Harvey's $11B Valuation: Enterprise AI Crushes Consumer Hype

Your weekly signal boost from 190,000+ articles, served with a DJ's ear for what actually matters.

So, What Actually Happened?

We scanned 190,000 articles this week so you don't have to. And the signal that cut through everything? AI just grew up. A legal AI startup hit an $11 billion valuation with nearly $200 million in recurring revenue, proving that vertical AI is building real businesses, not just fundraising decks. The same day, OpenAI killed its most viral consumer product and walked away from a billion-dollar Disney deal, because the enterprise pivot matters more than viral downloads. Meanwhile, CrowdStrike and IBM merged their AI agents into an autonomous security operations center that can respond to threats faster than any human team. And at the Gartner Data & Analytics Summit, 4,000 data leaders were told to stop measuring ROI and start measuring something called ”Return on Intelligence.” Nobody is connecting these dots. We are.

The Bottom Line: The consumer AI experiment is ending. The enterprise AI era just started. The companies that figured this out six months ago are now worth $11 billion.

The Tracks That Matter

1. A Legal AI Startup Just Hit $11 Billion. Its Revenue Nearly Doubled in Five Months.

Harvey raised $200 million at an $11 billion valuation, and the numbers behind this round tell a story that most AI funding announcements do not. Harvey's annual recurring revenue hit $190 million in January, up from $100 million in August. That is not a growth curve. That is a vertical line. For a company that builds AI tools for contract analysis, compliance, due diligence, and litigation, the pace of enterprise adoption is unprecedented.

What makes Harvey different from the dozens of AI startups chasing legal workflows? Reuters reports that the company has embedded engineering teams inside its largest law firm clients, building custom AI agents tailored to specific practice areas. This is not a generic chatbot with a legal skin. It is a purpose-built system that understands the difference between a merger clause and a force majeure provision. Co-founders Winston Weinberg and Gabe Pereyra (a former research scientist at Google DeepMind and Meta) built the company around the insight that legal work is one of the few domains where AI can create genuine value without replacing human judgment.

The broader signal: vertical AI companies that solve specific, high-value problems are generating real revenue while horizontal AI platforms struggle to monetize. Harvey's $190 million ARR makes it one of the fastest-growing enterprise AI companies on the planet. When the sector corrections come (and they will), companies with this kind of revenue trajectory survive. Companies with impressive demos and no ARR do not.

”The companies that succeed are going to be the ones that are relentlessly adapting.”
— Winston Weinberg, Harvey CEO

Here's what works: If you are evaluating AI investments (personal or corporate), look at revenue growth rate, not valuation. Harvey nearly doubled its ARR in five months. That is the metric that separates real enterprise AI from hype. Ask every AI vendor you work with one question: ”What is your ARR, and what was it six months ago?” The answer tells you everything about whether they are building a business or burning runway.

2. OpenAI Just Killed Its Most Viral Product and Walked Away from a Billion-Dollar Deal. That Is the Smartest Move They Have Made All Year.

OpenAI shut down Sora, its AI video generation app, and simultaneously ended a $1 billion partnership with Disney. The decision was immediate. No wind-down period. No transition plan. Just gone. Disney pulled out of the partnership after the app and its API were killed just months after launch, citing a shift in strategic priorities.

The reasons behind the shutdown stack up fast. Deepfake concerns were growing, with regulators and content creators pushing back against a tool that could generate photorealistic video from text prompts. Content moderation costs were climbing. Intellectual property lawsuits were multiplying. And most importantly, Sora was not generating meaningful revenue compared to the enterprise API business that actually pays the bills. Analysts note that OpenAI is entering its ”focus era,” trimming consumer products to concentrate resources on its core enterprise and robotics ambitions ahead of a potential IPO.

Here is what nobody is saying: killing Sora will not stop the flood of AI-generated video. Open-source alternatives already exist. But this move signals that the biggest AI company in the world has concluded that consumer AI products without clear enterprise monetization paths are not worth the regulatory, reputational, and operational cost. That is a maturity signal that every AI company should pay attention to.

Here's what works: Audit your AI product portfolio this week. For every AI feature or product you operate, answer three questions: (1) Does it generate revenue or reduce cost in a measurable way? (2) Does it create regulatory or reputational risk? (3) Would killing it free up resources for your highest-value initiatives? If the answer is no, yes, and yes, you have a Sora on your hands. Kill it before it kills your focus.

3. CrowdStrike and IBM Just Merged Their AI Agents. Cyberattacks Now Have 29 Minutes to Hide.

CrowdStrike and IBM expanded their strategic collaboration by integrating CrowdStrike's Charlotte AI with IBM's Autonomous Threat Operations Machine (ATOM), creating what both companies call an ”agentic SOC”: a security operations center where AI agents coordinate investigation and containment at machine speed. The integration extends CrowdStrike's Falcon platform into IBM's managed security services and global X-Force Cyber Range, offering immersive cyber crisis simulations.

The urgency is backed by a sobering number. The average eCrime breakout time has dropped to 29 minutes, with the fastest observed attack reaching lateral movement in just 27 seconds. That means from the moment an attacker gains initial access to the moment they spread across your network, your security team has less time than it takes to make a cup of coffee. Human-only response at that speed is not just difficult. It is impossible. The agentic SOC model puts AI agents in the first responder role, analyzing detections across endpoint, identity, and cloud environments before escalating to human operators for strategic decisions.

This is part of a broader pattern at RSAC 2026, where five major cybersecurity companies announced significant AI-powered capabilities in the same week. The industry is not experimenting with AI anymore. It is operationalizing it. When two of the largest security vendors on the planet merge their AI agents into a unified response engine, the message to every CISO is clear: autonomous security is not a roadmap item. It is a production requirement.

Here's what works: Ask your security team one question this week: ”How long does it take from detection to containment?” If the answer is measured in hours, you are operating at a speed that attackers left behind two years ago. Evaluate whether your SOC has AI-assisted triage and response capabilities. If it does not, the CrowdStrike-IBM partnership just showed you what your competitors' SOCs will look like by year's end.

4. Gartner Just Told 4,000 Data Leaders to Stop Measuring ROI. Here's What They Said to Measure Instead.

At the Gartner Data & Analytics Summit 2026, the analyst firm introduced a framework that should make every CFO uncomfortable: Return on Intelligence. The concept is built on three pillars. First, setting organizational ambitions (what you actually want AI to achieve, not just ”deploy AI”). Second, Return on Integrity, which means strengthening the technical foundations that make AI trustworthy. Third, Return on Individuals, which means empowering the workforce to use AI effectively rather than replacing them with it.

Alation's recap of the summit highlights a key tension: most organizations are still measuring AI success by traditional ROI metrics (cost saved, time reduced, headcount avoided) while the actual value of AI is increasingly about intelligence, the quality of decisions made, the speed of strategic pivots, the accuracy of market predictions. Traditional ROI captures the efficiency gains. Return on Intelligence captures the competitive advantage. They are not the same thing, and most companies are only measuring the first.

The timing matters. This framework arrives at the exact moment when enterprises are discovering that their AI investments are producing mixed results. The deployment is happening. The efficiency gains are real. But the strategic value, the ”intelligence” that was supposed to justify the investment, is elusive because nobody defined what it looks like or how to measure it. Gartner's framework gives data leaders a vocabulary to have that conversation with the C-suite.

Here's what works: Before your next AI budget review, replace the question ”What is the ROI?” with three specific questions: (1) What strategic ambition does this AI initiative serve? (2) Is the data foundation trustworthy enough to support it? (3) Can our people actually use the output to make better decisions? If you cannot answer all three, your AI investment is generating efficiency, not intelligence. That distinction will determine which companies lead in three years and which ones just saved money.

5. A Startup Nobody Has Heard of Just Raised $50 Million to Make the AI Chip Race Irrelevant

Normal Computing raised $50 million to develop thermodynamic computing, an approach that uses the physics of heat dissipation to perform AI calculations at a fraction of the energy cost of conventional chips. While every headline in AI hardware is about who builds the biggest, fastest GPU, Normal Computing is asking a fundamentally different question: what if you did not need a GPU at all?

The thesis is straightforward. Current AI chips burn massive amounts of energy because they fight against the laws of thermodynamics. Normal Computing's approach works with those laws instead, using the natural tendency of systems to find low-energy states as a computational mechanism. If it works at scale, the implications are enormous: AI inference costs drop by orders of magnitude, edge devices can run models that currently require data center hardware, and the energy consumption problem that threatens to derail AI adoption gets a physics-based solution instead of an engineering patch.

This is one of those stories that sounds like a science project until you realize the investors are serious. $50 million for a company that most AI professionals have never heard of is a bet that the computing paradigm itself is about to shift. Not incremental improvement. Not better chips. A different way of computing entirely. History says these bets fail nine times out of ten. History also says the tenth time changes everything.

Here's what works: If you are responsible for AI infrastructure decisions, add one line to your evaluation framework: ”Does this vendor's roadmap assume GPUs will always be the dominant AI compute platform?” If yes, they are optimizing for today's paradigm. Keep an eye on thermodynamic computing, neuromorphic chips, and photonic processors. The company that cracks post-GPU AI compute will make today's hardware leaders look like the mainframe vendors of the 1990s.

6. A $35 Billion Pharma Giant Just Hired AI to Find Cancer Drugs That Actually Work for Individual Patients

Tempus AI partnered with Daiichi Sankyo to use multimodal AI for antibody-drug conjugate (ADC) development, one of the most promising and complex areas in cancer treatment. Daiichi Sankyo, with a $34.6 billion market cap and 14% revenue growth, is betting that Tempus's PRISM2 model can combine pathology images with clinical data to discover biomarkers and stratify patients in ways that traditional clinical trials cannot.

What makes this partnership significant is the shift from research to production. AI in drug development has been a ”promising” field for a decade. This collaboration is not a research project. It is an operational integration where AI models directly inform which patients should receive which drug candidates, accelerating the path from laboratory to treatment. For ADCs specifically (drugs that combine an antibody with a cytotoxic payload to target cancer cells precisely), identifying the right patient population is the difference between a blockbuster drug and a clinical trial failure.

The broader pattern: AI in healthcare is moving from diagnostics and administrative tasks into the core of drug development itself. When a $35 billion pharmaceutical company with 14% revenue growth trusts AI to guide cancer drug candidate selection, that is not experimentation. That is a strategic bet that AI-guided drug development produces better outcomes than traditional approaches. The implications for every pharmaceutical company are stark: if your competitor is using AI to find better drug candidates faster, your traditional R&D pipeline is now a competitive liability.

Here's what works: If you work in life sciences, pharma, or healthcare technology, ask one question at your next leadership meeting: ”Where in our R&D pipeline could AI reduce time-to-candidate or improve patient selection accuracy?” The companies integrating AI into drug development today will own the patents that matter in five years. The ones that wait will be licensing those patents from their competitors.

Signal vs. Noise

🟢 Signal: Vertical AI is generating real enterprise revenue, not just raising capital. Harvey's ARR nearly doubled to $190 million in five months. That is not a valuation story. That is a business story. When an AI company in a specific domain (legal, in this case) can show that kind of revenue trajectory, it validates the thesis that vertical AI solves real problems. Watch for this pattern in healthcare AI, security AI, and financial AI over the next two quarters. The companies with real revenue are separating from the pack.

🟢 Signal: AI-powered security is going autonomous, and the clock is ticking for every SOC. CrowdStrike and IBM merging their AI agents into an agentic SOC is not a product announcement. It is a capability shift. When the average eCrime breakout time is 29 minutes and the fastest is 27 seconds, human-only response is structurally insufficient. The companies that operationalize AI in their security operations this year will have a defensible advantage. The ones that don't will be running a horse-drawn carriage on an expressway.

🔴 Noise: Consumer AI products are getting more attention than their business models deserve. Sora's shutdown is the clearest signal yet that standalone consumer AI apps without enterprise revenue are not sustainable. The deepfake concerns, the content moderation costs, the IP lawsuits, all of these were predictable. The noise is in the ”AI for everyone” narrative. The signal is in the enterprise applications that charge real money for real outcomes.

🔴 Noise: The AI chip arms race is drowning out a more important question. Every week brings another headline about bigger GPUs and faster training chips. But Normal Computing's $50 million raise for thermodynamic computing hints at something the hardware headlines miss: the next leap in AI compute might not come from making better chips at all. It might come from a fundamentally different computing paradigm. If you are making five-year infrastructure bets based on GPU roadmaps alone, you might be optimizing for yesterday's physics.

From the 190K

Three Security Companies, Three Acquisitions, One Week. The AI Security Convergence Is Accelerating Faster Than Anyone Expected.

We scanned 190,000 articles this week. Here is the pattern that only emerges at scale:

In the same week, CrowdStrike and IBM merged their AI agents into an autonomous SOC engine. A major data platform acquired two security startups and launched an entirely new security product designed to counter AI-specific attacks. And five cybersecurity companies announced significant AI capabilities at RSAC 2026. Three moves. Same direction. Same week.

This is not coincidence. This is convergence. The cybersecurity industry is realizing that AI is not just another tool to add to the stack. It is a new attack surface AND a new defense paradigm, simultaneously. Companies that sold data platforms are buying security startups. Companies that sold endpoint protection are building autonomous AI agents. The line between ”data company” and ”security company” is dissolving, because in an AI-native world, securing the data IS securing the business.

🔍 Below the surface: Compliance mentions spiked to 45 GDPR references, 25 CCPA, and 22 HIPAA in a single day's articles. But here is the shift: six months ago, those mentions were about compliance checklists. This week, they are about whether existing frameworks can handle AI at all. When the compliance conversation shifts from ”how to follow the rules” to ”are the rules enough,” new rules are coming.

By The Numbers

• $11 billion — Harvey's valuation after raising $200 million. Legal AI just became one of the most valuable verticals in enterprise software.
• $190 million ARR — Harvey's annual recurring revenue as of January, up from $100 million in August. That is 90% growth in five months.
• 29 minutes — Average eCrime breakout time. Fastest observed: 27 seconds. Your SOC has less time to respond than it takes to read this newsletter.
• $50 million — Normal Computing's raise for thermodynamic AI compute. A bet that the GPU era has an expiration date.
• $125 million — Qualified Health's funding to scale AI for health systems. Healthcare AI funding is accelerating while other sectors pull back.
• 45 GDPR mentions — In a single day's article corpus. CCPA hit 25. HIPAA hit 22. Regulatory density is not fading. It is intensifying.
• $1 billion — The Disney deal that collapsed when Sora was killed. Consumer AI partnerships are only as durable as the product strategy behind them.

Deep Dive: When AI Grows Up (The Enterprise Pivot Nobody Saw Coming)

You know that moment in a DJ set when the crowd stops dancing to the bangers and starts actually listening? The energy shifts. The floor gets quieter, more focused. The DJ switches from crowd-pleasers to the tracks that matter. The set goes from entertainment to experience. That is what happened to AI this week.

The Product Kill That Tells You Everything

When the biggest AI company on the planet kills its most viral consumer product, walks away from a billion-dollar entertainment deal, and pivots hard toward enterprise and robotics, that is not a product decision. That is a market signal. Sora was the crowd-pleaser: millions of downloads, viral clips, mainstream media coverage. But it did not generate enterprise revenue. It did not solve business problems. It created regulatory headaches, content moderation nightmares, and intellectual property liability. The DJ played the banger, the crowd cheered, and then the bills arrived. Enterprise AI is the headliner now. Consumer AI was the warm-up act.

The Revenue Proof

Harvey's numbers demolish the argument that AI cannot generate real enterprise revenue. $190 million ARR. Doubled in five months. In legal services, one of the most conservative, risk-averse industries on the planet. If AI can sell into law firms at that velocity, the ”AI monetization problem” is not an AI problem. It is a targeting problem. The companies struggling to monetize are the ones selling horizontal tools to everyone. The companies thriving are the ones solving specific, expensive problems for specific, paying customers.

The Measurement Shift

Gartner's ”Return on Intelligence” framework is the final piece. When the world's most influential analyst firm tells 4,000 data leaders that traditional ROI is insufficient for measuring AI value, it means the conversation has moved. Efficiency gains are table stakes. The competitive advantage lives in intelligence: better decisions, faster pivots, more accurate predictions. The companies still measuring AI by cost savings are measuring the wrong thing. They are tracking how much cheaper the kitchen got while their competitors are measuring how much better the meal tastes.

What Actually Works

1. Kill your Soras. Audit every AI product and feature for revenue contribution, regulatory risk, and strategic alignment. If it generates attention but not revenue, it is consuming resources your enterprise initiatives need.
2. Go vertical. Horizontal AI platforms compete on price. Vertical AI solutions compete on value. Harvey proved that domain-specific AI in a high-value industry generates revenue faster than any general-purpose alternative.
3. Measure intelligence, not efficiency. Replace ”how much did AI save us?” with ”how much better are our decisions since we deployed AI?” The first question captures cost reduction. The second captures competitive advantage.
4. Invest in the floor, not just the beat. CrowdStrike and IBM did not just build an AI product. They built the operational infrastructure (agentic SOC, crisis simulations, managed services) that makes AI useful in production. The technology is the beat. The operational readiness is the floor. Without the floor, nobody can dance.

The DJ who switches from crowd-pleasers to the real tracks loses some dancers and gains the audience that matters. AI just made that switch. The consumer crowd is thinning. The enterprise floor is filling. And the set is about to get serious.

What's Coming

Europe's Regulatory Frameworks Are Converging, and Most Companies Are Not Ready

A new analysis maps the convergence of four major EU regulatory frameworks: GDPR, the AI Act, the Digital Markets Act, and the Digital Services Act. These are no longer separate compliance exercises. They overlap, reinforce each other, and create compound obligations that most compliance teams have not mapped. If you operate in Europe and use AI, you are subject to all four frameworks simultaneously. The companies that treat them as separate checklists will discover the hard way that a violation under one framework triggers scrutiny under the others.

The White House Just Published a National AI Policy Framework

The White House released a National Policy Framework for AI that establishes federal guidelines for AI development, deployment, and governance. For every company selling AI to the U.S. government or operating in regulated industries, this framework will shape procurement requirements, compliance expectations, and liability standards for the foreseeable future. Read it this week, before your competitors build their proposals around it.

Making AI Agents Visible Is Becoming the First Governance Requirement

A new governance framework argues that the first step to governing AI agents is simply knowing they exist. As enterprises deploy autonomous AI agents across customer service, security operations, and internal workflows, most organizations cannot answer a basic question: ”How many AI agents are operating in our environment, and what are they authorized to do?” That question is about to become a compliance requirement. Start mapping now.

For Your Team

Friday's meeting prompt: ”Gartner just told 4,000 data leaders that traditional ROI is the wrong way to measure AI investments. They introduced 'Return on Intelligence' instead: ambitions, integrity of data foundations, and empowerment of individuals. Look at our AI initiatives: are we measuring what AI saves us, or what AI teaches us? What decision has AI actually improved in the last 90 days? If we cannot name one, what does that tell us about our measurement framework?”

The AI Maturity Audit Framework:

1. Revenue or risk? For every AI initiative, identify whether it generates revenue, reduces cost, or creates risk. If ”creates risk” is the only honest answer, you have a Sora. Kill it or fix it this quarter.
2. Vertical or horizontal? Horizontal AI tools compete on price and get commoditized. Vertical AI tools (legal, healthcare, security) compete on domain expertise and generate loyalty. Evaluate whether your AI investments are domain-specific enough to create lasting value.
3. Speed or intelligence? Are you measuring how fast AI completes tasks, or how much better your decisions are with AI? The first is automation. The second is intelligence. Gartner's framework says the second is what matters.
4. Infrastructure or product? CrowdStrike and IBM built an operational infrastructure (agentic SOC) before they built AI features. Is your AI deployment backed by operational readiness, or are you shipping features without the floor to support them?

Share-worthy stat: Harvey, a legal AI company, nearly doubled its annual recurring revenue from $100 million to $190 million in five months, reaching an $11 billion valuation. That is the fastest revenue growth in enterprise AI this year, and it happened in one of the most conservative industries on the planet.

Go deeper: Track AI enterprise adoption and security signals in real-time →

The Track of the Day

”The companies that succeed are going to be the ones that are relentlessly adapting.”
— Winston Weinberg, CEO of Harvey

Today's set: ”Changes” by David Bowie. Ch-ch-ch-ch-changes. Bowie wrote that song about turning and facing the strange. That is what AI did this week. The biggest company in the space killed its flashiest product. A legal AI startup proved that real revenue beats viral demos. Two security giants merged their AI agents because humans cannot keep up alone. And the world's biggest analyst firm told data leaders their measurement framework is obsolete. Every one of these is a ch-ch-change. The companies that turn and face it, that adapt relentlessly, will lead the next era. The ones still clinging to yesterday's playlist? They are about to hear the crowd walk out.

Your DJ signing off. Kill your Soras, go vertical, measure intelligence over efficiency, and remember: the set gets serious when the crowd-pleasers stop and the real tracks start. Every single time.

Yves Mulkers, your data DJ, mixing 190,000 articles into the tracks that actually matter.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: March 26, 2026 | Curated by Yves Mulkers @ Ins7ghts