Discover how AI is transforming software, labor-intensive industries, and creating defensible moats through proprietary data. Learn from leading venture inve...
AI Opportunity Beyond Models: Build Billion-Dollar Companies in 2025
Key Takeaways
- AI isn't just about models – The real opportunity lies in building AI-native applications that solve real business problems and generate measurable revenue
- Labor-market disruption is massive – With the U.S. labor market worth $13 trillion versus a $300 billion software market, AI is unlocking unprecedented market expansion opportunities
- Proprietary data creates defensible moats – Companies combining unique data assets with AI capabilities build sustainable competitive advantages that large labs cannot easily replicate
- Vertical software becomes sticky – AI-native solutions that become systems of record (owning end-to-end workflows) are significantly harder to displace than single-feature tools
- Speed and execution matter more than ever – The ability to rapidly deploy AI solutions and iterate based on real-world outcomes determines winners in this market cycle
The Four Major Product Cycles and Why AI Is Different
Since the late 1970s, technology has evolved through distinct product cycles: the personal computer era, the internet revolution, cloud computing, and mobile smartphones. Each cycle created trillions in market value by combining infrastructure innovation with application layer companies that solved real user problems.
The AI era, which began approximately two years ago, represents something unique. Unlike previous cycles where adoption took years to mature, AI adoption is accelerating at unprecedented speed. This acceleration exists because eight billion humans on Earth now have smartphones with cloud connectivity – the foundational infrastructure is already in place. When you combine this global reach with breakthrough AI capabilities, the addressable market for new applications becomes exponentially larger than in any previous technology cycle.
The statistics underscore this reality. Just two years ago, we had basic text and image capabilities. Today, we've achieved real-time audio interactions, video understanding, and reasoning capabilities that would have seemed impossible so recently. What's remarkable isn't just the technical progress – it's that this innovation is translating directly into business value. The majority of net new software revenue being generated today comes from AI-native applications, both at the infrastructure and application layers. This isn't theoretical; it's happening right now, and companies are growing from zero to $100 million in annual revenue in just one to two years, something virtually never seen in software before.
Traditional Software Going AI-Native: The Greenfield Opportunity
One of the most compelling investment themes in AI is the reinvention of existing software categories. When cloud computing emerged 15-20 years ago, every software category was disrupted. Companies that had dominated on-premise markets – selling shrink-wrapped software with high-profit margins – found themselves unable to transition to subscription-based cloud models. This created massive opportunities for cloud-native entrants.
We're witnessing the exact same pattern with AI. Every existing software market is now open for new entrants. Companies building AI-native versions of existing categories are winning because they're fundamentally rethinking the user experience, not just adding AI as an afterthought. Take design software as an example. Photoshop remains a fantastic business, but young designers starting their careers increasingly prefer Krea – an AI-native design tool with AI primitives built directly into the product's foundation. These AI-native designers aren't trying to convince Photoshop users to switch; they're the first cohort adopting design tools, and they're choosing AI-native options from day one.
This pattern is emerging across virtually every software category. From accounting (Rullet automating bookkeeping with 50+ AI features) to ERP systems, new entrants are reinventing categories specifically for the AI era. The key distinction between winners and losers in this space isn't whether they're using AI – it's whether they're building defensible moats beyond just having AI features.
Consider a company called Mercury, which built a neobank specifically for startups. Mercury didn't aggressively steal customers from Silicon Valley Bank until the weekend SVB collapsed. This exemplifies what we call a "greenfield" opportunity – targeting new customers or companies at an inflection point that lack existing solutions. The alternative, "brownfield" plays, involve convincing existing Mailchimp or NetSuite users to switch. Brownfield opportunities are dramatically harder because incumbents are sticky, even if their products aren't ideal. Mercury's success came from building a product specifically designed for a new cohort (startups) at exactly the moment they needed banking services and were evaluating options.
This greenfield versus brownfield distinction matters enormously. The most successful AI-native companies aren't trying to convince legacy software users to migrate; they're building for entirely new use cases or for users who are making their first software purchasing decision in that category.
Labor Market Transformation: The $13 Trillion Opportunity
While the global software market is worth approximately $300 billion, the U.S. labor market alone represents $13 trillion in annual economic activity. This massive gap represents the fundamental opportunity for AI: automating tasks previously performed by humans at scale. Unlike previous technology cycles where software merely improved efficiency within existing labor structures, AI is enabling entirely new service delivery models.
Consider debt collection, traditionally a labor-intensive industry. A company called Salient has deployed AI voice agents to handle consumer servicing entirely through phone, text, and email – eliminating the need for large human call center teams. But the real game-changer isn't cost savings (though those exist). Salient increased collection rates by 50%. This is fundamentally different from cost reduction; it's revenue generation. Consumer lenders can now increase their loan books without proportionally expanding human teams. In fact, many are shrinking their call center headcount while actually collecting more money.
The economics are revolutionary. Salient's AI agents speak multiple languages with perfect fluency, call consumers at optimal times, navigate complex compliance requirements across all 50 states and various county regulations, and do it all faster than humans while improving outcomes. No human can possibly keep track of every regulation across every jurisdiction – it's an impossible task. Yet Salient's AI system continuously ingests every new law and proposed statute from all governmental levels, making it nearly impossible to compete with.
Another powerful example is Eve, a legal AI platform serving plaintiff attorneys. These lawyers operate on contingency – they only get paid if they win. Eve makes them approximately five times more productive by owning the entire workflow from initial client intake (where voice agents collect evidence and sift through mountains of medical records and employment documents) through case evaluation, pre-litigation drafting, and courtroom support. Eve's AI drafts medical chronologies and demand letters in real-time, analyzing case characteristics to determine which cases are worth pursuing and which can be handled more cost-effectively.
The impact on attorney economics is dramatic. Previously, attorneys only took cases if they expected minimum recoveries of $50,000 (the economics had to work). With Eve's efficiency gains, they can now profitably take cases worth $5,000. The market expands dramatically. Eve isn't eliminating attorney jobs; it's expanding the number of cases attorneys can handle, enabling them to serve a much larger client population that was previously uneconomical to serve.
What makes Eve's moat particularly strong isn't just the AI voice agent – it's the proprietary data. Every case that flows through Eve's platform generates outcome data: which cases won, which settled, the reasons why, what tactics worked. This data isn't public; large AI labs like OpenAI can't train models against it. Eve uses this proprietary outcome data to continuously improve its intake process, helping attorneys make smarter case decisions. It becomes a reinforcing loop: better intake decisions lead to better outcomes, better outcomes generate more valuable training data, which further improves intake. Over time, using Eve becomes nearly mandatory for plaintiff attorneys. They're not using Eve because it's cheaper; they're using it because they can't afford not to in terms of competitive advantage.
This pattern – AI enabling entirely new service delivery models in massive labor-intensive industries – represents the single biggest market opportunity in AI. It's not about doing the same work cheaper; it's about expanding the addressable market by making previously uneconomical services economical at scale.
Building Defensible Moats: Why Systems of Record Win
The temptation when discussing AI is to assume that having AI capabilities is itself a defensible moat. It isn't. The challenge facing any AI-powered startup is that large labs like OpenAI are continuously releasing more capable models, and it's entirely possible they could replicate capabilities you've built. So what actually creates defensibility in the AI era?
The answer lies in three core concepts: owning the end-to-end workflow, becoming the system of record, and building proprietary data advantages that large labs cannot access.
Owning End-to-End Workflows
Eve exemplifies this principle. Rather than building just a voice agent or a document summarization tool, Eve deliberately architected its product to own the entire plaintiff attorney workflow – from intake through litigation outcome. This comprehensive coverage creates stickiness that isolated features cannot. An attorney living in Eve's product all day, managing every aspect of their cases through the platform, experiences tremendous switching costs. Building a competitor to replace Eve wouldn't mean building a better voice agent; it would mean rebuilding the entire operational system attorneys depend on.
Similarly, Toast – a vertical operating system for restaurants – demonstrates the power of end-to-end ownership. Toast helps restaurateurs manage every aspect of their business: integrating with DoorDash for delivery, managing staff payroll, handling inventory, processing payments. Toast initially seemed like a risky business (restaurants have high failure rates and limited software budgets), but by adding financial services like lending and payment processing, Toast became deeply embedded in restaurant operations. Traditional point-of-sale or payment processing companies find it nearly impossible to match Toast's comprehensive offering because they're specialized players competing against a vertical operating system.
Becoming Sticky Through Network Effects and Workflow Integration
The most defensible AI companies aren't single-feature solutions; they're platforms that users visit frequently because those platforms are essential to their work. Mercury became sticky not by being marginally better at banking, but by being built specifically for startups' actual operational needs – it integrated with their financial planning, payroll, and fundraising workflows. Similarly, ServiceTitan (field service management for contractors), Mindbody (for fitness and wellness), and Toast (for restaurants) all achieved massive valuations by becoming essential operational infrastructure for their respective verticals.
The data supports this: companies that build rich ecosystems around their core capabilities and own multiple workflow stages significantly outperform those offering isolated features. Customer retention is notably high for the best AI companies because they've become systems of record rather than bolt-on tools.
Proprietary Data: The Defensibility Layer Large Labs Cannot Match
Perhaps the most fascinating defensibility mechanism in AI is proprietary data access. While large language models trained on internet-scale data are powerful, they lack the specialized, domain-specific data that proprietary AI applications can access. This gap creates extraordinary opportunities for companies that can control or aggreg unique datasets.
OpenEvidence illustrates this perfectly. Two-thirds of American doctors use it weekly. It functions like ChatGPT but with exclusive licenses to critical medical journal data. When a doctor queries OpenEvidence about a treatment, they receive evidence-based answers grounded in millions of medical journal articles. In contrast, ChatGPT provides only moderately useful medical information because it lacks specialized medical data access. OpenEvidence's exclusive licensing creates a product that's substantially better than general-purpose AI, and it's protected by the very data licenses that competitors cannot easily replicate.
Similarly, vLex – a 26-year-old company – made a strategic decision to acquire every legal record in Spain. For years, selling raw legal documents offered limited growth potential. But when vLex integrated AI into their platform, the business quintupled in revenue. Suddenly, they could deliver finished, precise legal memos incorporating specific Spanish legal precedents – something invaluable to law firms. Generic AI cannot provide this level of Spanish legal specificity because it doesn't have access to the proprietary, aggregated dataset vLex controls.
This pattern repeats across industries. FlightAware aggregates ADSB transponder data (publicly available but scattered) into a comprehensive, curated dataset about global air traffic. PitchBook aggregates funding data. LexisNexis aggregates legal records. Bloomberg provides exotic financial data. Ancestry.com became valuable by digitizing genealogical records that existed but were scattered across various sources. The common theme: unique data access, when combined with AI processing, creates defensible competitive advantages.
The "Finished Product" Principle
A critical insight is that the value often lies not in the raw data itself – which might be freely available or inexpensive – but in transforming it into a "finished product." Instead of selling a raw subscription to PitchBook data, imagine using AI to generate a detailed memo comparing Company E's funding rounds to every other legal tech company's funding history since 1992. While PitchBook might charge a modest subscription for raw data, a finished, AI-generated analysis might command 10x higher pricing because it delivers specific, actionable value.
This principle creates enormous opportunities for "de novo walled gardens" – businesses built by digitizing, normalizing, and wrapping existing (often undervalued) data with AI. An entrepreneur who identified old blender manuals from the 1980s and 90s – previously worth almost nothing, available cheaply on eBay – could digitize, organize, and make them accessible through an intelligent interface. That previously worthless data becomes valuable in finished form.
The "why now" for these opportunities stems from technological progress. Just as Uber's emergence depended on smartphones and GPS reaching ubiquity, current AI capabilities allow companies to turn previously intractable data sources into immensely valuable businesses. The ability to process data intelligently at scale is now the limiting factor; data is often freely or cheaply available, but intelligently processing it was previously impossible. AI changes this equation entirely.
Vertical Software and the Case for Specificity
One of the most misunderstood aspects of AI's impact on software is the question of whether specialized vertical software can compete against horizontal platforms. The historical answer is clear: specialized vertical software companies can become massive businesses.
Vertical software achieves this by becoming the operational foundation for specific industries. When you're deeply embedded in an industry's operations – when users depend on your software throughout their workday and it integrates with their financial, payroll, and compliance workflows – switching becomes prohibitively expensive. The software becomes a system of record rather than a tool.
Toast's trajectory illustrates this principle. When Toast initially pitched to investors, skeptics pointed out that restaurants have high failure rates and spend little on software. This seemed like a fundamental market limitation. But Toast didn't try to sell enterprise software to restaurants; they built a vertical operating system. They integrated with delivery platforms, handled payroll, managed inventory, and eventually added financial services. Once embedded this deeply into restaurant operations, Toast became impossible to displace. Trying to switch restaurants from Toast to a traditional POS system would mean abandoning integrated financial services, delivery coordination, and payroll management.
Similarly, Workday disrupted human resources and financial management for medium and large enterprises not by being marginally better than SAP or Oracle, but by being specifically built for cloud-native, subscription-based operations. NetSuite achieved massive scale by providing a comprehensive, cloud-based ERP specifically designed for small and mid-market businesses, a segment larger legacy ERP vendors had largely abandoned.
The AI era is accelerating the emergence of specialized vertical software. Companies that were previously held back by technical limitations can now use AI to handle complexity that would have required enormous engineering teams. A small team can now build a comprehensive vertical operating system for a specific industry in ways that were previously impossible.
Why Aggregators Win in Consumer AI
An interesting dynamic in consumer AI is why aggregators of multiple AI models outperform single-model-dependent applications. The analogy is airlines: it's far more valuable to search for flights on Kayak (where you see every airline's inventory) than to check individual airline websites. Similarly, in coding, design, and content creation, users prefer tools that aggregate multiple AI models because each model excels at different tasks.
Claude is exceptional at reasoning and long-form writing. GPT-4 excels at certain creative and coding tasks. Gemini has strengths in multimodal understanding. These models aren't perfect substitutes; they're complementary. A designer wants access to multiple image generation models because one might excel at photorealism while another excels at illustration. A developer wants access to multiple coding assistants because they have different specializations.
Large AI labs and tech companies – by definition – can only use their first-party models. OpenAI's consumer applications naturally default to their own models. Google's offerings feature primarily their own models. This architectural limitation gives aggregators a structural advantage: they can offer users a genuinely superior experience by providing access to multiple model providers' capabilities.
This is why platforms that aggregate models are winning in consumer AI, and it's a durable structural advantage that's difficult for single-model providers to overcome.
The Investment Process: Finding, Picking, and Winning Deals
Beyond identifying market opportunities, the practical challenge for investors is finding exceptional companies before others do. The venture capital industry has evolved to recognize that sourcing deals – finding companies before they become obvious – requires becoming a genuine expert in specific categories.
This requires several components. First, becoming the recognized authority in a space involves publishing insightful research and analysis. When a venture firm publishes widely-read pieces like "Death of a Salesforce: Why AI Will Transform Sales" or comprehensive analyses of the top 50 enterprise AI applications, entrepreneurs building in those categories naturally want to talk to the authors. This creates inbound deal flow that's dramatically better than outbound prospecting.
Second, assembling a team of genuine category experts – people who are deeply knowledgeable about specific verticals and can provide real strategic value – attracts the best founders. A founder building AI for legal tech wants to work with investors who have deep legal tech expertise, who understand plaintiff attorney economics, who know the regulatory landscape. Generic venture investors are much less attractive.
Third, being willing to move quickly when exceptional opportunities emerge – what we call "process interrupt" – separates firms that win major deals from those that miss them. When a truly exceptional company appears, every top venture firm wants to invest. The firms that win are those willing to jump on opportunities immediately, not those that require extensive committee votes and consensus-building.
Fourth, there's a critical distinction between "adverse selection" and "positive selection" in deal sourcing. An inexpensive deal that's been on the market for six months is probably not a good one; that's adverse selection. Positive selection means meeting with the best companies – the ones that every top firm wants to invest in. These deals are highly competitive, but that competitiveness is exactly the signal you want. If a deal isn't competitive, it's probably not exceptional.
Finally, recognizing that deal-winning at the highest level requires leverage and seniority matters. When competing to win a superpower deal against firms like Sequoia, Accel, or Greylock, having senior partners with track records of building massive companies, who can offer board gravitas and strategic guidance, makes a material difference. The CEO of an exceptionally promising startup wants to work with legendary founders and operators, not just talented young investors. This is why senior partners who remain actively engaged (rather than semi-retired) are invaluable for winning the most competitive deals.
The Role of Culture in AI Adoption
One final observation that distinguishes successful AI companies from those struggling to gain traction is organizational culture. The companies that benefit most from AI are those that have genuinely internalized the question: "Can we use AI to improve this function?"
This isn't a technology question; it's a cultural imperative. Some of the most successful AI companies are those where leadership is constantly asking whether AI can improve existing processes, not just in product but in internal operations. This culture of continuous AI integration – asking how AI can improve sales, finance, operations, recruitment, and every other function – is becoming a key differentiator.
Companies that resist this mindset ("We're going to hire salespeople that play golf with lawyers and never use AI") will find themselves outcompeted by organizations that have embraced AI integration throughout their operations. The winners aren't those that treat AI as a special technology reserved for certain functions; they're those that see AI as a fundamental tool that improves virtually every operational aspect.
Conclusion
The AI opportunity extends far beyond developing better large language models. While foundational model companies like OpenAI will remain valuable, the most significant wealth creation will occur in application layer companies that solve real business problems with AI. The companies that achieve the highest valuations will be those that build defensible moats through proprietary data, own end-to-end workflows in their categories, and become systems of record that users cannot easily abandon.
The opportunity is genuinely massive: a $13 trillion labor market being disrupted by AI-native solutions, every software category being reinvented by newcomers, and entirely new service delivery models becoming economically viable for the first time. The next generation of billion-dollar companies are being built right now by founders who understand that the magic trick of AI isn't the AI itself – it's the defensible, customer-centric business models they build around it. For investors and entrepreneurs alike, the moment to act is now. The window for capturing this opportunity won't remain open indefinitely, and the teams that move quickly while thinking deeply about defensibility and business model design will define the AI era's winners.
Original source: The AI Opportunity that goes beyond Models
powered by osmu.app