Key Takeaways:

• Global markets are arguing about whether AI is a bubble. For most WA organisations, the more useful question is why so much AI spend produces nothing.
• MIT's 2025 research found 95% of enterprise generative-AI pilots deliver no measurable impact on profit and loss. Only 5% do.
• The single biggest predictor of success: projects delivered with a specialist partner succeeded ~67% of the time, versus ~33% for internal builds.
• The measurable returns are showing up in back-office and operational workflows exactly where connected systems matter, not flashy front-office tools.
• In WA, AI is already delivering hard ROI in the Pilbara, and WA Government has stood up its own AI Assurance Framework and AI Advisory Board.
• Dapth view (As a perth based digital consultancy) on whether the bubble pops or not, the winners treat AI as an operational-integration problem, not a technology-purchase problem.

The story everyone's telling right now

In June 2026, Apple did something it had never done mid-cycle: it raised prices on hardware, pointing to surging memory-chip costs driven partly by the AI infrastructure boom. Around the same time, the four largest US tech companies were on track to spend hundreds of billions on AI data-centre capacity in a single year, and commentators from Ray Dalio to Michael Burry were reaching for the word "bubble."

It's a genuinely interesting macro debate. It's also almost entirely irrelevant to whether your organisation gets value from AI.

Here's the honest position: nobody credible knows if it's a bubble. The technology is real, the revenue is real, and today's tech valuations are nowhere near the extremes of the 2000 dot-com peak, but the spending is running far ahead of the revenue AI currently generates. Anyone who tells you with certainty that it's a bubble is guessing. So is anyone who tells you it definitely isn't.

For a WA business owner, a government agency, or an operations leader, betting on the macro is a waste of energy. There's a far more useful question hiding underneath it.

What's the real question WA organisations should be asking?

Not "will the AI bubble burst?" but "why do most organisations spend money on AI and get nothing back, and how do I avoid that?"

That question has an actual, evidence-based answer. And it's the one almost nobody is putting in front of Western Australian businesses.

How often does enterprise AI actually fail to deliver ROI?

Far more often than the hype suggests. In August 2025, MIT's NANDA initiative published The GenAI Divide: State of AI in Business 2025 based on structured interviews with 52 organisations, a survey of 153 senior leaders, and analysis of 300 public AI deployments representing an estimated US$30–40 billion in pilot investment.

The headline finding: roughly 95% of enterprise generative-AI pilots delivered no measurable impact on profit and loss.Only about 5% achieved rapid, measurable value. (MIT NANDA, via Fortune)

Crucially, MIT's researchers were explicit that the failures were not primarily about the technology. The models work. Pilots stalled because tools couldn't retain feedback, adapt to a specific workflow, or integrate with the systems and data the business actually runs on. MIT called this the "learning gap." (The report was preliminary rather than peer-reviewed and drew some methodological criticism, but its central finding is directionally consistent with Gartner’s data and, as we’ll see, with what’s already happening in the Pilbara.)

Other research points the same way. Gartner has flagged that a large share of agentic-AI projects stall before production; consulting analyses repeatedly find that the minority seeing substantial returns are the ones who integrated deeply rather than bolting a chatbot onto an existing process.

Why do so many AI projects fail? Three patterns from the data

The MIT study, read alongside the broader evidence, surfaces three recurring reasons, and all three are fixable.

1. The money goes to the wrong place. More than half of enterprise generative-AI budgets went to sales and marketing tools, yet MIT found the biggest, most measurable ROI in back-office and operational automation cutting outsourcing, streamlining process-heavy workflows, reducing manual handling. The visible, front-office use cases attract the budget; the unglamorous operational ones deliver the return.

2. Build-it-yourself quietly fails more often. This is the finding WA leaders should sit with. In MIT's data, AI initiatives delivered through specialist external partners succeeded roughly 67% of the time, while internal builds succeeded only about a third as often. Going it alone doubled the failure rate. Enterprises consistently underestimate the integration, data, and change-management work that sits between a working demo and a production system.

3. Tools sit outside the workflow. Generic assistants are brilliant for individuals and hopeless for enterprises that need context, memory, and system integration. MIT found over 90% of employees already use personal AI tools at work ("shadow AI") even where official pilots failed proof the appetite is there, but that unmanaged, disconnected adoption isn't the same as organisational ROI.

Put simply: AI doesn't fail in the lab. It fails at the point where it collides with vague goals, disconnected systems, and messy data. That's an operational-integration problem not a technology problem.

Does any of this apply to Western Australia specifically?

Yes, and the WA picture is more advanced than the "wait and see" mood suggests.

Adoption is real but uneven. Nationally, the National AI Centre's monthly SME tracker found about 43% of Australian SMEs reported some level of AI adoption across the December 2025–February 2026 quarter and, tellingly, that number has plateaued rather than surged. Among businesses not adopting, around 65% cited distrust of AI decision-making or a preference to keep humans in control, and over half said AI simply "isn't relevant" to them. (National AI Centre) Trust and relevance not capability are the real barriers.

Perth is on the national AI map. CSIRO and the National AI Centre's ecosystem analysis identified Perth as one of Australia's significant AI company clusters, alongside Melbourne, Sydney and Brisbane. (National AI Centre)

WA Government has already built the guardrails. The WA public sector has a Government AI Policy and Assurance Framework, a WA Artificial Intelligence Advisory Board, and sector-specific guidance including a WA Health AI policy updated in February 2026 and the Department of Education's EdChat tool. (SafeAI-Aus resource index) For WA agencies and the businesses that serve them, AI governance is no longer optional or theoretical.

And in WA's biggest sector, AI is already paying its way. The Pilbara is arguably the clearest example on earth of AI delivering hard operational ROI. Rio Tinto's AutoHaul runs the world's first fully autonomous heavy-haul rail network around 50 driverless trains monitored remotely from Perth. BHP runs autonomous trucks and drills across its WA iron-ore operations and centralises operations from a Perth remote-operations centre; Fortescue operates a large autonomous truck fleet. Rio Tinto even partnered with BCG X on an AI scheduling platform run by its Perth-based integrated scheduling team. (S&P Global; BCG)

The lesson from the Pilbara isn't "buy AI." It's that those returns came from AI embedded into the operational core connected to equipment, data, and workflows not from a pilot running on the side. That's the exact pattern MIT found separates the 5% from the 95%.

What happens to your AI investment if the bubble does burst?

This is where the macro debate loops back and becomes practical. There are broadly two futures, and it's worth planning for both.

If the bubble deflates, capital tightens and a lot of thinly-integrated AI products disappear. The organisations left standing will be the ones whose AI is woven into operations and already producing measurable value because that's the spend nobody cuts.

If it doesn't burst, the pressure on AI providers to finally turn a profit means token and compute costs are likely to rise. Cheap experimentation gets more expensive. Again, the organisations that win are the ones who built for efficiency and real ROI rather than open-ended experimentation.

Either way, the strategic conclusion is identical: treat AI as an operational-integration decision, tie every use case to a measurable outcome, and don't try to do the hard integration work alone. That's a robust position under both scenarios which is exactly what you want when the macro is genuinely uncertain.

Dapth view as a Perth AI Consultant 

The bubble question makes headlines. The ROI question drives business outcomes.

For most WA organisations, the real risk isn’t that AI is overhyped. It’s investing in pilots that never deliver measurable business value just as research suggests happens with the majority of enterprise AI initiatives.

The organisations seeing real returns share three common traits. They focus AI on operational and back-office workflows where outcomes are measurable. They integrate AI into existing systems and processes rather than running disconnected pilots. And they work with experienced specialists to bridge the gap between a proof of concept and a production-ready solution.

That’s the principle behind Dapth’s approach to AI. The value isn’t in the tool itself it’s in how well it’s integrated into your business. If you’re exploring where AI could deliver measurable returns across your operations, we’d be happy to help you identify the opportunities worth pursuing, and the ones that are likely to waste time and budget.

Sources

1. MIT NANDA, The GenAI Divide: State of AI in Business 2025 reported via Fortune, Aug 2025. fortune.com
2. National AI Centre (Australia), AI adoption insights: December 2025 to February 2026. ai.gov.au
3. National AI Centre / CSIRO, Australia's artificial intelligence ecosystem: growth and opportunities. ai.gov.au
4. SafeAI-Aus, State & Territory AI Resources (index of WA AI Policy & Assurance Framework, WA AI Advisory Board, WA Health AI policy, EdChat). safeaiaus.org
5. S&P Global Market Intelligence, A peek at AI revolution in mining. spglobal.com
6. BCG X, How an iron-ore producer modernised mining operations with AI (Rio Tinto, Perth). bcg.com