It seems every other day another AI startup bursts onto the scene, promising to change the world with a slick new interface powered by the latest and greatest large language model. It’s a gold rush, no doubt. But what happens when the gold is just someone else’s gold, repackaged in a slightly different box? According to one Google executive, the party might be over sooner than many of these fledgling companies think, bringing the question of AI startup viability into sharp focus.
This isn’t just abstract speculation. It’s a direct warning shot fired from the very heart of the generative AI engine room. If you’re building an AI company today, you need to listen very carefully.
The Two Endangered Species of the AI Kingdom
In a recent candid discussion with TechCrunch, Google’s Vice President, Darren Mowry, laid out a stark reality. He identified two specific “archetypes” of AI startups that he believes are facing an existential threat. These aren’t just struggling businesses; they’re models that might be fundamentally flawed in the current market.
– The LLM Wrapper: You’ve seen these everywhere. A startup takes a powerful foundation model like OpenAI’s GPT or Google’s own Gemini, puts a thin “product layer” on top—perhaps a new user interface or a pre-loaded prompt for a specific task—and calls it a new product. The problem? Mowry argues the value they add is minimal. As he put it, “If you’re really just counting on the back-end model to do all the work…the industry doesn’t have a lot of patience for that anymore”. The LLM wrapper limitations are becoming painfully obvious; when the underlying model provider can easily replicate your feature in an afternoon, your entire business is built on borrowed time.
– The AI Aggregator: This model, used by companies like Perplexity and OpenRouter, seems clever on the surface. Instead of betting on one horse, they offer users a choice of different AI models through a single interface. The idea is to be the one-stop-shop for AI. The snag, however, lies in the generative AI economics. These aggregators are essentially middlemen, and middlemen always get squeezed on margins. As the model providers get more competitive on price and features, the aggregator’s slice of the pie gets thinner and thinner.
Think of it like this: an LLM wrapper is like buying a generic ready-meal from the supermarket and serving it on a fancy plate at your “restaurant.” An aggregator is like a restaurant that doesn’t cook anything itself but has delivery menus from every takeaway in town. In both cases, where’s the unique value? Where’s the reason for a customer to stay loyal to you?
Differentiation or Death
The core of Mowry’s warning is a classic business lesson that the tech industry seems to relearn every decade: you need a moat. Without genuine startup differentiation, you’re just a temporary feature, not a sustainable business. In a market this saturated, standing out isn’t just a good idea; it’s a matter of survival.
So, how does a startup build that all-important moat?
– Go deep with vertical specialisation: Instead of trying to be everything to everyone, the smartest startups are becoming indispensable to a select few. They are embedding AI deeply into a specific industry’s workflow. Take Cursor, an AI-first code editor. It’s not just “ChatGPT for developers”; it’s a purpose-built environment designed from the ground up to augment how developers write, debug, and understand code. Another prime example is Harvey AI, which has created a platform specifically for legal case analysis and research. These companies aren’t just wrapping an LLM; they are building a complex, valuable service around it, often using proprietary data and workflows that are incredibly difficult to replicate.
– Build true horizontal differentiation: This is the road less travelled because it’s much harder. It involves creating a fundamentally new capability that works across different industries. This isn’t about offering a menu of models, but perhaps creating a novel method for model interaction, a unique way of chaining processes together, or a user experience so intuitive it becomes the new standard. The bar here is exceptionally high.
We’ve Seen This Film Before, and It Didn’t End Well for the Middlemen
For anyone who’s been around the tech block a few times, this story sounds eerily familiar. Mowry himself draws a direct parallel to the early days of cloud computing, a comparison I find spot on. A decade ago, a flurry of startups emerged as “cloud resellers.” They bought computing resources in bulk from Amazon Web Services (AWS) and sold them on, perhaps with a simple management dashboard attached.
For a while, it worked. But what happened next? AWS, seeing the opportunity, simply built its own, more sophisticated, enterprise-grade management tools and offered them directly to customers, often at a lower cost. The resellers were wiped out almost overnight because their value proposition had evaporated.
The AI aggregators of today are the cloud resellers of yesteryear. They face the exact same margin pressure and the same risk of being made redundant by the very platforms they rely on. The lesson from history, as cited in the TechCrunch article, is that you cannot build a long-term, defensible business by simply being a tollbooth on someone else’s highway.
So, Where’s the Smart Money Headed?
Despite the dire warnings, the future for AI startups isn’t all gloom. The opportunity is simply shifting. Instead of building flimsy wrappers, the real growth is in creating the tools that enable others. According to industry analysis mentioned by Mowry, developer platforms had a record-breaking year for investment. This is the classic “selling shovels in a gold rush” strategy. Building the infrastructure, the developer tools, and the specialised platforms is where deep, sustainable value lies.
We’re also seeing potential in consumer-facing tools, but only those that solve a tangible problem in a unique way. It’s not enough to be a slightly better chatbot. A successful consumer AI tool needs a clever hook, a delightful user experience, or a function that is simply impossible without AI.
The era of easy wins in AI is closing. The market is maturing, and the low-hanging fruit has been picked. AI startup viability now depends on building what Mowry calls “deep, wide moats.” This isn’t about having the slickest demo or the catchiest name. It’s about creating genuine, defensible value through deep expertise or true innovation.
The question for every founder and investor in this space should be: are you building a castle with a moat, or are you just putting a fancy sign on someone else’s tent? Which AI startups do you believe have built a moat strong enough to withstand the coming storm?
