On February 5, analyst Brad Reback cut Microsoft's stock to hold, citing AI spending pressures on Azure's cloud unit. He projected Microsoft Cloud gross margins would compress to 63% from 67%. The price target fell to $392 from $540. Stifel had held a buy rating on Microsoft for years.
That downgrade arrived eleven weeks after Microsoft announced what may be the largest AI infrastructure deal in history: a package in which Anthropic committed to spend $30 billion on Azure compute capacity, Microsoft invested up to $5 billion in Anthropic, and Nvidia committed up to $10 billion. On April 6, Anthropic confirmed its revenue run rate had topped $30 billion, up from $9 billion at the end of 2025. More than 1,000 enterprise customers are now spending over $1 million annually with the company, more than double the count in February.
That is the pipes paradox. In 1999, Cisco Systems sold the infrastructure that powered the early internet. Its routing and switching hardware was indispensable; at peak, Cisco commanded a market cap above $500 billion. When the dot-com bubble burst in 2000, Cisco's stock fell 88% from $79 to $9.50. The company that owned the pipes did not capture the value created by the people using them.
The analogy has been making the rounds in infrastructure circles for two years. The current version: if AI application companies capture the value their models create, and if those companies rent compute from Azure rather than owning it, then Azure earns thin margins on high-volume commodity business while its customers pocket the upside.
Microsoft Cloud gross margin has moved in a narrow band over the past four quarters: 67% in the September 2025 quarter, 69% in the December 2025 quarter, 68% in the January 2026 quarter, and 67% again in the March 2026 quarter, according to Microsoft's earnings reports. The compression is real, but the trajectory is not the straight line downward that the Stifel downgrade implied. Margins ticked back up between Q3 and Q4 of fiscal 2025 before slipping again. The January 2026 quarter included explicit acknowledgment that AI infrastructure investment was the driver, and management cited demand continuing to exceed available supply, which complicates the story of a customer pullback.
Anthropic's $30 billion Azure commitment does not translate to $30 billion in 2026 spend. The commitment is spread across a multi-year horizon. Based on available reporting, Anthropic's total cloud compute spend in 2026 is projected at roughly $1.9 billion across all providers, according to The Information. The Azure slice is estimated at roughly $600 million, against $30 billion in annualized revenue. The $30 billion figure is a forward commitment, not a current-year number. At current growth rates, that commitment becomes a larger fraction of revenue over time, which is the structural concern Stifel was flagging, but it has not arrived yet.
There is a meaningful counterargument. Microsoft Azure's customer base spans enterprise software, productivity, and security workloads. It is not solely dependent on Anthropic. Custom silicon is the traditional escape route from commodity infrastructure economics. Microsoft's Maia 200 AI accelerator, announced in January, delivers 10 petaflops of FP4 compute and claims threefold FP4 speed over Amazon Trainium with superior FP8 performance versus Google TPUs. But Maia 200 is not yet a meaningful offset to Nvidia GPU rental income, and the transition from commodity to custom takes years.
D.A. Davidson analyst Gil Luria told Reuters that the main feature of the Anthropic-Microsoft-Nvidia partnership is reducing the AI economy's reliance on OpenAI. That framing — the deal is strategically useful because it breaks an OpenAI dependency — is distinct from the margin question. It may be the more honest answer to why Microsoft did the deal.
The pipes paradox has not been resolved. Azure margins are compressed but not collapsing. Anthropic's Azure spend is growing but remains a small fraction of revenue. The market's downgrade is a bet that both trends move in the wrong direction simultaneously: margins keep falling and Anthropic's share of Azure revenue keeps rising. Whether that bet is right is the open question. The Cisco comparison works better as a warning about who holds pricing power in a gold rush than as a prediction of 88% stock decline. Cisco survived and eventually recovered. The more relevant question is not whether Microsoft gets punished but who sets the terms when the cycle normalizes.
