By the time Avathon announced its deal with Barrick on September 23, its software was already running at some U.S. sites, including Nevada Gold Mines. That is what The Information reported on September 22, citing Avathon's chief executive.
The same report says Barrick's North American business agreed to a five-year partnership. A person with direct knowledge of the discussions estimated it would generate between $10 million and $20 million a year in sales for Avathon.
That is an estimate of Avathon's revenue. Barrick has disclosed no contract price of its own, and we still need to see which mine results would pay for the software.
Barrick expects to float a minority stake in its North American gold business before the end of 2026, according to its August 10 results. We want the operating case before anyone asks us to value the technology inside that offering.
The Rollout Has Already Begun
Avathon's September 23 announcement names the work it wants its platform to do: spot hazards, connect ore flow with processing decisions, predict equipment failures, coordinate spare parts and improve mine planning. Barrick's North American staff, the release says, retain operational judgment and control.
In that announcement, Barrick chief executive Mark Hill put the ambition this way:
AI has the potential to change how a mining company safely operates at its core.
The release gives no measured change in recovery, downtime or safety, and no Barrick-disclosed spending figure.
The Information says Avathon and Barrick began working together in December, with software at some U.S. sites since the second quarter. It says the Nevada Gold Mines rollout is expected to finish over the next couple of quarters, and that the Dominican Republic is next, according to Avathon chief executive Pervinder Johar.
Those rollout dates come from Johar's interview. Barrick's September announcement sets no site-specific milestones. But a system already running at Nevada gives owners a nearer test than a distant research project: can a mine manager point to an operational result?
A camera that flags an unsafe condition has to lead to an intervention. A maintenance model that predicts a failure has to get a crew and a part to the equipment before it stops. At a mill, better information has to change a processing decision and help recover more gold from the ore already delivered.
Each link can fail even if the software identifies the problem correctly. A list of applications cannot show whether a repair got faster or recovery improved at a named mine.
A predicted failure may help only if the maintenance crew has time to act, the needed part is in stock, and the mine can make the repair without moving the stoppage to another day. We want a record of completed actions, not just a count of alerts.
The Price Has Two Layers
The Information reported on September 22 that Barrick will pay a base fee for Avathon's platform, plus fees tied to results. Its estimate of $10 million to $20 million in annual sales comes from a person with direct knowledge of the discussions, not from a Barrick contract disclosure.
The report does not define which outcomes trigger the performance fee. Would a predicted repair earn a payment when a warning appears, when a machine stays running, or when the mine produces more?
Nor does Avathon's September 23 release give the exact fee Barrick will pay, the cost of putting the platform into each mine, or a measured benefit. The Information does not say how much of Avathon's estimated annual sales would come from the base fee or from payments linked to results.
Ore grade, mining areas and plant availability can change during a software rollout. A higher grade or a completed repair can lift recovery or output without help from a new model.
Barrick credited stronger second-quarter production partly to the ramp-up at Cortez in its August 10 results. It said lower grades at Carlin and Cortez were among the reasons gold costs rose from a year earlier. Calling the next improvement an AI gain without separating such changes would tell us little.
We would ask for the date a system went live at each operation, the operating measure it was meant to change, and the result beside the same mine's earlier record. For equipment, that might be unplanned downtime. For processing, it might be recovery under comparable ore conditions. For safety, the record must show incidents and interventions, rather than incidents a model says it might have prevented.
Without those comparisons and the actual spending, even a reported annual vendor revenue range cannot yield a credible payback period for Barrick.
If Avathon gets paid for an avoided shutdown, someone must define what would have happened without its alert. If it gets paid for extra recovered gold, the comparison must account for changes in ore grade and plant settings. We cannot put an estimate of savings in the same column as actual mine cash flow and call the difference a return.
A Company Average Cannot Prove a Mine Result
Barrick produced 796,000 ounces of gold company-wide in the second quarter of 2026, with all-in sustaining costs of $1,866 per ounce, according to its August 10 results.
They are not a baseline for Nevada Gold Mines, let alone a measure of what Avathon has changed there. Barrick's quarter includes operations outside the proposed North American listing. Its company-wide cost figure reflects far more than any one technology contract.
Barrick posts a separate mine statistics supplement for the second quarter alongside its August 10 results. Those mine figures give us a better starting point than the group average, but they predate a completed Nevada rollout.
A change at Nevada would be more useful if Barrick disclosed the measure at that operation before and after deployment. One mine's mill recovery might improve while its ore grade changes, or its downtime might fall after a scheduled overhaul. We would want management to explain those changes alongside the software's deployment date. Otherwise a group average can move in the right direction while leaving the cause hidden.
If you own B in New York or ABX in Toronto, the tickers in Avathon's September 23 release, the upside is more recovered gold, fewer stoppages or safer work at the mines you own. The risk is that an estimate of a vendor's sales gets treated as evidence of a miner's future cash flow.
Our earlier piece, Agnico Just Said No to Barrick, followed the planned North American IPO. We cannot add an AI premium to those mines without a measured operating gain.
Barrick's August 10 release said the IPO remained on track for completion by year end. Before December 31, we will check the next quarterly operating release and any offering document for the deployment scope, spending terms and site-level measures. If those details are absent, we will keep valuing the business through the ounces, costs and cash flow Barrick actually reports.
We will watch the Nevada Gold Mines rollout over the next couple of quarters and keep B and ABX tied to Barrick's reported operating results. We will give the company no credit for a recovery gain until it measures one.
The model can find a fault. The mine has to deliver the gain.
Seek the truth and be prepared,
Equedia
Sources
- Avathon, Selected to Power an AI-Native Mining Operating Model for Barrick's North American Business
- The Information, Barrick Gold Inks Five-Year AI Deal Before IPO
- Barrick Mining Corporation, Second Quarter 2026 Results
- Barrick Mining Corporation, Second Quarter 2026 Mine Statistics
- The Equedia Letter, Agnico Just Said No to Barrick
Disclaimer: This letter is for informational and educational purposes only and does not constitute investment advice. Past predictions and performance are not indicative of future results. Please see our full terms of use and disclaimer at equedia.com.
