Darren Mowry oversees Google’s global startup organization across Cloud, DeepMind, and Alphabet. He spends his days watching which AI startups get traction and which ones stall out. In a recent appearance on TechCrunch’s Equity podcast, he singled out two business models that he says are running on borrowed time.
“If you’re really just counting on the back-end model to do all the work and you’re almost white-labeling that model, the industry doesn’t have a lot of patience for that anymore,” Mowry told TechCrunch. Startups fitting this description, he says, have their “check engine light” on.
The two categories he’s flagging: LLM wrappers and AI aggregators.
What’s an LLM Wrapper?
LLM wrappers are startups that build an interface on top of someone else’s model - Claude, GPT, Gemini - and sell access to that interface. They add a UX layer, maybe some prompt engineering, possibly some workflow automation. But the actual intelligence comes from the underlying foundation model.
The problem is existential: as the models get smarter, the value of the wrapper shrinks. Features that differentiated a product six months ago become standard capabilities in the next model release. ChatGPT itself keeps adding features that make third-party wrappers redundant. Claude added Projects. Gemini added Deep Research.
Wrapping “very thin intellectual property around Gemini or GPT-5” signals you’re not differentiating yourself, Mowry said. You’re renting someone else’s innovation and calling it your product.
What’s an AI Aggregator?
AI aggregators are a subset of wrappers. They connect users to multiple models through a single interface - query GPT-5 for one task, Claude for another, Gemini for a third. They typically add routing logic, monitoring, governance, or evaluation tooling. Examples include AI search startup Perplexity or developer platform OpenRouter.
Mowry’s message is clear: “Stay out of the aggregator business.”
The business model made sense when navigating multiple AI APIs was complicated and enterprise buyers wanted a single vendor. But cloud giants have noticed. Microsoft’s Azure AI and Amazon’s Bedrock now bundle multi-model access as standard offerings. What was once a startup’s unique value proposition is becoming a platform feature.
Aggregators face the same squeeze as wrappers: their margins compress while switching costs drop. If you’re just a middleman with no proprietary technology, bigger players will eventually absorb your function.
We’ve Seen This Before
Mowry draws a parallel to the early days of cloud computing. When AWS launched, a crop of startups sprang up to resell infrastructure - essentially wrapping Amazon’s services in a slightly different package and marking them up.
Most of them died. When Amazon built its own enterprise tools and customers learned to manage cloud services directly, the middlemen got squeezed out. The only survivors were those that added real services: security, migration, DevOps consulting. Not reselling, but building.
AI is following the same pattern, just faster. The cycle from “hot new category” to “commoditized feature” is compressing. What took cloud computing years is taking AI months.
What Actually Survives
For a startup to “progress and grow,” Mowry says, it needs “deep, wide moats that are either horizontally differentiated or something really specific to a vertical market.”
He points to Cursor, the AI coding assistant, and Harvey AI, the legal AI platform. Both are technically wrappers - they use foundation models rather than training their own. But they’ve built substantial additional value. Cursor integrates deeply with development workflows. Harvey has accumulated legal-specific training data and domain expertise that generic models lack.
The difference between a dying wrapper and a surviving one comes down to what happens beyond the API call. If you’re just making it easier to talk to GPT, you’re vulnerable. If you’re building proprietary datasets, specialized workflows, or deep vertical expertise, you have something to defend.
The Numbers Tell the Story
The shakeout is already visible in the data. SimpleClosure’s 2025 “State of Startup Shutdowns” report documented a 2.5x year-over-year increase in Series A shutdowns, with AI wrappers disproportionately represented.
Jasper AI, once valued at $1.5 billion as a content writing wrapper, cut its internal valuation by 20% amid slowing growth. ChatGPT’s launch created a formidable low-cost competitor practically overnight, and users started asking why they should pay premium prices for what felt like the same underlying technology.
The pattern is consistent across the category. Easy to start, hard to sustain. Early traction from differentiated UX, followed by margin compression as the underlying model gets better and alternatives proliferate.
Why This Matters
If you’re using AI tools, the wrapper extinction wave might actually be good news. Features currently locked behind paid subscriptions will migrate to the base models. Competition will drive prices down. The survivors will be the ones that genuinely add value, not the ones that just reskin existing capabilities.
If you’re building an AI startup, the lesson is stark. Wrapping is not a strategy. It’s a starting point that you need to evolve beyond quickly, before the next model update makes your differentiation disappear.
And if you’re investing in AI startups, Mowry’s framework offers a simple test: what does this company do that survives the next model upgrade? If the answer is “nothing,” the check engine light is on.