Power supplies were the last boring part of a PC build — a mature product, quietly getting cheaper for a decade. That run is over. PSU prices in the U.S. have been climbing since mid-2025, manufacturers announced a fresh round of increases for 2026, and the same AI datacenter boom that made RAM shockingly expensive is now working its way through the power-supply supply chain. This article covers what happened, when it started, whether relief is coming, and how to buy well in the meantime.
For live prices while you read, our ATX power supply and SFX power supply price lists rank current units, and the price-per-watt hub shows which wattage tiers are actually good value today.
Yes, prices have risen — in two distinct waves
Unlike RAM, which roughly quintupled in a year, power supplies have climbed the way a tide comes in: several separate increases stacking on top of each other.
The first wave was tariffs, starting in spring 2025. The U.S. reciprocal-tariff regime announced in April 2025 hit power supplies harder than most PC parts, because PSU manufacturing is concentrated in exactly the countries the tariffs target. As of 2026, a power supply imported from China — where most mainstream units are built — carries a 45% import duty. A $100 unit lands at $145 before freight or retailer margin. Units built in Taiwan or Malaysia land at $119, and the Philippines at $120 — better, but still a fifth more expensive than the sticker used to imply:
The second wave is raw materials and components, and it's the one arriving now. In January 2026, a major Chinese components distributor warned customers that PSU prices would rise another 6–10% in 2026 (CPU coolers 6–8%), with upstream factories refusing new orders at old prices from January 6. The stated cause: the metals a power supply is made of. A PSU is, at heart, copper windings, tin solder, and silver contact plating wrapped around silicon — and all three metals spiked. Copper hit record highs above $13,000 per ton on the LME in January 2026, tin passed $50,000 per tonne in late February, and silver has more than doubled from its early-2025 lows on industrial demand.
There's a quieter third force: wattage inflation. Flagship GPUs now draw what whole systems used to — NVIDIA's RTX 5090 has a 575W rating and transient spikes near 600W, with a 1000W minimum recommendation and a 1200W ATX 3.1 consensus among system integrators. Even where per-unit prices held steady, the unit a high-end builder needs moved two tiers up the price ladder.
Why this is happening: AI is eating the power chain
The tariff math is simple policy. The component squeeze is the same story that broke the RAM market, one level down the parts list.
An AI datacenter is, from a supplier's point of view, a gigantic power-conversion project. Every accelerator rack pulls demand through the full chain: front-end power supplies, DC-DC stages, power-factor correction, gate drivers, magnetics, and enormous quantities of capacitors. The consequences through 2026, documented by supply-chain trackers: multilayer ceramic capacitors became the third-largest cost category in AI infrastructure, passive component prices rose 15–30%, MOSFET makers pushed through 10–20% increases, and lead times on power semiconductors stretched to 20–30 weeks. Those are precisely the parts inside a desktop PSU.
The manufacturers themselves are pivoting toward the money. Delta Electronics and Lite-On — two of the largest power-supply makers on earth, and the OEMs behind many consumer brands — spent Computex 2026 showcasing megawatt-class rack power for AI datacenters. FSP, one of the biggest consumer-PSU OEMs, themed its entire booth "Powering AI Together". A server power shelf sells for far more per watt than a retail ATX unit, so engineering attention, component allocations, and factory capacity flow to the datacenter first. Consumer lines aren't being abandoned — but they're now the lower-priority customer of their own factories, competing for the same capacitors, MOSFETs, and transformer cores.
If this sounds familiar, it should: it's the same crowding-out mechanism that sent RAM prices up 400%. PSUs are catching a milder version of the disease because a power supply, unlike a DRAM chip, isn't a single-source wafer product — but the direction of pressure is identical.
Is there hope? Realistically: a plateau, not a rollback
The honest near-term answer is that announced increases are still arriving, not receding. The 6–10% distributor increases took effect in early 2026, copper forecasters see prices holding between $11,000 and $15,000 per ton through 2026, and AI datacenter construction — the demand engine underneath all of it — is still accelerating into 2027.
But there are genuine reasons PSU prices should level off rather than follow RAM into orbit:
- A PSU is not supply-capped the way DRAM is. Building more power supplies doesn't require a $20 billion fab. Component makers are expanding passive and power-semiconductor capacity, and new lines come online through 2027 — capacitors and MOSFETs respond to price signals far faster than memory wafers do.
- The market is fiercely competitive. Dozens of brands share a handful of OEMs, PSUs are durable goods with 7–12 year warranties, and demand is elastic — builders simply keep old units longer when prices climb. That disciplines retail pricing in a way the memory oligopoly never faces.
- Manufacturing is walking out of the tariff zone. Premium lines already ship from Taiwan and Vietnam, and every point of origin shift cuts the landed-cost gap in the chart above from 45% toward 19–20%.
- Last-generation stock is a pressure valve. ATX 3.0 units — fully capable for mid-range builds — get discounted as ATX 3.1 replaces them, which keeps real deals appearing even while list prices rise.
What would actually reverse prices, rather than flatten them? Three changes, roughly in order of impact: tariff relief or continued relocation of assembly out of China (the single biggest line item — no other factor adds 45% at the border); metal prices cooling, which depends on mine supply catching up to electrification and AI grid buildout — most analysts put that beyond 2026; and the AI power-component squeeze digesting, as expanded capacitor and power-semi capacity lands in 2027. Longer-term, the industry's shift to GaN (gallium nitride) designs helps structurally — GaN units need smaller transformers and fewer passives, which means less copper per watt.
None of those move this quarter. Our expectation for 2026: mainstream ATX units settle at a plateau roughly 15–25% above 2024 street prices, with high-wattage ATX 3.1 flagships holding premium pricing while supply chases datacenter margins instead.
How to buy a PSU in this market
Buy once, buy quality — this is the one component where that hedge is rational. A good ATX 3.1 unit with a native 12V-2×6 connector and a 10-year warranty will span two or three GPU generations. Amortized over that span, even inflated 2026 pricing is tolerable; buying a marginal unit twice is not.
Right-size the wattage. The consensus for an RTX 5090 build is 1200W ATX 3.1; a mid-range build with a 250W-class GPU still lives happily on 750–850W. Wattage you don't draw is money spent on headroom you'll never use — check the price-per-watt hub to see which tiers are efficiently priced right now, because the answer shifts week to week.
Watch for last-gen and OEM value. Our trackers list everything from flagship modular units to OEM pulls under $10 — the price spread in this category is enormous, and the sub-$100 shelf still holds honest 80 Plus Gold units:
Set an alert instead of waiting for a crash. Nobody credible forecasts PSUs getting meaningfully cheaper in the next 12 months, but individual units dip constantly as retailers clear stock. Watch a specific model on our lists and buy the dip when it fires.