Everything About AI
NextFin News - Pervinder Johar, chief executive of industrial AI company Avathon, said in a recent interview on the Tech Disruptors program that the next phase of artificial intelligence will not be won by chatbots but by software that can act inside the physical operations of energy grids, mines, factories and logistics networks. The claim is specific: Avathon's Autonomy Platform does not stop at prediction or analytics; it takes autonomous action across critical infrastructure, and its newest commercial contracts tie part of the company's compensation to measurable operational outcomes.
The timing matters. The interview, published November 3, 2026, arrives weeks after Avathon announced a multi-year agreement with Barrick's North American gold business to deploy its Physical AI platform across safety, production, asset reliability, supply chain, mine planning and exploration — a deal that puts the company's "autonomous operations" thesis to its most visible test yet. It also follows a December 2025 expansion of Avathon's partnership with Google Cloud, which integrated the platform with Vertex AI and Gemini Enterprise for the energy sector.
The Situation: From Digital Mining To An AI-Native Operator
Avathon is not a household name in the way that the large language model vendors are, but its footprint in heavy industry has widened quickly through 2026. The Pleasanton, California-based company, founded in 2013, is privately held with a $1.4 billion valuation set in its January 2022 Series D, according to private-market data providers. Its investor roster — Temasek, March Capital, Verizon Ventures, CME Ventures, Energize Capital and Boeing among them — reads like a roll call of strategic capital from the industries it is trying to transform.
The core product is the Avathon Autonomy Platform, built around what the company calls a Computational Knowledge Graph: a layer that connects assets, processes, people, constraints and operational data, then lets AI agents reason across that context to support decisions and coordinate workflows. Unlike a dashboard that alerts a human, the platform is designed to close the loop — generating work orders, adjusting dispatch, automating compliance tasks and, in defined domains, taking action without human intervention. It plugs into existing industrial systems, including SCADA, weather feeds, market data, compliance records and work-order systems.
The Barrick deal, announced September 23, 2026, is the clearest window into how this works in practice. Barrick's North American business selected Avathon as a strategic technology partner to build an AI-native mining operating model across its gold assets. Initial applications span six areas: safety monitoring through computer vision; production and recovery optimization from mine to mill; asset reliability and predictive maintenance; supply chain intelligence; mine planning; and exploration. Avathon's platform has been operational since the second quarter of 2026 at several U.S. sites, including Nevada Gold Mines, with full deployment across the Nevada complex expected within one to two quarters of the announcement and a subsequent rollout to Barrick's Dominican Republic operations in the pipeline. Analysis of the agreement describes it as a five-year arrangement covering roughly 200 AI models, with a base fee for platform use plus additional performance-based fees tied to the operational outcomes actually achieved; no dollar values have been publicly disclosed.
"Mining has historically been organized around individual functions, systems and processes," said Pervinder Johar, Avathon's chief executive. "The opportunity now is to connect those pieces into one intelligent operating system. Barrick's North American business is taking a fundamentally different approach by putting AI at the center of how the enterprise understands its operations, makes decisions and acts. This is the evolution from digital mining to an AI-native mining company, where human expertise and Physical AI work together across the entire value chain to drive safer operations, higher productivity, greater recovery and stronger returns."
Two features of the arrangement are worth flagging. First, the performance-linked fee structure transfers part of the execution risk from the operator to the software vendor — a rarity in mining technology contracts at this scale. Second, despite the "autonomous" branding, the release is explicit that Barrick's mining professionals retain operational judgment, accountability and control, backed by governance safeguards. The pitch is AI plus human intelligence, not lights-out replacement — at least for now.
Why This Is Different From The Last Industrial AI Cycle
Industrial AI has a credibility problem earned over two decades. The sector has watched successive waves — expert systems in the 1990s, predictive analytics in the 2010s, digital twins in the 2020s — arrive with grand promises and settle into pilot purgatory, where proofs of concept never graduate to enterprise-wide deployment. The natural question is whether Avathon is selling another dashboard with a better name.
Three things argue that this cycle is structurally different, though none is decisive on its own.
First, the technology stack has crossed a threshold. Agentic AI — models that can plan, call tools and execute multi-step workflows — combined with foundation vision and language models gives software a way to act, not just report. Avathon's integration with Google Cloud's Vertex AI and Gemini Enterprise is a concrete example: anomaly detection, predictive maintenance and energy forecasting are no longer static models but agents that can automate complex workflows across geographically dispersed assets. The computational knowledge graph matters here because it supplies the context that generic models lack; an AI agent that knows the difference between a compressor station constraint and a maintenance-window constraint can act safely in a way a general-purpose model cannot.
Second, the commercial model is shifting from license fees to outcome-based pricing. The Barrick structure — a base fee plus additional compensation contingent on measurable operational gains — aligns the vendor's incentives with the customer's P&L and lowers the barrier to adoption for operators who have been burned by seven-figure software projects that never moved a metric. But it also changes the vendor's economics: a meaningful share of revenue becomes contingent on measurable operational gains, which compresses margins during ramp-up and demands deep integration into the customer's workflow. This is not a land-and-expand SaaS motion; it is a partnership that requires the vendor to understand the customer's business as well as the customer does.
Third, the installed base of physical automation is large enough to absorb AI retrofits without waiting for a robot replacement cycle. The worldwide operational stock of industrial robots reached 5.079 million units in 2025, with 603,000 new installations during the year, an 11% annual increase, according to industry data compiled by Globe Market Research. Amazon operates more than 1 million robots across its fulfillment network. Avathon's approach targets the software and decision layer above that hardware — the scheduling, dispatch, maintenance and planning logic that determines whether a million robots, or a fleet of wind turbines, or a mining complex actually performs near its theoretical capacity.
The counter-argument is serious and should not be waved away. Public industrial AI vendors are still struggling to prove profitability. C3.ai, one of the closest public comparables, reported fiscal third-quarter 2026 revenue of $53.3 million but a GAAP gross margin of only 17% and a GAAP net loss of $0.94 per share. If a scaled public player cannot turn industrial AI into durable profit, a private company still building toward scale faces an uphill climb. The gap between "autonomous action" in a press release and audited, recurring, high-margin revenue is where most industrial AI theses have gone to die.
The Second-Order Consequence: Whoever Maps The Asset Class Owns The Workflow
The first-order read of the Avathon story is straightforward: autonomous operations reduce downtime and lift productivity. The second-order question is less discussed and more important — who captures the value as this layer becomes the operating system for physical industries?
The answer points to the knowledge graph, not the model. Foundation models are becoming commoditized; the durable moat in industrial AI is the structured representation of a specific asset class — what a SAG mill's throughput depends on, how a gas turbine degrades, which maintenance crew has which skills, what the compliance calendar requires. Once that graph is built and wired into daily decisions, switching costs are enormous. The vendor that maps an industry first does not just sell software; it becomes the system of record for how that industry operates. That is why Avathon is moving sector by sector — energy, then mining, then logistics and aerospace — rather than trying to be a horizontal platform from day one.
This has cross-asset implications. If autonomous operations prove their value, capital expenditure in heavy industry gradually shifts from incremental headcount and reactive maintenance toward software, compute and data infrastructure. Operating margins for energy and mining companies could expand if downtime and recovery variability actually fall — Avathon claims up to a 17% reduction in operating costs, up to 20% less unplanned downtime and a 4% to 9% boost in annual revenue for renewable energy operators, though these are company-stated figures rather than independently audited results. Conversely, vendors that fail to demonstrate measurable outcomes under performance-linked contracts will find their revenue contingent and their funding windows narrow.
The market backdrop is expansive but definitions vary wildly. Globe Market Research projects the physical AI market at $115.9 billion in 2026, growing to $1.01 trillion by 2030 and $2.13 trillion by 2035 — a figure that sweeps in industrial robotics, autonomous vehicles, humanoids and healthcare systems. MarketsandMarkets uses a narrower definition, forecasting $1.50 billion in 2026 rising to $15.24 billion by 2032 at a 47.2% compound annual growth rate. The spread between the two estimates is itself instructive: "physical AI" is still a category being invented, and the number an investor attaches to the opportunity depends entirely on which slice of it they count.
The Adversarial Case: Autonomy Is A Spectrum, Not A Switch
The strongest challenge to Avathon's thesis is not that the technology will fail, but that "autonomy" will prove to be a spectrum that stalls well short of the lights-out vision. The Barrick release itself concedes that humans retain operational judgment and control. In safety-critical industries — mining, energy, aviation — regulators, insurers and boards will not hand decision rights to agents quickly. The realistic endpoint for this decade is decision support that automates the routine and escalates the exceptional, not fully autonomous operations.
History is the second, heavier line of attack. Comparable autonomy deployments in mining have taken the better part of a decade to reach full scale: Rio Tinto's AutoHaul autonomous rail program and BHP's Jimblebar autonomous haulage system are the precedents, and analysis of those programs places material financial benefits from Barrick's program more realistically in the 2028-2030 window rather than in near-term quarterly results. If that timeline holds, Avathon's five-year contract may be mostly infrastructure-building through 2027, with the economics that justify the valuation arriving later than the market expects.
A third line of attack runs through the economics. Outcome-based pricing is attractive to customers precisely because it transfers risk to the vendor, but vendors that accept that risk need balance sheets deep enough to absorb periods when a customer's operations underperform for reasons outside the software's control — ore grade variability in mining, weather in renewables, commodity prices in logistics. Avathon's $1.4 billion valuation and private-market funding give it room to experiment; a public-market version of the same model would face quarterly scrutiny that outcome-based revenue does not fit neatly.
There is also the competition problem. Avathon is not alone. C3.ai has pushed hard into manufacturing and defense; hyperscalers are building their own vertical AI stacks; and a new generation of physical AI startups — Skild AI raised $1.4 billion in January 2026 to build general-purpose intelligence for controlling robots; Agility Robotics reported more than $300 million in multi-year orders for its Digit humanoid — are attacking adjacent layers of the same opportunity. In mining specifically, the vendor that wins is not necessarily the one with the best model but the one that survives the longest sales cycle and integrates deepest into legacy systems.
What To Watch: The Falsifying Signals
The thesis that Avathon is leading a structural shift from predictive analytics to autonomous operations rests on observable milestones, not rhetoric. The most important near-term signal is Barrick's operational reporting. Barrick's fourth-quarter 2026 and first-quarter 2027 operational reports are the earliest windows where throughput, recovery and safety data from the Avathon deployment could appear. If those reports show no measurable improvement attributable to the platform — or if the full Nevada Gold Mines deployment slips beyond the one-to-two-quarter window the companies indicated — the "AI-native operator" narrative loses its strongest evidence.
At the company level, revenue scale and margin trajectory matter. Avathon's private-market funding and $1.4 billion valuation give it runway, but the path from here to a profitable public business is narrow. If Avathon cannot demonstrate a clear path to the kind of gross margin that public industrial AI vendors have struggled to reach, and if its revenue growth stalls while performance-linked contracts compress margins, the structural thesis weakens toward the cyclical "pilot purgatory" outcome that has claimed earlier waves.
Broader category signals cut the same way. If the physical AI market grows toward the conservative MarketsandMarkets trajectory rather than the expansive Globe Market Research one, the opportunity is real but smaller and more contested — a winner-take-most software market becomes a fragmented services market. Investors should watch whether outcome-based contracts spread beyond Avathon or remain a negotiating concession; if competitors refuse to follow, the model may be a sales tactic rather than a durable pricing regime.
Conclusion: A Structural Shift With A Cyclical Gate
The central judgment: Avathon is riding a structural change, not a cyclical wave. The shift from AI that predicts to AI that acts inside physical operations is a regime change in industrial software, driven by agentic models, knowledge-graph context and outcome-based commercial structures that did not exist in prior cycles. That shift will not revert on its own; once an operator's decisions are orchestrated by a software layer, the workflow does not go back to spreadsheets and siloed systems.
But the structural leg sits on top of a cyclical adoption gate. Deployment in heavy industry moves in multi-year waves tied to capital budgets, commodity prices and labor markets. The next 12 to 18 months will separate vendors with deployed, measurable outcomes from vendors with demos. Avathon has accumulated credible evidence inside roughly a year: a multi-year Barrick commitment with a platform already live at Nevada Gold Mines, a Google Cloud integration for energy, a strategic partnership with Aramco Digital to accelerate industrial AI in Saudi Arabia, a Bengaluru facility expansion and an Avathon Physical AI Lab planned with IIT Roorkee.
Scenario map:
- Base case: Avathon delivers measurable gains at Barrick by 2027, outcome-based pricing spreads to two or three more anchor customers, and the company approaches a public listing or a strategic sale at a premium to its $1.4 billion valuation.
- Upside case: the Barrick deployment becomes the reference architecture for mining and energy, the knowledge graph becomes the industry operating system, and Avathon re-rates into the top tier of industrial AI platforms alongside or above the public comparables.
- Downside case: deployment slips, reported operational gains are muted, outcome-based contracts compress margins, and Avathon joins the long list of industrial AI companies that raised heavily but never proved scalable profit.
The short-term trade is sentiment and funding headlines; the medium-term trade is deployed revenue and gross margin; the long-term trade is whether the knowledge graph becomes the operating system for physical industries. Those are three different investments, and they will not all pay off on the same timeline.
The bottom line: Avathon's bet is that the value in industrial AI has moved from the model to the workflow — and the Barrick deal is the first real test of whether a software company can get paid for outcomes rather than promises.
Explore more exclusive insights at nextfin.ai.
