Amazon's 200MW Swedish Wind Deal Is a Footnote to a $220 Billion Power Problem

Amazon said on August 28 that it had signed long-term agreements to buy electricity from four wind farms in Sweden as it expands its data-center capacity there. The headline figure is about 200 megawatts, which reads as one more clean-energy bullet point in a corporate green portfolio. Before letting the round number do the work, take it apart, because who the megawatts are attached to — and which of them are not even generated yet — tells you more about how Amazon now spends $220 billion a year than the contract itself does.

Break up the 200 MW and you get 199 MW spread across four power purchase agreements. Three of them, together 67 MW, cover wind farms in southern Sweden that are already selling power to the grid: Boarp, Dållebo, and Fågelås, a portfolio named Saena that French investment manager Mirova bought last year from Swedish developer Eolus. The fourth contract is the telling one: Amazon agreed to offtake 132 MW from Fageråsen, a 189 MW wind farm in Dalarna County that is still under construction and not expected to reach commercial operation until 2028.

That contract for a farm that does not exist yet is the structural point. A power purchase agreement is not an ownership stake. Amazon puts no turbines on its balance sheet; it signs a long-term promise to buy electricity on agreed terms. What that promise buys on the other side is a bankable revenue stream — precisely the thing an unbuilt 189 MW project needs to finance its construction. Amazon's balance sheet is, in effect, part of the wind farm's capital. In exchange, Amazon gets what its data centers actually consume: a fixed, long-term price for electrons plus the renewable-energy attributes, in a market where wholesale power prices can swing sharply when a cold Nordic winter tightens supply.

The four contracts lift Amazon's committed Swedish wind portfolio to roughly 985 MW across nine utility-scale projects — near a gigawatt — tied to its three data-center sites in Eskilstuna, Katrineholm, and Västerås. Amazon says the full portfolio, once operational, will generate enough electricity annually to supply more than 350,000 Swedish households. It was already the largest corporate buyer of fossil-free energy in Sweden in 2025, and globally it has contracted more than 40 GW of carbon-free capacity across over 700 projects in 28 countries; its European PPAs have backed 16.5 GW of new capacity over five years. The cadence is accelerating, too: the Swedish deals came in the same summer Amazon signed Germany's largest-ever corporate PPA.

Now put the scale next to the demand side. A gigawatt of wind nameplate produces only about a third of its rating across an average year — so the entire roughly-1-GW Swedish commitment works out to roughly 3 terawatt-hours annually, in the same ballpark as the 350,000-household figure Amazon gives for it. The reason the arithmetic is worth doing is not the peaking of wind; it is the ratio. Amazon has said it plans to deploy more than a million Nvidia GPUs starting this year and another two million across 2027 and 2028. Every one of those chips is, among other things, an appliance for turning electricity into compute. Power is the input that decides whether the rest of the machine can run, which is why the company keeps signing contracts in units far smaller than its ambitions.

This is where the investor's real number sits. Amazon's 2026 capital spending guidance stands at roughly $220 billion — up 76% from the $125 billion it actually spent in 2025 — with nearly all of it going to AWS data centers, power capacity, and AI chips. The wind agreements are the cheap end of that capital stack: a handful of long-term power commitments measured against a $200 billion-plus annual spend, bought so the expensive end — the chips and the concrete — never idles for lack of electrons.

The cost side of that trade is already visible. Amazon's trailing free cash flow has swung from roughly +$18 billion a year ago to somewhere around negative $8 billion to $12 billion across accounting cuts, while it sells at least $25 billion of bonds to fund the buildout. The revenue side is intact: Q2 cloud revenue rose 37% year over year to $42.2 billion, the fastest AWS growth in eighteen quarters, and AWS's AI business has passed a $25 billion annualized run rate. The temptation is to read the quarter's $5.75 EPS beat as proof the model works; decompose it and more than $53 billion of the $62.6 billion net income was a non-cash mark-to-market gain on Amazon's Anthropic stake. The real earnings engine is being built, and paid for, at the same time.

So take the 199 MW at face value: it is immaterial to revenue and earnings, and Amazon has not disclosed the contract prices, so no one can say what it pays per megawatt-hour. What the announcement does is confirm where the constraint sits. Amazon's buildout is not going to stop for money, chips, or customers — its AWS backlog stands near $500 billion — but it can stop for power and grid connections, which are local, finite, and slow. The convertible detail to watch is whether the gap between contracted power and committed capacity stays closed in each region, and, on the income statement, whether AWS growth stays near 30-37% while the depreciation on a $220 billion annual build rate starts loading onto earnings. That arithmetic, not megawatts of wind, decides whether this is the decade's best capital allocation or a bill that comes due.