AI expands the surface area of venture capital

Much of today’s AI discourse centers on a single question: won’t OpenAI and Anthropic eat all the app layer opportunities?

It’s an understandable fear given the relentless pace at which the big labs are shipping. (It’s also a narrative encouraged by the labs themselves as they prep for IPO).

Every platform shift creates questions about where value will accrue. In the early days of the internet, it was widely believed the telcos and browsers would capture most of the value. Yet the real winners were the applications built on top. Similarly, in the SaaS revolution it was assumed that the cloud infrastructure providers were the ultimate winners. Yet AWS enabled the creation of Salesforce, Shopify, Datadog or thousands of other valuable software businesses.

Is this time different? Our view is a resolute No. While the big labs will undoubtedly be enormous winners, their success isn’t even the most important consequence of the AI revolution. Rather, it’s that AI dramatically expands the number of venture-scale opportunities available to founders.

Every sector is suddenly investable (and bigger)

Historically, SaaS was limited to automating structured, repetitive and digital tasks. It didn’t penetrate the services economy - a segment six times larger than traditional software - because services roles were messier and inherently human. Humans were required to answer phones, read documents, send emails, interpret language, exercise judgment and handle exceptions. It was simply too complex to codify in if-this-then-that software.

This view of the world is now completely outdated. With the rise of LLMs, along with voice and vision models, software is reaching (and in many cases exceeding) human parity in reading, writing, reasoning, communicating and navigating unstructured information. It is more appropriate to think of software as labor, rather than the tool used by labor. The implications of this are profound. Entire sectors of the economy that were previously 100% services are now transitioning into the domain of software.

This opportunity is much larger and much more interesting than SaaS 1.0. Not only can businesses be built in any sector, they can also be substantially larger by virtue of being able to tap into labor budgets. A company might spend $10m on recruiters and $500k on recruiting software. $5m on claims adjusters and $250k on claims software. $3m on salaries for debt collectors and $100k on collections software. As software moves closer to performing the work itself, the addressable market expands dramatically.

Why the Labs Won't Capture Everything

So if every market is in play and bigger than ever, why won’t the big labs pursue these opportunities themselves?

First, they’re focused on a bigger prize. Despite some loud proclamations about app-layer initiatives, the labs have mostly retreated from those pursuits. Instead, they’ve redoubled their focus on winning the arms race on horizontal capabilities that matter across every use case (like coding, reasoning, memory and agentic behavior). Improving the intelligence layer itself is likely a multi $T opportunity and has taken priority.

Secondly, beyond market size prioritization, competing at the app layer triggers an innovator’s dilemma. Anthropic’s historic revenue ramp has come off the back of enterprise API usage, as more application-layer businesses infuse AI into their solutions. If Anthropic builds applications themselves, they compete with many of the companies who would otherwise be their customers, and jeopardize their revenue base.

Yet it would be inaccurate to claim that the labs are pure infrastructure plays; in some industries (like legal and finance), they’ve been building primitives for customers to implement themselves. At first glance this might look like a threat to application layer companies. In reality, it’s far from it. Plugins and primitives might get you 80% towards solving a vertical-specific problem. However the last mile - which often requires domain expertise, judgment and deep customer empathy - is what really matters to fully address the customer’s pain point. Solving for this last mile usually requires building industry-specific workflows, integrations, managing compliance, exception handling and providing customer support. These aren’t the most technically interesting parts of any business offering, but they’re critical when selling to enterprises.

Thirdly, AI is enabling entirely new business models that are orthogonal to the priorities of the labs. Consider the rise of the AI-native services firm: law firms, recruiting agencies, insurance brokerages, and other service businesses that are being rebuilt from the ground up using AI. This hybrid business type requires building a services firm alongside a technology company. Consumed by the race towards AGI, it would be hugely distracting for the big labs to also build thousands of vertically-focused services firms in-house just to compete in those spaces.

Where value accrues

As you may gather, our perspective is that for most application layer startups, the direct threat from OpenAI/Anthropic is overstated. However, as AI capabilities become widely available, we do believe the main threat comes from other startups themselves. (In some cases, there is also a threat from incumbents who are willing to burn the boats on their existing business and fully embrace AI).

As a result, we ask ourselves structural questions around defensibility at the time of making an investment. Ironically, you have a higher chance of building defensibility by building hard things beyond the AI (i.e. AI is table stakes). That might mean building towards a system of record opportunity, accumulating proprietary data that makes your product offering better as it scales, curating offline supply into a marketplace, embedding deeply into customer workflows or securing unique distribution channels. These advantages are difficult for other startups to replicate because they are AI-proof.

Commercial real estate brokers use our portfolio company Terrakotta as their core dialer software when calling prospects. Broker conversations on the platform surface proprietary data (such as cap rates in a specific geography, for a specific property type) that can be anonymized and sold to other stakeholders, like REITs and developers. The business has a clear network data effect as more brokers sign up for the core dialer product

The Opportunity Ahead

SaaS 1.0 created several hugely valuable companies but was ultimately constrained to a relatively small slice of the economy. AI expands the reach of software into industries, workflows, and labor markets that were previously inaccessible. Markets that were once too operational, too services-heavy, or too dependent on human labor are suddenly becoming fertile ground for venture-backed startups.

This thesis shapes how we invest. We're less interested in companies whose primary advantage is access to a model and more interested in founders using AI to attack large labor markets, build hard things beyond the AI, and create proprietary assets that compound as the underlying models improve. In our experience, these opportunities often look unusual, operationally complex, and difficult to categorize - exactly the type of businesses that many investors initially overlook.

If the SaaS era was about digitizing work, the AI era is about transforming how work gets done. We believe that shift will create an extraordinary number of category-defining companies—and that we're still in the earliest innings of what may become the largest expansion of venture-scale opportunities in decades.