Our Take
The AI infrastructure trade was real; execution risk was always the question, and it's arriving now.
Why it matters
GPU and chip makers powered a two-year bull run on the assumption of durable scarcity and pricing power. Margin compression signals that assumption is breaking, which reshapes valuation and forces investors to distinguish between capacity growth and actual profit growth.
Do this week
Infrastructure teams: audit your GPU procurement contracts for price escalators and lock in multi-year rates before Q2 earnings calls reset vendor guidance.
The margin pressure is real
The trade that drove GPU and chip maker valuations higher for two years rested on a simple mechanic: acute scarcity of AI compute, combined with heavy customer demand, meant suppliers could sustain premium pricing. Bloomberg's reporting shows that assumption is eroding. Revenue is growing, but margins are not keeping pace (per Bloomberg). Competition has intensified, customers are becoming more price-sensitive, and capacity additions by both incumbents and new entrants are expanding supply faster than demand growth alone would justify.
This is not a failure of demand. Data center and cloud infrastructure spending on AI remains robust. The issue is structural: when multiple suppliers can meet the need, pricing power shifts to buyers. That is exactly what is happening across the GPU and semiconductor market now.
Valuation assumptions are cracking
Equity markets priced in two beliefs: (1) AI compute would remain scarce for years, and (2) suppliers would earn outsized returns on that scarcity. Both are now in question. Revenue growth alone no longer justifies the multiple expansion that supported the sector's run. Investors are rotating from "this company will grow" to "at what margin will this company grow," and the answer is becoming messier.
For practitioners making infrastructure spend decisions, this is a reprieve. Customers who faced take-it-or-leave-it vendor terms a year ago now have leverage. Spot pricing is becoming more competitive. Multi-year contracts are negotiable. The scarcity premium is dissolving.
Lock contracts before guidance resets
If you manage cloud infrastructure or GPU procurement, the window for favorable long-term pricing is now. Vendors are still committed to revenue targets, which means they will offer multi-year discounts to lock bookings before Q2 and Q3 earnings calls reset guidance downward. Once Wall Street reprices for lower margins, vendor willingness to negotiate rates on new contracts will decline. Current-quarter results will force a conversation inside every chip maker about sustainability of their pricing strategy. That conversation will end in lower prices. Get ahead of it.
For chip makers themselves, the challenge is capital allocation. You can chase volume (and compress margin further) or constrain supply (and risk losing share to competitors willing to price lower). The comfortable middle ground that existed six months ago is gone. That shift, not the demand slowdown, is what earnings guidance will reflect this quarter.