OpenAI Wants a Device of Its Own While Google Turns the Entire Internet Into One AI Interface

The discussion of “OpenAI Wants a Device of Its Own While Google Turns the Entire Internet Into One AI” yields a practical test: the race moves beyond apps — Altman and Ive discuss hardware, Apple prepares glasses, Google links AI Mode to Chrome, Gmail, and Docs, Meta translates the world through Ray-Ban; the winner must occupy the camera, voice, screen, and workflow at once.

The “OpenAI teams up with Apple's designer” issue should be assessed with one constraint in mind: the partnership with Jony Ive is meant to close a gap — OpenAI has no experience manufacturing, distributing, and supporting hardware, and history shows how hard it is to launch a new device class and scale it.

The decision in “Can OpenAI build hardware?” depends on one criterion: building a strong model and building a mass-market device are different professions — OpenAI can ship iterations of ChatGPT, but it has no experience in manufacturing, logistics, and hardware support.

The decision in “AI news straight from Computex” depends on one criterion: Computex announcements matter not for their loudness but for whether they change access to compute and hardware; for the user what counts is not the presentation but a product that actually ships and works.

The “Google's new AI Mode for $270” topic becomes clearer once this point is included: Google already has Android, Chrome, Gmail, Docs, search, and its own models, and AI Mode can connect it all into one context — but Google has shown repeatedly that owning every component does not guarantee a coherent product.

The practical meaning of “Veo 3 — the best video generation model” is that Veo 3 raises the bar in video generation, but the best model alone does not make a product — what matters more is control over the result and whether the generation is embedded in the user's workflow.

The boundary of the “GPT at $20 vs. $200: which LLM will the market choose?” case is defined by this point: the choice between the $20 and $200 tiers is decided not by a benchmark but by the specific task — the expensive mode is justified only where it delivers a repeatable result at a clear cost with error control.

The boundary of the “Meta Ray-Ban, Apple, Google: the era of glasses begins” case is defined by this point: apple is preparing its response and controls the iPhone. Microsoft is strong in workplace software, but it lacks a model of the same status and depends on partners.

In the context of “800 million weekly GPT users - are the figures correct?,” this criterion applies: claims of hundreds of millions of weekly users sound impressive, but new models do not always create a new product. A person still opens a chat and manually moves the result into a CRM, documents, and email.

The decision in “CRM will not be in the current form anymore!” depends on one criterion: a real agent appears when the boundary between chat and system disappears — the model knows the customer history, proposes an action, and carries it out in the interface; but that requires permissions, memory, integrations, and trust, not just intelligence and a nice device.