Around $17 trillion in infrastructure investment is either deployed or committed over the next few years, global chairman of PwC, Mohamed Kande, told an audience at IFS Industrial X Unleashed in New York this week.
In a chat with IFS CEO Mark Moffat, Kande described today’s industrial boardroom as a place of two dominant emotions: excitement about what AI can unlock and genuine fear of missing the train or choosing the wrong technology stack.
The shift, both agreed, is that AI is no longer an IT side project. It has become a core capital-expenditure decision that will determine whether multi-billion-dollar assets remain competitive or turn into stranded white elephants over the next 20–30 years.
From “if” to “how” and fast
Twelve months ago, board-level AI conversations were still centered on back-office efficiency (finance, HR, IT), Kande said. Today, they are about operational reinvention: how to run plants, grids and fleets differently, not just cheaper, he added.
Kande calls this “industrial AI,” AI embedded so deeply into physical infrastructure that the software layer effectively becomes part of the asset itself.
He said that automated inspections, predictive maintenance and remote operations are now discussed in the same breath as capacity expansion and safety.
The stranded-asset risk is real and immediate
Industrial assets are designed for 20–30-year lives; frontier AI technologies evolve on much shorter cycles, Kande warned. Build a new plant or grid extension without an intelligent, upgradable layer, and it risks obsolescence long before the financing is paid off.
Kande cited his recent visit to a South Korean “dark factory,” fully robotic, lights on only when humans enter. His exact words to the audience: “The future isn’t coming – it’s already here.” Around 18% of industrial companies have moved AI into live operations, he said, and the rest are racing to catch up.
Beyond pilots: What the 5% who scale actually do
Moffat referenced research showing that most AI pilots fail, focusing instead on the small group of organizations that do scale successfully. According to the MIT study he cited, these successful projects share common traits: they are integrated into real workflows, they focus on narrow and high-value use cases, and they have clear paths from pilot to wider deployment.
Both speakers emphasized that boards are now asking for operational metrics and scale-up plans, not just demonstrations. The message was practical rather than promotional; a checklist of what works in the minority of cases that succeed.
“Jump In” but with controls
Kande advised directors to “jump in.”
“Whoever isn’t making mistakes right now isn’t trying hard enough.” Speed beats perfection, but governance cannot be an afterthought, he claimed, adding that in safety-critical environments, where a bad decision can stop a supply chain or endanger lives, that means ring-fenced experiments, full audit trails and pre-defined stop-loss triggers.
Vendor lock-in is another obvious boardroom concern, he says, hard-wiring a single proprietary stack into billion-dollar assets is a bet on skills availability, regulatory acceptance and the vendor’s long-term survival.
Mark added that industrial AI is for the majority of industrial workers who work on the shop floor, or up a tower, not at desks, and the primary driver is the augmentation of a stretched, ageing workforce rather than wholesale replacement.
The new boardroom imperative
For companies staring at trillions in committed capex, AI has become an infrastructure choice, not a software choice, the speakers stressed.
They emphasized that the question is no longer whether to embed intelligence, but which version of the intelligent layer they are prepared to live with for the next three decades.
Kande concluded: “You do not invest $17 trillion in a space without an outcome. Innovation is coming.”
