<div id="readability-page-1"><div>The shovel turned at Chililabombwe on April 29, and the soil it lifted had been flagged, three years earlier, by a machine-learning model running on servers in California. Zambia's president Hakainde Hichilema sat in the audience. Mfikeyi Makayi, who runs KoBold Metals' Zambia operation, stood at the podium and gave the day its quotable line: “We must go fast, we shall go fast, and we will not take shortcuts.” A few rows away, the U.S. ambassador, Michael Gonzales, used his remarks to frame Mingomba as a flagship of the Washington-Lusaka relationship.KoBold Metals, the Berkeley company behind the project, has spent three and a half years drilling and characterizing what its algorithm flagged as a high-grade copper deposit roughly 1,700 meters below surface. The story for everyone watching commodity markets is not whether the algorithm was right in this case. It almost certainly was. The story is whether the speed advantage AI created at the front end of mining survives contact with the back end, where humans sink shafts through some of the wettest underground rock on earth.Strip away the press releases and the algorithm does one thing. It ingests four kinds of data: drilling logs, satellite imagery, geophysical surveys and decades-old core records. The output is a probability map of where copper is likely to sit. The model predicts. People drill, log, assay and verify. KoBold's CEO Kurt House has told the Zambian government the company drilled more than 120,000 meters at Mingomba in roughly two and a half years, more than was completed at the legacy license over Zambia's entire post-independence history. What the model is good at, in plain terms, is reading old data nobody had bothered to digitize and noticing patterns no single geologist could hold in her head at once.The compression happened at the front of the mining cycle, where geologists used to spend the better part of a decade chasing leads on a wing and a prayer. KoBold's pitch to its backers (Bill Gates' Breakthrough Energy Ventures, Andreessen Horowitz, T. Rowe Price and Durable Capital Partners, plus earlier checks from Jeff Bezos and Sam Altman) is that the math of mineral exploration has changed because the model can rank a thousand prospects faster than a roomful of geologists can argue about ten.Whether that math holds at scale is a separate question. What it has done at Mingomba is take the project from acquisition in late 2022 to groundbreaking in April 2026. By industry standards, that is briskly fast.By industry standards. That phrase is doing a lot of work.S&amp;P Global Market Intelligence has tracked the lead time from copper discovery to production for years. Recent figures put the global average at around 17 years. The IEA's 2025 Critical Minerals Outlook cites a similar range. In the United States the average climbs to roughly 32 years, weighed down by permitting backlogs that can outlast two presidential terms.KoBold's compression at Mingomba is from acquisition to groundbreaking, not from discovery to first metal. Production is targeted for the early 2030s. From late-2022 acquisition to first copper in, call it, 2031, that is still eight or nine years. Faster than the industry average, yes. Not the four-year miracle a casual reader of the headlines might assume.The honest read is that the algorithm did the part that ate the most calendar, finding and characterizing a deposit, and gave the rest of the schedule back to engineers, regulators and capital markets. Those clocks tick on their own, and they are mostly mechanical.At 1,700 meters the engineering domain shifts. Hydrostatic pressure rises, rock temperature climbs, and water (always the enemy in the Copperbelt) becomes a permanent line item. The Zambian Copperbelt hosts some of the wettest underground mines on the planet; flooding control consumes electricity, pumps and capital across the entire life of a mine.KoBold, for all its money and modeling firepower, has never built a mine anywhere. Its track record is in finding them. The engineering study at Mingomba is not yet finished. Makayi has said the final cost estimate is not expected until early 2027, with shaft sinking proper beginning around the same time. The smelting and refining route is still being weighed, with an in-Zambia smelter on the table but not committed.Roughly $2.3 billion is the working construction figure cited by Bloomberg, with Africa-side reporting putting the upper bound closer to $2.5 billion. The number will move when the engineering study lands. Mingomba's grade, the share of copper inside each tonne of rock, is where KoBold's case rests. The deposit is among the highest-grade undeveloped copper resources anywhere, and that is the only reason the depth math works at all. Low-grade copper at 1,700 meters does not pencil out. The smelter question is not a technical sideshow either: a Zambian smelter would mean more local jobs and a fatter share of the value chain, but it would also stretch the build.This is where the rubber meets the road for AI mineral exploration. The model can find the rock. It cannot pump the water, sink the shaft or negotiate the smelter offtake. Whether the speed gained upstream survives the grind downstream is the question the next six years will settle.The reason any of this matters beyond Zambia is the supply curve.Copper is the metal everything electric runs through. Data centers, electric vehicles, transmission lines, every gigawatt of new wind and solar. Copper sits in the middle of all of it. The IEA projects a 30% supply shortfall by 2035 under current pipelines, with declining ore grades and a collapse in new discoveries doing most of the damage.Prices have been telling the story for a year. LME 3-month copper ran above $13,000 a tonne in early January 2026 before easing into a $10,000 to $12,000 range through April. Goldman Sachs Research, in a late-2025 note, pegged the first-half 2026 average at $10,710 a tonne, with a longer-term $15,000 forecast for 2035 if supply does not catch up.Mingomba's targeted 300,000 tonnes a year would be roughly 1% of global mined copper. It does not solve the deficit. It does not even bend it noticeably. What it does is qualify as one of the very few large new-supply projects to break ground anywhere in 2026, in a year when the industry is short on greenfield activity. Whether Mingomba arrives on time will be a market signal regardless of the spot price.That is the analytical heart of the AI exploration argument. If the model genuinely speeds up the front end and the rest of the chain holds, the world gets more copper online faster. If the back end stretches the schedule out, the model's contribution shrinks to a footnote.Then there is the politics, which never sleeps.ZCCM Investments Holdings, Zambia's state-owned mining vehicle, holds 20% of Mingomba; KoBold holds 80%. The split was set in the original 2023 joint-venture announcement. On April 28, the day before the groundbreaking, ZCCM-IH chairman Phesto Musonda said publicly that Zambia would like to lift its stake to 25%. No formal negotiations have been disclosed.KoBold has put an estimated $250 million to $300 million into Mingomba so far across discovery, drilling and pre-construction. The U.S. embassy has framed the project as a bilateral marquee, useful when Washington is trying to break China's grip on the African critical-minerals trade. Hichilema has said the project demonstrates that Zambia is open for business.Resource nationalism has been spreading across Africa for two years. Burkina Faso, Niger, Mali and Ghana have either rewritten mining codes or raised royalties. The Lobito Corridor, the rail line being built to move Zambian and Congolese copper out through Angolan ports rather than South Africa, is itself a bet on the same decade. Mingomba is a long bet that Western capital, AI-driven discovery and a Zambian state partner can hold the line for a full decade of construction and ramp-up. Those parties do not have identical interests. The deal can survive that, but only if everyone keeps the math honest.Makayi closed her remarks with the line everyone took home. We must go fast. We shall go fast. We will not take shortcuts.The honest version of the question her project is asking is whether the algorithm's head start can hold for the eight years it will take to turn a probability surface into produced copper. If Mingomba lands on time and on budget in the early 2030s, AI mineral exploration earns its place as a commercial methodology and not just a research tool. If the schedule slips, the lesson for copper markets is that algorithms can find deposits faster than humans can mine them, and the bottleneck has simply moved underground.Either outcome is a story. Watch the engineering study in early 2027. Watch the shaft sinking. Watch the ZCCM-IH stake conversation. And keep the depth in mind: 1,700 meters. That is the part the model cannot do.</div></div>

廣發中證稀有金屬ETF

國泰中證有色金屬礦業主題ETF

南方中證申萬有色金屬ETF

有色金屬

國泰中證有色金屬ETF

KoBold Metals 正在贊比亞投資 23 億美元用於一個高品位銅項目，利用人工智能識別礦牀。該項目位於明戈姆巴，已從收購迅速推進到破土動工，預計將在 2030 年代初開始生產。然而，仍然面臨挑戰，包括工程複雜性和對冶煉解決方案的需求。該公司之前從未建造過礦山，最終的成本估算將在 2027 年公佈。項目的成功取決於銅礦牀的質量以及將要實施的工程解決方案的有效性

KoBold's $2.3 Billion AI Copper Bet Hits Rock In Zambia

福布斯

普信金融

標普全球

KoBold 在贊比亞的 23 億美元 AI 銅礦投資遇到了困難