New details on Apple-Google AI deal revealed, including Nvidia chips: report

Apple is reportedly planning to use Google’s Gemini model to distill smaller on-device AI models for Apple Intelligence, while routing some Siri queries through Google Cloud. The company is also considering acquisitions, including Liquid AI, to help shrink models for local processing, and has recently approved Nvidia confidential compute to support privacy protections in the cloud. The report suggests Apple will keep the Private Cloud Compute branding even as some AI workloads move beyond Apple-owned servers.

Analysis

This is a quiet de-risking of Apple’s AI stack: the strategic value is no longer “own every model layer,” but “own the user experience while renting scarce compute and frontier capability.” That shift lowers time-to-market for Siri, but it also means Apple’s margin narrative becomes more dependent on external infrastructure pricing and partner leverage than investors may appreciate. The bigger second-order winner is Google: Apple effectively validates Gemini as the default enterprise-grade model for high-stakes consumer inference, which strengthens Google Cloud’s credibility even if the economics are partly opaque.

The most interesting nuance is that Apple’s privacy posture is being preserved through branding and confidential compute, not by end-to-end control of the workload. That creates a new constraint set: every incremental Siri capability now trades off latency, privacy, and unit cost, which should cap how aggressively Apple can expand AI feature breadth before the economics become visible. Nvidia benefits as the “picks-and-shovels” layer if confidential compute becomes the standard way premium consumer AI is routed through cloud, because the security feature embeds NVDA deeper into inference infrastructure rather than just training demand.

Near-term, the catalyst is WWDC in days; the market will likely react more to architecture disclosure than feature demos. Over months, the real risk is that Apple’s AI experience remains good enough to support device upgrades but not differentiated enough to re-rate the multiple. The contrarian read is that the stock may be underpricing the strategic benefit of Apple avoiding a capex arms race while still shipping credible AI, but overpricing the idea that this protects long-term ecosystem moat if Google becomes the model layer behind the scenes.

The main tail risk is execution slippage: if latency or reliability is poor, users will blame Apple, not the cloud partner, and AI credibility can deteriorate quickly within 1-2 product cycles. A second risk is partner dependence—if Google re-prices model access or Cloud usage, Apple’s gross margin headroom on AI features compresses just as expectations rise.

Key Decisions for Investors

• Long GOOGL vs short AAPL into WWDC, 1-3 month horizon: buy the partner that monetizes model + cloud usage while capping Apple’s strategic optionality; prefer a modest size because Apple can still benefit from upgrade-cycle enthusiasm.
• Add NVDA on pullbacks, 3-6 month horizon: confidential compute is a subtle but durable extension of NVDA’s moat into privacy-sensitive inference; risk/reward improves if cloud inference security becomes a broader requirement beyond Apple.
• Buy AAPL call spreads into the event, 2-8 weeks: use upside-defined exposure to a positive launch reaction, but avoid outright long equity given the risk that the market eventually focuses on dependency and margin leakage.
• Relative-value pair: long GOOGL / short AMZN or MSFT on AI-cloud credibility, 3-6 months: Apple’s adoption reinforces Google’s AI platform legitimacy more directly than it helps other cloud incumbents.
• If AAPL rallies sharply on WWDC, fade into strength after the event: the architecture is strategically mixed, so a post-event pop may be an opportunity to trim rather than chase.