Anish Acharya: AI app layer is real — top companies growing 0 to $80M ARR faster than anything seen before

Summary

Anish Acharya, who leads the consumer team at Andreessen Horowitz, makes a direct case that the AI application layer is producing real businesses at a pace that has no historical precedent. The top-decile companies he sees aren't going from $0 to $1M or $2M ARR in 12 months — they're going from $0 to $10M, $0 to $15M, and in some cases $10M to $80M. He frames these numbers as categorically different from anything the venture market has seen before.

The "wrapper" critique that dominated early AI investing discourse is, in his view, a dead argument. Fine-tuning, open-source models, and model proliferation collapsed the binary between training a foundation model from scratch and building something derivative on top of one. Anyone still raising the wrapper concern, he says, hasn't thought it through.

Business model quality

One counterintuitive dynamic Acharya highlights: the fact that tokens cost real money has made early-stage AI companies better businesses. Because founders face genuine cost structures from day one, they're forced to charge immediately and deliver value that justifies the price. Consumer AI in particular has broken the "field of dreams" model — build it, hope for ads — that defined the prior generation. Consumers are paying $50 a month or more at launch, and the question founders should be asking isn't whether they can charge, but what their $1,000-a-month tier looks like. His framing for where consumer spend goes in the future: food, rent, and software.

Model routing

Open source hasn't collapsed pricing so much as it has created a routing layer. Companies are directing queries to different models based on capability and cost — open-source for tasks where the trade-off makes sense, Claude for coding or long-form writing, OpenAI for general-purpose reasoning. That dynamic is part of what's pushing OpenAI up the stack; with its Windsurf acquisition and broader application-layer moves, it's responding to no longer being the sole model provider.

Consumer-to-enterprise bleed

Acharya argues the consumer/enterprise boundary is dissolving. Expense cards are the mechanism — individuals inside large companies are buying software on their own, bringing consumer-grade expectations to work tools. He cites Krea (ticker: KRE), one of A16Z's portfolio companies, as an example: a technically sophisticated product with a distinct aesthetic that an IT buyer might not understand but that is gaining meaningful enterprise adoption because employees inside those companies want the best tools available.

Where incumbents win and lose

His read on the hyperscalers is that new technology reliably helps incumbents extend their lead in existing markets — Microsoft gets better at delivering the word processor, Google gets better at search. The real threat to Google isn't someone beating it on search; it's that there's a new front door to the internet. New categories are where upstarts will dominate, and that's where he expects M&A activity to concentrate.

On monetization, he's skeptical that ads will define AI the way they defined the web. His argument is that ChatGPT is essentially a command-line product, and wherever the interface evolves from there will determine the monetization model. He wouldn't be surprised if it's something other than ads entirely.

The sharpest practical observation he makes: very few people in the industry are actually using the products they're evaluating. Investors who answer "ChatGPT" when asked what they use daily are leaving alpha on the table. For anyone trying to build intuition about where the application layer is going, he treats daily hands-on use as non-negotiable.