How the ‘funniest chart in tech right now’ shows how almost all analysts were ‘very wrong’ on Apple
A chart from A16Z titled “Apple on Capex: ‘Nah, we’re good'” might be the most humbling visual Wall Street has seen in a while. Sourced from FactSet data as of February 9, 2026, it shows Apple’s quarterly capex sitting nearly flat since 2016—down 19% year-over-year—while Amazon (+42%), Microsoft (+89%), Alphabet (+95%), and Meta (+48%) have each rocketed into the $20–40 billion-per-quarter range. Together, these four hyperscalers are on track to spend roughly $635–700 billion on capex in 2026, most of it poured into AI data centres and GPU clusters. Apple’s projected budget for the entire year? A little over $14 billion—essentially flat year-over-year.The chart’s been called “the funniest in tech right now.” But the real joke landed on the analysts who spent two years calling Apple’s restraint a strategic disaster.
The consensus got Apple’s AI playbook backwards
Through 2024 and into mid-2025, the conventional narrative was that Apple was losing the AI race. Siri delays, no proprietary frontier model, and zero appetite for warehouse-scale GPU buildouts drew downgrades and warnings that the company was one to two years behind competitors. Apple’s Q4 2025 capex actually dropped 17% to just $2.4 billion—a rounding error next to what peers were burning per quarter.What analysts missed was that Apple wasn’t sitting out the AI era. It was building for a different version of it. The M-series unified memory architecture—optimised for edge inference since the M1 launched in 2020—turns out to be exactly what on-device AI needs. An OpenClaw-fuelled ordering frenzy has created genuine Mac shortages, with high unified memory Mac Studio orders now stretching to 54-day wait times. The new M5 Max, with 128GB of unified memory and 614GB/s bandwidth, runs a Llama 70B model quantized to Q6 entirely on a laptop—no data centre in sight. Early benchmarks show it hitting 30 tokens per second on Llama 70B, outperforming clusters that cost $40,000 just eighteen months ago.Apple also struck a deal to use Google’s Gemini to power the next generation of Siri and Apple Intelligence—reportedly worth about $1 billion a year. That’s access to a top-tier model for pennies on the dollar compared to building one from scratch.
The balance sheet tells the rest of the story
Apple posted record Q1 fiscal 2026 revenue of $143.8 billion, up 16% year-over-year, with diluted EPS of $2.84—a 19% jump. It held $145 billion in cash and returned $32 billion to shareholders in a single quarter. The hyperscalers, meanwhile, had a rougher time. Amazon, Google, and Microsoft saw a combined $900 billion in market value erased after earnings reports, as investors grew uneasy about the gap between AI spending and actual returns.Whether Apple’s spending gap represents a failure of vision or an exercise of it depends on a question nobody can answer yet: will AI models become interchangeable commodities? Apple appears to be betting they will. If it’s right, the flattest line on the graph was the smartest one all along.
