The editorial argues Apple's choice wasn't about benchmarks or context windows but about competitive asymmetry. OpenAI is building a device with Jony Ive and Anthropic has Claude as a destination product, while Google's Pixel is a rounding error against iPhone — making Google the only frontier lab that can be a permanent supplier rather than a future competitor.
The 563-upvote HN thread spent most of its energy debating 'why Google?' by cycling through Gemini's benchmark scores, context window comparisons, and nostalgia for the Safari default-search deal. This conventional read frames the decision as a straightforward vendor evaluation where Google's model specs won out.
The editorial acknowledges this conventional read as 'true but trivial' — Apple's own foundation models remain restricted to on-device tasks like notification summaries, forcing them to outsource heavy reasoning. But it argues this framing misses the more interesting strategic calculation about which vendor poses the least long-term competitive threat.
Google pays Apple north of $20 billion annually to be Safari's default search engine, a figure disclosed during the DOJ antitrust trial. That arrangement has trained both companies on the mechanics of one paying the other for distribution while remaining nominally independent — making the Gemini-in-Siri deal a natural extension of an existing playbook.
Apple's long-delayed Siri overhaul shipped this week, and the partner powering the heavy reasoning is Google's Gemini — not OpenAI, not Anthropic, not Apple's own foundation models, which remain restricted to on-device tasks like notification summaries and writing tools. The Apple Intelligence page now lists Gemini as the default cloud model behind the new App Intents-driven Siri, with a per-query opt-in dialog and a privacy-preserving relay that strips IP addresses before requests hit Google's servers.
The 563-upvote HN thread spent most of its energy on the obvious question — "why Google?" — and most of the answers missed the point. The top comments cycled through benchmark scores, context window comparisons, and Apple-Google search deal nostalgia. The actual answer is structural: Apple chose the only frontier lab that is constitutionally incapable of competing with the iPhone.
OpenAI is building a consumer device with Jony Ive. Anthropic has Claude as a destination product with rapidly growing direct revenue. Both have explicit ambitions to own the user relationship. Google has Pixel — a product whose global market share remains a rounding error against iPhone, and which Google has shown no willingness to subsidize into relevance. From Apple's perspective, that asymmetry is the entire point.
The conventional read is that Apple is behind on AI and had to pick a vendor. That's true but trivial. The interesting read is that Apple evaluated three frontier labs and chose the one whose corporate strategy makes them a permanent supplier rather than a future competitor.
Consider the cash flow that's already running in the opposite direction. Google pays Apple somewhere north of $20 billion annually to be the default search engine in Safari — a number disclosed during the DOJ antitrust trial. That deal has trained both companies on the mechanics of one paying the other for distribution while remaining nominally independent. The Gemini-in-Siri deal is the same shape with the polarity flipped: Apple pays Google for inference, Google pays Apple for search placement, and the net cash flow stays favorable to Cupertino. Neither company has to disclose the inference contract terms because the search contract dwarfs it.
Now imagine the same deal with OpenAI. Microsoft owns a sizable chunk of OpenAI's economics. Apple does not negotiate from a position of strength against Microsoft — they negotiate from a position of mutually assured indifference, which is worse. ChatGPT's consumer app already cannibalizes Siri usage on iOS. Embedding OpenAI deeper would accelerate the substitution. Sam Altman has publicly speculated about an AI-native device. None of this points to a stable supplier relationship.
Anthropic is even less workable from Apple's seat. Claude is the model developers respect most, but Anthropic's growth thesis depends on enterprise API revenue and direct consumer subscriptions — exactly the surfaces Apple wants to control. Anthropic also has no ad business to subsidize unprofitable inference, meaning any deal would need to price closer to actual cost. Google can eat the margin because Gemini inference is a rounding error against YouTube and Search.
The HN commenters who framed this as Apple "settling" for the weaker model are reading the spec sheet instead of the org chart. Apple didn't pick the best model. Apple picked the model whose owner has the least incentive to ever turn this relationship into a knife fight.
There's a secondary signal worth flagging: Apple's own foundation model team has been at this for years and the on-device models still cap out at writing assistance and summarization. The cloud-tier work was supposed to be theirs. It isn't. That's the second admission buried in this announcement — Apple's internal frontier effort has not produced something competitive, and the company has decided that's acceptable as long as the partner is contained.
For iOS developers, the most concrete change is that App Intents finally has a model behind it that can actually do multi-step orchestration. The original Siri Shortcuts framework assumed users would manually compose actions; the new version lets Gemini-backed Siri compose them on the fly from natural language. If you maintain an iOS app, your App Intents surface area just became the equivalent of a 2010-era REST API — the integration layer that determines whether you exist inside the new agent loop or sit outside it.
This is not optional work to defer. The apps that exposed rich, well-typed intents during the iOS 18 cycle are going to surface in Siri's reasoning chains for years. The apps that didn't will get routed around. Audit your `AppIntent` conformances, make sure your `entityType` definitions actually expose the nouns users care about, and write parameter summaries as if they were OpenAPI descriptions — because that's effectively what Gemini is reading.
For non-Apple-platform developers, the relevant takeaway is the precedent. The largest consumer software vendor in the world just publicly endorsed the model-as-utility framing — that frontier models are inputs you buy from a supplier, not capabilities you build. Every product team that's been hand-wringing about whether to fine-tune their own model has just been given air cover to stop pretending and call the API.
The next twelve months will reveal whether Apple's containment strategy holds, or whether Google quietly does to Siri what AWS did to every retailer that used it — learn the workload, then ship the competing product. Google's restraint here is not a permanent feature of its corporate personality; it's a function of Pixel being structurally subscale. If that changes — if Google ever decides the phone business is worth subsidizing for real — the most consequential supplier relationship in consumer tech turns into the most consequential conflict. Apple is betting that won't happen. Based on the last decade of Pixel earnings, that's a reasonable bet. For now.
I didn't really see anything that knocked my socks off. Mostly, it's the promise that Siri now works in the way in which they said it would work a few years ago, when it didn't. I do like the addition of Siri in the context menu, though. I can see that being useful.
The demo Mike Rockwell gave at WWDC was interesting. He kinda showed off Siri as like the Star Trek computer for your phone. I hope this is the direction Apple is going to continue in. Having AI as a user interface is way more interesting than chat bots, image editors, or copy editing.
They promised Apple Intelligence with iPhone 15 Pro and more recent models.Now [relevant parts of] Siri AI is restricted to iPhone 17 / iPhone Air and more recent models.People who believed Apple and bought an iPhone 16 to use with Apple Intelligence are getting the shaft.
None for the EUIs it available in China at least or is this another “50% of the userbase gets nothing new in the OS update” year?Edit: https://x.com/wongmjane/status/2064052590992916840?s=46Lol
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
It's interesting how anemic the use cases seem to be - we see the same things recycled over and over: "reword my email", "remove object from picture", "add a reminder", "summarise my text message which was already only 20 words long" etc etc. As if these