Everyone Saw Mozal Coming. It Closed Anyway.

In March 2026 the Mozal smelter in Mozambique stopped making aluminum. Its electricity agreement ran out. Running, it drew a constant 950 megawatts.

South32 had spent years trying to replace it, talking to the Mozambican government, to the Cahora Bassa hydro company that supplied the plant, and to Eskom across the border. In August 2025 it told the market those talks did "not provide confidence" that Mozal would get affordable power past March 2026. It stopped relining pots, stood down contractors and took an impairment of US$372 million. Carrying value fell to $68 million.

Here is the part worth taking away. Nobody was caught out. On the day it closed, South32's chief executive said the company had "engaged extensively" with the Mozambican government, Eskom and other stakeholders "over the past six years" and still could not secure affordable power beyond March 2026. It had written the plant down seven months earlier. The date was never the missing information.

An expiry date tells you when you will be at the table. It tells you nothing about what you will be holding when you get there. Mozal arrived at a negotiation it could not win, exactly on schedule. The hydro resource had weakened, Eskom would not price at a level that kept the smelter competitive, and Mozal had no alternative supply and no generation of its own.

That matters because smelters buy nearly all their power. Alcoa, the most integrated producer in the industry, made about 11 percent of what its smelters used in 2025 and bought the rest.

The obvious way to find the next Mozal is to sort every plant by when its contract ends. I did that, and not one of the hits meant what the date implied.

Five alarms, and what each one actually was

Massena West, New York. Alcoa's filing says the New York Power Authority contract expires in March 2026. The next sentence says the smelter signed a ten year replacement in October 2025 for 240 megawatts, effective the day the old one ended, extendable by two further five year terms.

Portland, Victoria. The filing says hedges covering 587 megawatts expired on 30 June 2026. The next sentence says nine year replacements for the same 587 megawatts were signed in 2023 and 2024, effective 1 July 2026. No gap.

Baie-Comeau, Quebec. The filing says Hydro-Quebec contracts for Alcoa's three Quebec smelters expire on 31 December 2029. Two sentences later it says the Baie-Comeau contract renews automatically through February 2036.

Those are all one company, so I ran the same test on two more.

Grundartangi, Iceland. Century says its Icelandic agreements "expire on various dates from 2026 through 2036," and the same sentence closes with "(subject to extension)." The qualifier is inside the parenthesis at the end of the date range.

Mount Holly, South Carolina. Century's table shows Santee Cooper supply through December 2031. Elsewhere the filing explains that date is itself an extension, finalized in October 2025, and that it exists so the smelter can return to full production.

Sort those five honestly and they are not one thing.

Three were simply done. Massena West, Portland and Baie-Comeau were replaced or renewed before the date arrived, and the company said so in the same document. Portland's replacement was signed two to three years ahead. Massena West's landed about five months out.

One was handled and created a new date. Mount Holly's December 2031 is an extension finalized in October 2025. Near-term, solved. In 2031, real.

And one is not resolved at all. Grundartangi's agreements "expire on various dates from 2026 through 2036 (subject to extension)." Subject to extension means an extension is possible. It does not mean one exists. I read that as an answer the first time through, and it is not.

So a date that looks alarming usually means somebody already fixed it. These get settled well ahead, and the fix gets disclosed when it lands. The screen is reading the announcement, not the risk.

And the genuinely unresolved case has no date at all. Rio Tinto states that Tomago "faced the risk of closure before 2030 due to challenges in securing a competitive energy solution after its current electricity contract expires." No replacement is described because there is not one, and Rio publishes no expiry date for the plant. It surfaced as risk language in prose, which is where an unsolved problem tends to live.

Where the exposure actually sits

Read to the end of every paragraph and seven plants reach a genuine reset by 2031. Lista in Norway in 2027. Mosjoen, also Norway, and the Icelandic price reopener in 2028. Becancour and Deschambault in Quebec at the end of 2029, both coming off the same Hydro-Quebec contract on the same day. Mount Holly in South Carolina and Hillside in South Africa in 2031.

Grundartangi's contracts end at different times between 2026 and 2036. That spreads the risk. But Century only says they can be extended, not that they have been.

Two others sit outside the framing entirely. Century's Sebree in Kentucky runs to May 2028, but at a variable rate set by market prices through MISO, so the date tells you when the arrangement ends and nothing about what it costs before then. A term without a price is not cover. Alcoa's Warrick smelter in Indiana runs on a co-located coal plant supplying substantially all of its power. No counterparty, no expiry, and a different problem instead.

The far end of that road is Alba in Bahrain, which owns five power stations and took one complex to 2,481 megawatts in December 2024. Alba also has a Chief Power Officer. That is where this argument finishes: a smelter that stopped treating electricity as something it buys and gave it a seat at the top table.

What actually separated them

Five things told the real exposures apart from the false ones, and not one of them was the expiry date. Each is answerable from a public filing.

1. Does it renew by itself? Baie-Comeau renews automatically through 2036. Its two sister plants do not. Same supplier, same page, completely different position.

2. Does the price reopen before the contract ends? Alcoa's Iceland supply runs to 2048 and reprices in 2028. A twenty-two year gap between those two numbers.

3. What is the price tied to? Fixed, the metal, or inflation. Roughly 70 percent of Grundartangi's power is priced off the London Metal Exchange, and Alcoa carries embedded derivatives indexing power to the same benchmark out to 2036, so those bills fall when aluminum does. Hillside's rises with South African inflation whether the metal cooperates or not.

4. Who else shares that supplier and that date? Becancour and Deschambault both come off Hydro-Quebec on 31 December 2029. One site negotiating alone is procurement. Two sites negotiating against the same counterparty on the same morning is a portfolio, and it is the fastest way to weaken your own hand.

5. Can they switch you off, and are you paid for it? Eskom keeps the right to interrupt Hillside. If you carry that and did not get a discount, someone mispriced it.

Then ask the sixth, which is the one nobody writes down. What happens if nothing replaces it. Mozal is the honest answer. Not a sensitivity, a plant on care and maintenance.

The date is not the risk. Being unable to replace the power is.

Methodology

Every asset here is from company reporting. Alcoa, Century Aluminum and Rio Tinto from annual filings for financial year 2025. South32 from exchange releases, including the Mozal update of 14 August 2025 and its August 2021 release finalising the Hillside energy supply agreement.

No delivered power price is claimed for any smelter. Producers do not publish it, and a number offered for it would be a guess in a suit. What they do disclose is counterparty, term, renewal rights and pricing basis, which is enough to judge durability without pretending to know cost.

Two rules came out of getting this wrong. Read to the end of the paragraph, because the disproof of an expiry date tends to sit beside it. Then check anything expiring within about eighteen months of a filing date against what came afterward, because a filing is fixed and the world is not.

Coverage. Disclosure follows where a company lists, not how good its assets are. American, Australian and European producers report this. Chinese and Russian producers do not, and China is roughly three fifths of world output. This reads the part of the industry that publishes, which is also the part that competes on cost.

All analysis by Tafel Power from public sources.