Microsoft’s early AI lead has become a test of faith
TL;DR:
- Microsoft’s shares have fallen 18% over two years despite an AI boom that lifted the wider market by a third, as investors question its returns.
- Capital spending has soared from 12% of revenue in 2022 to about 50% this year, while its Copilot assistant has been slow to catch on.
- CEO Satya Nadella is betting on a new layer of “AI-native” software that learns a company’s workflows — a market rivals are also chasing.
Two years ago Microsoft looked like Big Tech’s best-placed AI player, its OpenAI partnership the envy of rivals. Today, as the Financial Times notes, the picture is markedly different: its stock has fallen 18% during a boom that has enriched almost everyone else, and its price/earnings ratio sits at a multi-year low.
The revolution devours its own
Providing the compute behind ChatGPT propelled Microsoft to the front of generative AI, but the same wave is now unsettling its core software business. Copilot has been slow to land, its shift towards building in-house models has spooked investors over cost, and frontier labs have muscled into its territory — Anthropic’s Cowork agent is exactly the kind of product once expected from Microsoft itself. The one certainty is spending: capital expenditure has climbed from 12% of revenue in 2022 to roughly half today.
Nadella’s answer, set out in a recent essay, is that AI-native applications are unlike static software — they improve as they absorb a company’s “workflows, domain knowledge, and accumulated judgment”. Capturing that will mean training internal models and building a learning loop around human workers. To sell it, Microsoft has stood up a dedicated adoption division and plans to embed 6,000 engineers on customer sites. It is not alone: Amazon Web Services is assembling “forward-deployed engineers”, and OpenAI and Anthropic have struck private-equity alliances to push deeper into business.
Looking forward
Nadella’s warning to customers doubles as a warning to Microsoft: firms that fail to train their own models risk “ceding value to a few models that eat everything they see” — and if the general-purpose labs win that ground, Microsoft’s own software franchise is hollowed out too. With earnings due this month, Wall Street wants proof the AI capex is converting into growth. For UK enterprises weighing Copilot against rivals, the episode is a reminder that even the best-resourced incumbent is still searching for returns — a caution that pairs with fresh data on firms struggling to prove AI’s value.