The AI deals making headlines won’t drive Australian productivity. Here’s what will.

When Westpac CEO Anthony Miller stood on stage at Microsoft’s AI Tour in Sydney last month and described the bank’s 35,000-seat Microsoft Copilot rollout as “a bet on people,” he made one observation that the rest of the AI industry should pay close attention to.

“It only drives productivity when it’s diffused right across the community and right across society.”

Miller likened AI to electricity. The CEO of one of Australia’s largest banks, speaking at Microsoft’s own event, was telling them that 35,000 seats are not the answer. Diffusion is the answer.

He is right. And the question of how AI actually diffuses across the Australian economy, not just across the largest enterprise tier, is the most important question in AI right now. Most of the headlines coming out of Microsoft, Anthropic, and OpenAI miss it entirely.

What Westpac actually represents

Westpac’s deployment is a Microsoft win, and a real one. 35,000 employees plus contractors and service providers across Australia, scaled from a 15,000-employee pilot in 2025. It is the largest Copilot deployment in financial services in Asia Pacific. Microsoft 365 was already deployed across the bank, and adding Copilot was the path of least resistance.

Path of least resistance is exactly what it is. M365 incumbency at the F1000 tier means Microsoft wins enterprise AI deals through distribution leverage, not product superiority. The buyer already has the licensing relationship, the security review is partly done, the procurement path is short. Copilot wins by being already there.

That is defensible at the very top. Westpac is in the top half of Australia’s largest 50 companies by market cap. They have the budget, the in-house security team, the IT department, and the M365 deployment to make Copilot economically viable. The same dynamic plays out at CBA, NAB, ANZ, Telstra, BHP, and Rio Tinto. The top 50 to 100 companies in Australia.

It does not extend to the rest of the economy.

How Anthropic and OpenAI are routing around

If you cannot out-distribute Microsoft on M365 ground, you go around. That is exactly what Anthropic and OpenAI are doing right now, and the deals are public.

Anthropic is in talks with Blackstone, Hellman & Friedman, Permira, and General Atlantic to form a $1 billion joint venture, modelled on Palantir’s forward-deployed engineer playbook. Anthropic anchors with $200 million. The PE consortium contributes the rest. The venture exists for one purpose. To embed Claude into the operations of the PE firms’ portfolio companies, which collectively run into the thousands.

OpenAI’s parallel deal, internally known as DeployCo, is in advanced talks with TPG, Bain Capital, Advent International, and Brookfield Asset Management. Pre-money valuation of around $10 billion. The PE firms commit roughly $4 billion. They get equity stakes plus influence over how OpenAI’s technology is deployed across their portfolios. OpenAI is reportedly offering preferred equity with a guaranteed annual return to attract them. Combined, the four firms hold over 1,200 portfolio companies.

OpenAI’s enterprise platform, Frontier, anchors a separate programme called Frontier Alliances, which pairs OpenAI’s forward-deployed engineers with BCG, McKinsey, Accenture, and Capgemini for deployment into other corporate clients. Anthropic announced its own $100 million Claude Partner Network in March, anchored by Accenture, Deloitte, Cognizant, and Infosys.

Microsoft amended its OpenAI agreement on April 27 this year to formally remove OpenAI’s cloud exclusivity. OpenAI can now serve its products across any cloud provider. The structural lock that gave Microsoft an automatic distribution advantage on every OpenAI workflow is gone.

The pattern is clear. Anthropic and OpenAI are building parallel distribution that does not depend on Microsoft. PE consortia for direct enterprise embedding. Big Four implementation partners for Fortune 1000 clients. Embedded distribution through platforms like Xero, which has integrated Claude directly into its accounting workflow and reaches every Australian SMB and bookkeeper that uses it.

That is a lot of distribution. It is also entirely concentrated at the top of the market.

The gap Miller is talking about

None of these channels reach the 99 percent of Australian businesses that sit between 5 and 500 staff. The accounting firm in Subiaco with 80 employees. The mining services contractor in Karratha with 250 staff. The legal practice in Perth CBD with 45 lawyers. The construction firm in Welshpool with 120 employees and operations across three states.

These businesses are where the Australian economy actually lives. They employ most of the workforce. They contribute the bulk of national output. They are not on Microsoft’s enterprise sales call sheet, and they are not in any PE firm’s portfolio. They cannot economically deploy Copilot at $30 per user per month for a workforce of 80, because the productivity ROI math gets thin once you scale beyond power users.

They also do not have a Chief Data Officer. They do not have an in-house AI governance function. They do not have an SDLC team that can specify, test, deploy, and monitor an AI agent. They have a financial controller who knows Xero, an IT manager who knows Microsoft 365, and a CEO who reads Forbes and wants to understand what to do about AI.

This is the gap Miller was pointing at when he said productivity only comes when AI is diffused. None of the deals in the headlines reach this layer of the economy.

Where MSPs sit

Managed service providers are the only channel that does reach this layer of the Australian economy. We deploy IT, cybersecurity, and managed services across thousands of small and mid-market Australian businesses every day. We already sit inside their Microsoft 365 environments. We already manage their security stacks. We already know their data, their workflows, and their compliance obligations.

PwC’s recent global research on managed services partnerships frames the dynamic at the macro level. The top 20 percent of high-performing organisations in PwC’s survey of 2,000 business leaders are 4.2 times more likely to use managed services partners for strategic advantage. Companies that use MSPs strategically beat their cost-only counterparts by 43 percentage points of performance premium. They are 1.6 times faster to market and 2.4 times more innovative. Two thirds of the top performers in PwC’s research use managed services partners at the most mature levels.

PwC’s conclusion is direct. No company can go it alone. Top performers team up with managed services partners to address scarce talent and fast-moving technology, in support of the operating model changes that drive business model reinvention.

What does that look like for AI in practice? It looks like the boring infrastructure that decides whether AI is safe to use at all.

An AI acceptable use policy that staff actually follow because it is enforced by deny-by-default blocking on the network. Shadow AI discovery that maps every consumer AI tool currently in use across the business. A Microsoft 365 permissions audit before Copilot is deployed, because Copilot inherits whatever access already exists, and most businesses have permission gaps they do not know about. Data classification, sensitivity labelling, DLP policies that warn or block staff from pasting confidential information into AI platforms, and audit trails that satisfy compliance.

On top of that, an SDLC harness for AI agents. How they get specified, tested, deployed, monitored, and rolled back. How they integrate with the workflow software the business already runs. How they get governed when the next model release changes their behaviour.

This is the work that determines whether a 100-staff Australian business gets real AI productivity or a flashy demo that goes nowhere. It is the work that no PE forward-deployed engineer will do for a sub-$10 million revenue client. It is the work that no Big Four implementation partner will pitch outside Fortune 1000. It is the work MSPs do every day.

The complexity of delivering this layer well also tells us something about how the MSP industry itself will change. AI governance is not a free add-on. Building deny-by-default network controls, M365 permissions audits, DLP policy engines, sensitivity labelling, SDLC harness for AI agents, and the team capability to deploy and monitor it all costs real money. Smaller MSPs without the scale to invest cannot credibly deliver this layer. The Australian MSP industry will consolidate as a result, and the consolidation will be driven by AI capability gaps as much as by cost synergies. The same PE roll-up pattern playing out at the LLM provider tier will play out at the MSP tier directly below it, because the structural logic is identical. Distribution into the mid-market needs scale.

What we expect to see next

Four predictions, all falsifiable inside 18 months.

First, at least one Australian PE firm will announce its own LLM partnership. Quadrant Private Equity, BGH Capital, Pacific Equity Partners, Mercury Capital. Any of these has the portfolio scale to make the same calculation Blackstone and TPG just made. Watch for the first announcement, probably in the second half of 2026.

Second, Microsoft will lose material Australian enterprise share in regulated industries that prioritise data privacy, even with M365 incumbency. Health, legal, financial services outside the big four banks, and government will be the first to move. The Anthropic dispute with the US Department of Defense over military safeguards is not a story here, but the underlying preference for vendors with stronger data ethics positioning will be.

Third, the MSP layer will become the dominant deployment channel for AI in Australian SMB and mid-market by mid-2027. The first Australian MSP to publish a formal, branded AI governance framework as intellectual property will earn lasting market positioning. Expect that announcement before the end of this year.

Fourth, expect material consolidation in the Australian MSP space inside the next 24 months. Mid-size MSPs without the capital or technical depth to build AI governance and SDLC capability will be acquired by larger MSPs and PE-backed MSP roll-ups. The first Australian MSP acquisition explicitly framed as an AI capability play probably happens in the first half of 2027.

Diffusion is where productivity actually happens

Anthony Miller is right. AI only drives productivity when it is diffused across the community and across society. Westpac’s 35,000 Copilots are necessary but not sufficient. The Anthropic and OpenAI PE plays will reach thousands of portfolio companies, which is real progress, but still only the upper tier of corporate Australia.

Diffusion happens in the layer below. The accounting firms, the construction businesses, the medical practices, the engineering consultancies, the legal practices, the manufacturers, the agribusinesses. The businesses where most Australians work. The businesses that make most of the national output. They are too small for Microsoft’s direct enterprise team, too obscure for Big Four implementation partners, too thin to economically deploy Copilot at scale, and not in any PE portfolio.

This is the Australian AI productivity question. It will be answered by the layer that already operates inside this segment of the economy. That is the MSP layer.

We are building it because Miller is right, and because someone has to.

If you are navigating AI adoption in your business and want to understand what governance, deployment, and managed AI services actually look like in practice, talk to us on 1300 EPIC IT. We have a full breakdown of our AI services, and our AI Governance and Managed AI service pages cover the detail.