Ben Thompson: Apple still doesn't get the Vision Pro — just put a camera courtside and stop editing

Summary

Ben Thompson's core argument is blunt: Apple is producing TV for a device that is not a TV, and it is killing the Vision Pro's best use case before it has a chance to scale.

The NBA live sports experience on Vision Pro — geo-locked to Southern California and produced by Spectrum — demonstrated that the underlying technology works. Cameras are small enough to sit courtside and under the basket, latency is not an issue, and the immersive presence is genuinely compelling. Thompson describes a three-second NBA clip shown only on Vision Pro's launch day, before being pulled from retail demos, as the single most persuasive demonstration of the device he has seen. The clip has not appeared since, possibly due to rights complications.

What followed for the live NBA broadcast was a self-defeating production stack: dedicated pregame hosts, bespoke play-by-play announcers, multiple camera angles, and a production truck actively cutting between views every few seconds. Thompson turned off the immersive feed mid-game and switched to conventional TV because the constant camera cuts made the experience physically uncomfortable. His comparison is precise — it is like buying a courtside seat and being forced to stand up and move to a different location every thirty seconds.

The fix, in his view, costs roughly $30,000 per venue. The Blackmagic Ursa Cine Immersive, a dual-fisheye 16K camera, is a one-time capital outlay. Install it in every arena, every concert venue, sell virtual attendance to every live event in the world, and the marginal cost per additional viewer approaches zero. The content problem and the hardware adoption problem solve each other simultaneously. Thompson frames the $3,500 Vision Pro price as trivially justified if the device becomes a universal live-event portal — he notes he previously flew from Taiwan to the US specifically to attend NBA games, spending well over $3,500 per trip.

The same production failure appears across all Apple's immersive content. The Metallica experience opens with a genuinely stunning walk through a crowd toward James Hetfield, then cuts away abruptly. An MLS season-review video — Apple holds the MLS rights in a deal Thompson characterises as one of the most expensive sports overpays in history — contained approximately 56 cuts. The Alicia Keys content has the same problem. The pattern is consistent enough that Thompson rejects the charitable interpretation that rights-holders are forcing Apple's hand. Apple's own production team is applying television grammar to a spatial medium.

Thompson draws a direct parallel to the Apple Watch. The Watch launched with a diffuse feature set — digital heartbeats, wrist drawings, fashion positioning — before consolidating around fitness and health monitoring. That anchor use case drove adoption, which drew developers, which created the platform. The Vision Pro needs the same forcing function. His candidate is live events. Without a mass-market install base, developers have no incentive to build for the platform, as illustrated by a developer who joined immediately after the Vision Pro announcement, spent three months building for it, and abandoned the effort when it became clear the addressable market was too small to justify continued investment.

On Meta's Quest as a competitor, Thompson is dismissive on the immersive content side. He acknowledges Quest has advantages in specific categories like gaming but says those use cases do not interest him. He notes that Apple effectively invented the immersive camera pipeline that Blackmagic is now commercialising, giving Vision Pro a qualitative edge in the experience that matters most to him. A James Cameron-linked source suggested Meta may be two years behind Apple on display technology for next-generation Quest hardware.

On Apple's reported deal to integrate Gemini into Siri, Thompson sees clear logic for both sides but raises a structural concern. The Google search default was a clean, loosely-coupled relationship — Apple could credibly threaten to switch providers, which is why value flowed toward Apple at roughly $20 billion per year. AI integration is fundamentally different. It requires deep embedding into the operating system, which means Apple is more dependent on Google than Google is on Apple. Thompson doubts Apple can simultaneously develop a competitive in-house LLM while running Gemini in production — citing his time at Microsoft where developer subsidies without user demand failed — and worries that Apple's leverage in future negotiations will erode as a result. He also flags that intermediate reasoning steps and user interaction data will accrue to Google regardless of how Apple structures its privacy protections.

Mark Gurman has reported Apple paying approximately $1 billion for the Gemini integration, though Thompson notes that figure is likely blended into the broader search revenue arrangement.

On OpenAI's failure to build an advertising business, Thompson is direct to the point of frustration. He argues Facebook and Google advertising are already the world's most profitable and most mature AI agent systems — an advertiser states a cost-per-acquisition target, the platform executes autonomously. OpenAI had a window in 2023 when it was the only meaningful consumer LLM to launch ads, absorb the inevitable negative press cycle, iterate, and arrive in 2026 with a functioning revenue stream. Instead, it has a religiously anti-advertising culture imported partly from Meta talent, is burning cash at scale, and will eventually be forced to launch ads into a competitive market where Gemini and others already exist. Thompson calls the delay company-imperiling.

His view on inventory dynamics in social advertising is worth noting for Meta investors. Every time Meta has introduced a major new ad surface — Stories in 2018-2019, Reels two years ago — the price-per-ad dropped sharply as inventory expanded, the market panicked, and the stock fell. Both times, Meta subsequently figured out monetisation and the stock recovered significantly. He expects the same dynamic to play out as generative AI creates new ad inventory surfaces, and cites Grok's inference speed as specifically relevant to real-time ad auction latency requirements.

On Meta's acquisition of Manus, Thompson's read is that it is primarily a talent acquisition rather than a product bet. He is sceptical Meta has a credible enterprise software strategy and doubts Manus will scale into a material standalone revenue line. The more plausible path is the team contributing to Meta's agentic workflow capabilities internally.