Google’s Vice President of the Global Startup Organisation, which includes Cloud, DeepMind, and Alphabet, Darren Mowry, issued a warning this month, February, that two particular categories of AI companies are likely to fail when the “gold rush” phase of generative AI comes to a close.
The explosion in generative AI created a startup every minute. The two once-hot business models, LLM wrappers and AI aggregators, are beginning to resemble cautionary tales as the dust settles.
The two kinds of startups that might likely not make it:
LLM Wrappers: These businesses overlay third-party models, such as GPT-4 or Gemini, with a simple user interface. “White-labelling” has been unpopular in the industry because these items don’t provide much unique value.
AI aggregators: These intermediaries offer a single interface or API to route queries between several distinct LLMs. As model suppliers develop their own business features, governance layers, and intelligent routing tools, these companies are subject to obsolescence and pressure on their margins.
These startups with these hooks have their “check engine light” on, according to Darren Mowry, head of Google’s worldwide startup business, which includes Cloud, DeepMind, and Alphabet.
In essence, LLM wrappers are startups that apply a product or user experience layer to big language models that already exist, such as Claude, GPT, or Gemini, in order to address a particular need. A firm that uses AI to assist students in their studies would be one example.
On this week’s episode of Equity, Mowry stated, “The industry doesn’t have a lot of patience for that anymore if you’re really just counting on the back-end model to do all the work and you’re almost white-labelling that model.”
Mowry stated that very thin intellectual property wrapped around Gemini or GPT-5″ indicates that you’re not setting yourself apart.
For a firm to “progress and grow”, he stated that users have to have deep, wide moats that are either horizontally differentiated or something really specific to a vertical market.” The GPT-powered coding assistant Cursor and the legal AI assistant Harvey AI are two instances of the deep-moat LLM wrapper type.
Put differently, startups can no longer hope to open a ChatGPT store in the middle of 2024 and acquire traction with their offering by just slapping a user interface (UI) on top of a GPT. Currently, creating sustained product value is the challenge.
AI aggregators, which are firms that combine several LLMs into a single interface or API layer to route queries between models and provide customers with access to numerous models, are a subset of wrappers. These businesses usually offer an orchestration layer with tools for governance, monitoring, and evaluation. Consider the developer platform OpenRouter or the AI search firm Perplexity, which offer access to several AI models through a single API.
“Stay out of the aggregator business,” Mowry advises prospective entrepreneurs, despite the fact that many of these platforms have found traction.
He claims that users want “some intellectual property built in” to make sure they’re directed to the appropriate model at the appropriate moment based on their needs, not because of access or computation limitations behind the scenes, which is why aggregators aren’t seeing much development or advancement these days.
Mowry has been involved in the cloud industry for many years. He began his career with AWS and Microsoft before establishing himself at Google Cloud, and he has witnessed the evolution of this. He claimed that the current state of affairs is similar to the early years of cloud computing, when Amazon’s cloud business began to take off in the late 2000s and early 2010s.
A plethora of firms emerged at that time to resell AWS infrastructure, positioning themselves as simpler entry points that offered support, billing consolidation, and tooling. However, the majority of those startups were forced out as Amazon developed its own enterprise tools and users discovered how to directly administer cloud services. Only those who added genuine services, such as security, migration, or DevOps advice, managed to survive.
As model providers move into enterprise features themselves, potentially displacing intermediaries, AI aggregators are currently under comparable margin pressure.
Mowry is optimistic about mood coding and developer platforms, which saw a record-breaking year in 2025 with significant investment and consumer traction from firms like Replit, Lovable, and Cursor, all of which, according to Mowry, are Google Cloud clients.
Mowry also predicts growth in direct-to-consumer (DTC) tech that puts powerful AI tools directly in customers’ hands. He highlighted the chance for students studying cinema and television to use Google’s AI video generator, Veo, to make stories come to life.
Apart from artificial intelligence, Mowry believes that biotech and climate tech are experiencing a boom due to the influx of venture capital into these fields and the “incredible amounts of data” that companies can access to provide genuine value “in ways we would never have been able to before.”
The early days of cloud computing are comparable to this. After Amazon released its own enterprise-grade tools, startups that resold AWS infrastructure were swiftly forced out. New entrepreneurs should concentrate on fields like biotech, climate tech, and developer platforms that have deep vertical-specific value or proprietary data.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
