AI Earnings Season: Giants Are Picking Different Paths in AI Development This round of AI earnings season, what matters most isn’t revenue—it’s how the giants are signaling their priorities with capital. On the AI journey, no one wants to compete head-on with the others. Google is the most comprehensive. It builds everything in-house: cloud, chips, and models. Its TPU chips generated tens of billions in annual revenue, with 2025 capital expenditures projected at around $92 billion. Google wants full control over the entire stack. Amazon isn’t competing on models—it’s focused on building Trainium chips. Trainium2 generates billions in annual revenue, with over a million units in production. It supplied over half a million chips to Anthropic for Project Rainier and also sells compute power to OpenAI. It doesn’t build the smartest brain—it sells the machines that run it. NVIDIA has taken “selling shovels” to the extreme. It invested up to $100 billion in OpenAI to co-build at least 10 gigawatts of data centers housing millions of GPUs. It’s supplier, investor, and shareholder—all at once—betting its balance sheet directly on the outcome. Microsoft plays the platform game, having long distanced itself from exclusive ties with OpenAI. Azure AI Foundry integrates OpenAI, Claude, Grok, Mistral, and thousands of other models, serving over 70,000 customers. It treats AI capabilities and enterprise-facing agents like utilities—any model that performs well can plug in. Meta is the most aggressive—and most awkward. Its 2025 capital spending is projected at $66–72 billion, rising further in 2026. It’s all-in on models, pushing its Hyperion cluster toward 5 gigawatts. But its revenue still relies almost entirely on advertising; AI hasn’t yet generated a single dollar of direct profit, and Reality Labs lost $4.5 billion. All investment—with returns still floating in the sky. Elon Musk is the wildest. SpaceX exchanged $6 billion in stock for Cursor and controls the Memphis Colossus—equivalent to one million H100 GPUs—and rents out compute power to Anthropic and Google, generating roughly $26 billion annually. He’s bundling compute, models, and applications into one basket. Apple is the outlier. Its fiscal 2025 capital spending is just $12.7 billion, rising to about $14.3 billion in 2026. It doesn’t build large models—it integrates ChatGPT now and will likely add Gemini later, using its own Private Cloud Compute for hybrid leasing. But there’s no free lunch: rising hardware and tariff costs are squeezing margins, and iPhone revenue is already showing signs of fatigue—the price hikes are being eaten by costs. In this same AI tide, some are building full stacks, others are selling shovels, and some are burning cash to race on models. No path is right or wrong—but eventually, every bill comes due.