Alex Kantrowitz on Apple's subdued WWDC, on-device AI as a developer unlock, and Google's Sergey Brin comeback
Jun 17, 2025 with Alex Kantrowitz
Key Points
- Apple retreated sharply on AI at WWDC 2025, spending more airtime on UI redesigns than artificial intelligence, as the company's quality standards clash with generative systems' probabilistic nature.
- Apple's Services division, up 13% in 2024, faces existential threats from a lost antitrust case against Google over search payments and sustained legal pressure from Epic Games on App Store fees.
- On-device AI's real unlock for developers is eliminating inference costs entirely, lowering barriers for solo builders who previously couldn't absorb cloud compute bills for experimental features.
Summary
Apple's WWDC 2025 marked a striking retreat from the AI ambition it projected a year earlier. Alex Kantrowitz, who covered the event for CNBC from the broadcast riser, notes the conference spent more time on the redesigned Liquid Glass UI for iOS than on artificial intelligence — a stark contrast to the vision-setting around Apple Intelligence at WWDC 2024. The core problem is Apple's perfectionism colliding with AI's probabilistic nature: the technology doesn't behave consistently enough for a company with Apple's quality bar. Kantrowitz's suggested fix is an acquisition of Perplexity, which would give Apple a controllable, integrated AI search product without having to build unpredictable generative systems from scratch.
Apple's financial exposure makes the AI stumble consequential. iPhone unit sales are declining, with 2024 revenue held flat only by higher average selling prices. MacBook grew roughly 2% last year. The one division growing with conviction is Services, up 13% in 2024 — but that division contains two significant liabilities. First, the $20-plus billion annual payment from Google for default search placement is directly threatened by a federal antitrust ruling Google has already lost. Second, App Store in-app payment fees — approximately 30% of in-app transaction value — are under sustained legal pressure from Epic Games CEO Tim Sweeney. If the Google payment disappears, Services contracts sharply, Apple's software-company valuation multiple compresses, and the $3 trillion market cap rationale weakens materially.
The on-device AI debate centers on a less-discussed unlock: zero marginal cost inference for independent developers. The thesis is not that Apple's small on-device model outperforms cloud-hosted frontier models — it doesn't — but that eliminating the inference bill removes the biggest barrier for solo, non-venture-backed developers who historically built apps like Flappy Bird without worrying about cloud costs. Viral AI features have already caused surprise cost events that scared off smaller builders. Free on-device inference, combined with falling software development costs via tools like Cursor, GitHub Copilot, and Codex, could generate a long tail of lightweight AI-native apps. The analogy offered is GPS: drop the cost to near zero and Uber becomes possible. Kantrowitz was skeptical initially but acknowledged the consumer-app explosion hasn't happened yet even with powerful cloud models available, leaving open the possibility that cost, not capability, has been the binding constraint.
Sergey Brin is genuinely back in a technical capacity at Google, not just making ceremonial appearances. Google's current position reflects Sundar Pichai consolidating DeepMind and Google Brain under Demis Hassabis, ending years of internal resource competition between the two divisions. The result is visible in Google I/O's volume of AI product releases, with NotebookLM cited as a standout. Gemini's integration into Gmail is functional but inconsistent — Kantrowitz estimates it correctly executes complex email queries roughly 50% of the time. Google is in what Kantrowitz frames as active reinvention, comparable to Satya Nadella steering Microsoft away from Windows dependency toward Azure — a transition that took years but repositioned the company entirely.
The OpenAI-Google infrastructure deal reflects a pragmatic reality: OpenAI is GPU-constrained, its Microsoft hosting arrangement is unsettled, and Google's TPU capacity offers a viable offload. Google Cloud CEO Thomas Kurian has publicly said Google would welcome OpenAI models on its platform — a next step worth watching. The resulting dynamic is a frenemy structure where OpenAI depends on Google infrastructure while competing directly with Google on the consumer assistant and chatbot layer. Whether OpenAI qualifies as Big Tech at a $300 billion valuation remains an open question, though Kantrowitz sees it as directionally on that trajectory.