Apple vs Google

The Research Gap: Nobel Prizes vs. Neural Engines

The asymmetry in fundamental AI research between these two organizations is stark. Google DeepMind has produced multiple generational breakthroughs — AlphaFold earned the 2024 Nobel Prize in Chemistry for solving protein structure prediction, a problem that had eluded biologists for fifty years. Its Gemini Deep Think system has autonomously solved open mathematical conjectures. Apple publishes comparatively little research; its contributions are concentrated in on-device efficiency, neural engine optimization, and differential privacy. Apple's strategy treats frontier model development as a commodity input (evidenced by its adoption of Gemini for Siri's cloud reasoning) while investing in the last mile: making AI work reliably and privately on a phone in your pocket.

The Agentic Economy: Service Mesh vs. Open Protocols

Both companies are positioning for the agentic economy, but from opposite ends. Apple's App Intents framework is a closed-ecosystem power play: every iOS app that adopts it becomes a tool that Siri and Apple Intelligence agents can discover and invoke, creating what may be the most natural agentic service mesh in consumer computing. Apple Pay provides transactional closure. Google's approach is protocol-first and open: the A2A protocol enables inter-agent communication across platforms, the Agent Development Kit (ADK) provides scaffolding for multi-step agents, and the Universal Commerce Protocol aims to standardize how agents transact across the web. Google's bet is that openness wins at scale; Apple's bet is that control wins at quality.

The Capex Divergence and What It Signals

Alphabet's planned $175–185 billion in 2026 capital expenditure — roughly double its 2025 spend — represents the largest AI infrastructure investment in corporate history. This funds TPU clusters, data center buildouts, and the compute required to train and serve Gemini at global scale. Apple's capital allocation is an order of magnitude smaller for AI-specific infrastructure, reflecting its conviction that the marginal intelligence gains from massive cloud compute are less valuable than controlling the device interface. This divergence is not just a spending difference — it reflects fundamentally different theories about where value accrues in the AI stack. If foundation models commoditize (as Apple appears to believe), then control of the device and user relationship matters most. If frontier capability remains a durable moat (as Google bets), then scale of compute and data is decisive.

Spatial Computing vs. Embodied AI

Apple's Vision Pro represents the most ambitious consumer bet on spatial computing since the iPhone, but commercial traction has been disappointing — production was halted after roughly 390,000 units shipped in 2024, and Q4 2025 shipments collapsed to an estimated 45,000 units. Apple is pivoting toward lighter-weight AR glasses expected in late 2026. Google DeepMind has no consumer headset but is pursuing embodied AI through robotics partnerships — integrating Gemini Robotics models into Boston Dynamics' Atlas humanoid and Agile Robots' industrial systems. These represent different visions of how AI enters the physical world: Apple through wearable computing overlaid on reality, Google through autonomous machines that operate within it.

Platform Economics and Distribution

Apple controls the device that two billion people carry, along with the App Store's distribution economics (and its contentious 30% commission). Google controls the default information access layer for three billion Android users, plus Search, YouTube, Chrome, and Workspace. In the agentic economy, distribution increasingly means which platform's agents users default to. Apple's advantage is that it controls the OS-level integration point — Siri is the gatekeeper. Google's advantage is that its services are already the default integration targets for most agentic code: Gmail, Calendar, Drive, and Maps APIs are where agents go to get things done. The question is whether the OS layer or the service layer proves more durable as AI agents reshape computing.

The Model Supply Chain Paradox

Perhaps the most revealing signal in this rivalry is Apple's reported decision to adopt Google's Gemini models to power the new Siri — a move reflecting Apple's internal assessment that large language models may become commoditized inputs not worth the cost of proprietary development at frontier scale. This creates a paradox: Apple's flagship AI assistant may run on its competitor's models, while Apple focuses on the integration, privacy, and user-experience layers above. For Google, this validates the massive capex investment — even competitors become customers. For Apple, it represents a calculated risk that the orchestration layer matters more than the intelligence layer.