Kingsoft Cloud (KC) Q1 2026 Earnings Transcript

Kingsoft Cloud reported Q1 revenue of RMB 2.704 billion, up 37.2% year over year, with AI business revenue rising 91% to RMB 998 million and becoming the majority driver of public cloud revenue for the first time. Adjusted EBITDA jumped 134.7% to RMB 748 million, while AI cloud gross billings reached RMB 1.0 billion and April token-service revenue was 53x January levels. Gross margin fell to 13% from 17% sequentially due to higher server costs and upfront deployment expenses, but management said demand remains strong and kept 2026 CapEx guidance at RMB 15 billion to RMB 20 billion.

Analysis

KC is transitioning from a cyclical capacity story to a scarcity-driven pricing story. When utilization is constrained by upstream supply rather than demand, the economics favor the few operators with ready rack access, integrated procurement, and the ability to reprice contracts faster than the market can source alternative compute. That creates a second-order winner set beyond KC itself: GPU/rack landlords, network equipment vendors, and upstream component suppliers with tight inventory, while smaller “virtual GPU” resellers likely face margin compression as customers migrate toward whoever can guarantee delivery.

The key message for the next 1-2 quarters is not revenue growth; it is whether KC can convert AI token demand into durable pricing power before capex intensity outruns working capital. Flexible contract periods are a subtle tell: management is trying to shorten the lag between cost inflation and price reset, which should protect near-term gross profit, but it also introduces more volatility into reported margins and may imply customer concentration risk if the pricing discipline breaks. If supply chain bottlenecks ease, the market could quickly re-rate the company from “scarcity beneficiary” to “normal cloud operator with heavy depreciation,” which would pressure the multiple.

The contrarian miss is that this is less a broad AI cloud adoption trade than a near-monopoly on constrained capacity in specific end-markets. That makes the next catalyst sequence highly path-dependent: backlog conversion and capex execution matter more than headline billings. In other words, this can work even if the macro AI trade cools, as long as inference demand remains the marginal consumer of scarce compute; but if customer prepayments slow or Xiaomi-linked demand normalizes, the operating leverage cuts the other way because depreciation and lease costs are already locked in.

Key Decisions for Investors

• Long KC into the next 1-2 earnings prints only on pullbacks, targeting a 15-25% upside if pricing power persists and backlog converts; downside is a fast de-rate if gross margin fails to stabilize.
• Pair trade: long KC / short a basket of US-listed AI infra resellers or smaller cloud proxies over the next 3-6 months; thesis is that scarce capacity and ecosystem control will concentrate economics in the best-supplied operator.
• Buy 3-6 month KC call spreads rather than stock to express the AI-inference scarcity thesis with defined downside; structure for a move higher on continued pricing commentary, but avoid outright equity given capex and margin volatility.
• Monitor for a short signal in 1-2 quarters: if capex stays near the high end of guidance while gross margin does not recover, reduce exposure aggressively—this is the point where depreciation can outrun monetization.
• Overweight upstream beneficiaries with tight AI supply exposure on any dip: GPU distribution, networking, and memory names should be supported by KC-like demand signals for at least several quarters if the supply bottleneck persists.