Capital in the AI industry is flowing from abstract large language models toward managing users' real-world tasks. Companies are monetizing not text generation, but autonomous action — from ordering pizza to running an entire business end-to-end.

The "Free Hook" Strategy and the Hardware Bet

OpenAI suddenly lifted limits on text queries for free ChatGPT users, adding a "Think" feature (analogous to extended reasoning) for complex requests. In parallel, the company is prepping its own AI device — a smart speaker priced between $300 and $400.

In my view, this is a classic monetization pivot. When basic text becomes a free commodity, you have to make money on hardware screens and agentic subscriptions. The $400 speaker is OpenAI's attempt to build its own ecosystem, one that doesn't depend on Apple's and Google's app stores. Free ChatGPT here is simply a tool for capturing user data to train future hardware products.

Google Maps Becomes a Super-Agent

Google has added agentic features to Google Maps: the service can now autonomously order food and book hotels on the user's behalf. This is a transformation of a navigation product into a full-fledged business assistant that monetizes transactions, not ad clicks on maps.

This move shows how giants with massive user bases outflank pure-play AI startups. Google simply embeds agents where the user already is, saving millions in customer acquisition costs.

Infrastructure Contracts and Company Automation

Startup Mirendil signed a contract with Google Cloud worth over $100 million to scale compute capacity for "self-improving AI" workloads. Meanwhile, startup Naïve raised $28.5 million to automate the routine processes of setting up and running a business. Naïve positions itself as a tool capable of autonomously handling companies' operational work.

Both deals illustrate opposite ends of the B2B market. Mirendil is attracting massive capital for fundamental infrastructure (compute power), while Naïve is monetizing the application layer — turning "silent" AI into agents that replace corporate management. Investors clearly see value in replacing entry-level human labor with algorithms.

Open Weights as a Market Capture Tool

NVIDIA, together with some two hundred companies, signed an open letter titled "Open Weights and American AI Leadership," promoting open World Models for physical AI. At the same time, startup Liquid AI released LFM2.5-2.6B, an open-weights model built specifically for agentic tasks. The key feature — the model runs on the edge, including Raspberry Pi-class microcomputers, with no need for cloud GPUs.

This is an economically calculated strategy. By giving away compact models for free, corporations commoditize the bottom layer of the AI market, forcing competitors to spend on cloud compute while NVIDIA itself profits from selling physical hardware for the next technological paradigm.

AI Price Doesn't Correlate with Benchmarks

Alibaba released the Qwen 3.8-Max model, claiming it is second only to Claude Opus 5. However, independent tests showed that the actual cost of performing agentic tasks with the new model can differ significantly from its advertised efficiency.

For the market, this means the end of the era of marketing through "benchmark supremacy." Enterprise clients now buy AI based not on leaderboard rankings, but on total cost of ownership. If a cheaper model handles an agentic task 5% worse but costs ten times less, businesses will choose it.

Bottom Line

The AI market is stratifying rapidly. The mass B2C segment is being devalued and turned into free bait. All the real margins and venture capital are flowing into three areas: hardware devices with screens, B2B agents capable of replacing employees, and the invisible infrastructure for training systems. In the coming quarters, expect a wave of M&A among startups that built their businesses solely on paid text-generation APIs.

Sources

  1. TechCrunch: Mirendil inks $100M+ Google Cloud deal
  2. TechCrunch: OpenAI's new AI smart speaker will reportedly sell for between $300 and $400
  3. NVIDIA Blog: Into the Omniverse: How Open World Models Push the Frontier
  4. TechCrunch: Naïve raises $28.5M to automate the grunt work
  5. TechCrunch: Google Maps adds agentic features
  6. TechCrunch: ChatGPT brings unlimited text chats to free users
  7. VentureBeat: Liquid AI's new model brings powerful AI agents to devices
  8. VentureBeat: Qwen 3.8-Max and Claude Opus 5 show why raw benchmark scores don't predict the bill