billion dollar valuation reveals what investors believe, or need to believe, for the number to make sense.
If Harvey eventually trades at ten times revenue, which would still be a premium multiple for a mature software company, the valuation implies approximately eight hundred million in expected ARR at maturity. If the market is generous and Harvey earns a fifteen-times multiple, you still need roughly 530 million.
Either scenario requires the company to grow three to eight times from current levels while maintaining or improving margins. Against Harvey's trajectory, this is plausible. Against the history of legal technology companies, it is ambitious. Clio, one of the most successful legal tech companies ever built, reached three hundred million ARR over seventeen years. Harvey is being valued as though it will surpass that in perhaps three.
The Legal Tech Context
Here is what makes the legal AI valuation question different from the broader SaaS valuation question. Legal technology has historically traded at a discount to general software multiples. The market is relatively small: the entire US legal services market is approximately 350 billion dollars. It is fragmented across practice areas, firm sizes, and jurisdictions. Adoption cycles are long. Buyers are conservative and demanding.
Previous legal technology success stories achieved their outcomes over many years. Relativity reached unicorn status in e-discovery, but it took time. Various contract lifecycle management players were acquired at meaningful but not spectacular multiples. The largest legal tech acquisition before 2025 was measured in hundreds of millions, not billions.
Harvey's eight billion dollar valuation would make it larger than most legal services providers, not just legal technology companies. That either reflects a genuine paradigm shift in how the legal market values technology, or it reflects the kind of optimism that historically precedes corrections.
The Bull Case, Taken Seriously
I want to be fair to the case for these valuations, because dismissing them entirely would be intellectually lazy.
The total addressable market is genuinely large. Legal services globally represent roughly nine hundred billion dollars. If AI tools capture even a modest percentage of that spend, rather than competing only within the traditional legal tech budget, the addressable market dwarfs historical legal technology categories. Harvey already serves corporate legal departments at companies like Comcast, KKR, and PwC, not just law firms. The corporate legal market may ultimately be larger and stickier than law firm sales.
Network effects may be real. AI models improve with usage data. Harvey's early position with sophisticated customers provides training data that competitors lack. If this creates compounding quality advantages, early market share could translate to durable competitive position. This is different from traditional software, where the product is largely static between versions.
Switching costs may be higher than they appear. Once lawyers develop habits around specific AI interfaces, once firms build governance frameworks around particular tools, once training programmes reference specific platforms, the cost of switching rises. This is not theoretical. I have watched organisations struggle to migrate between document management systems for years. AI tools may embed even more deeply.
The LexisNexis partnership changes the equation. Harvey's strategic alliance with the largest legal research platform in the world creates distribution and content advantages that a standalone startup could never achieve. If this partnership deepens, Harvey's competitive position becomes significantly more defensible.
The Bear Case, Taken Seriously
Revenue multiples compress. They always do, eventually. The 2020-2021 SaaS bubble saw companies trading at fifty times revenue fall to ten or fifteen times within eighteen months, regardless of continued growth. Legal AI is not immune to the same dynamics.
Competition is intensifying from both directions. Microsoft, Google, and the major legal research providers all have or are developing AI capabilities. Harvey competes not just with other startups but with companies that have vastly larger resources, existing customer relationships, and distribution advantages. At the same time, open-source models are improving rapidly, potentially commoditising capabilities that Harvey currently monetises.
The AI infrastructure cost question. Running sophisticated AI models at scale is expensive. Gross margins for AI-first companies differ meaningfully from traditional SaaS, where marginal costs approach zero. If inference costs do not decline as quickly as revenue grows, the path to profitability stretches.
The "sweetheart deal" problem. Industry reporting suggests early adopters received favourable pricing to build the customer base. As Harvey moves from land-grab to monetisation, the question is whether customers who adopted at promotional rates will accept full pricing. My experience from other technology transitions suggests this is harder than companies expect.
Slower-than-expected impact. MIT economist Mert Demirer observed: "I will expect some impact on the legal profession's labour market, but not major." Bloomberg Law's 2025 survey found smaller-than-expected changes from AI in every workload category. If the practical impact of AI on legal work proves incremental rather than transformational, valuations built on disruption narratives will not hold.
The Historical Pattern
I want to share what I have seen before, because the pattern is instructive.
In the late 1990s, significant investment flowed into legal technology. The internet was going to transform legal services. Valuations soared. The correction came, and it was painful. Companies like Westlaw and Lexis survived and ultimately thrived, but at more modest valuations than peak enthusiasm suggested.
In 2015-2016, another wave of legal tech investment produced mixed outcomes. Some companies achieved meaningful scale. Others were acquired at modest multiples or failed. The market, as it always does in professional services, tempered the enthusiasm.
The lesson is not that legal technology is a bad investment. Clearly it is not. The lesson is that the legal market has structural characteristics, conservatism, long sales cycles, risk aversion, fragmentation, that have historically moderated the technology valuation cycles that other industries experience.
Whether AI is different enough to override these structural factors is the central question. I think it partially is. AI genuinely changes what is possible in ways that previous legal technology did not. But I think the moderation will come. It always does.
What This Means for Practitioners
If you are not an investor, why should you care about valuation multiples?
Because they predict behaviour. Companies valued at eighty times revenue must grow aggressively. That means aggressive sales tactics, pressure to expand deployments, incentives to overpromise capabilities. It also means these companies will invest heavily in product development, which benefits users.
Because they signal market risk. If a significant correction occurs, some of the companies you depend on may not survive. Vendor evaluation should include financial resilience, not just product capability.
And because they reveal expectations. An eight billion dollar valuation on a legal AI company says the market believes AI will capture a significant share of legal spend. If that expectation is even partially correct, every legal organisation needs a strategy for how AI will affect their economics, whether they adopt the technology or not.
The Prudent View
The growth is real. The technology works. The market opportunity is genuine. And the valuations are, by any historical standard, extraordinary.
History suggests that extraordinary valuations in professional services technology tend to moderate over time. Whether the moderation comes through companies growing into their valuations or through the valuations declining is the only question. I suspect the answer, as it usually is, will be a bit of both.
Alex Blumentals is the founder of Twin Ladder. He has watched technology valuation cycles for three decades and has learned that the market is usually right about the direction and wrong about the timing.