Every quarter, I look at where enterprise AI budgets are actually being deployed versus where the conference talks suggest they should be. Q1 2025 shows some interesting divergences.

The Headline Numbers

Based on conversations with CFOs and technology leaders across Australian enterprises, here’s where AI budgets are landing:

Productivity tools: 40% of AI spend Data infrastructure: 25% of AI spend Custom AI development: 20% of AI spend AI consulting and strategy: 15% of AI spend

These numbers have shifted notably from a year ago. Let me break down what’s happening.

Productivity Tools Dominate

Microsoft Copilot is eating a huge portion of enterprise AI budgets. The combination of familiar interfaces, existing Microsoft relationships, and broad applicability makes it the default choice for many organisations.

What I’m hearing:

• Large enterprises are rolling out Copilot to 20-50% of knowledge workers
• Monthly costs of $30 AUD per user are being accepted without much resistance
• Productivity gains are hard to measure but broadly felt
• The alternative of building similar capabilities in-house is seen as unrealistic

This isn’t necessarily optimal resource allocation. It’s the path of least resistance – which, in enterprise procurement, is often the path that gets chosen.

Data Infrastructure Gets Real Investment

The second-largest spending category is less visible but arguably more important. Enterprises are finally investing seriously in the data foundations that AI requires.

Specific investments:

• Data quality initiatives (deduplication, standardisation, enrichment)
• Modern data platforms (Snowflake, Databricks adoption continues)
• Data governance tooling
• Integration work to connect siloed systems

This is the unsexy spend that enables everything else. I’m encouraged to see it getting proper budget attention rather than being treated as an afterthought.

Custom Development Plateaus

Here’s an interesting shift: custom AI development spending has plateaued. A year ago, everyone wanted to build bespoke AI solutions. Now, the appetite for custom development has cooled.

Why? Several factors:

• Platform capabilities have improved (Bedrock, Azure AI, etc.)
• The cost and complexity of custom LLM work became clearer
• Pilot projects produced mixed results
• Maintenance burden of custom AI is significant

The enterprises doing custom development now are more targeted: specific use cases where off-the-shelf doesn’t fit, not general “we should build AI” initiatives.

Consulting Spend Patterns

AI consulting spend breaks into two categories:

Strategy work: Helping organisations figure out where AI fits, building business cases, governance frameworks. This is steady but not growing dramatically.

Implementation help: Hands-on assistance getting AI deployed. This is where consulting spend is increasing. AI consultants Sydney are seeing strong demand as enterprises move from strategy to execution.

The shift from “help us think about AI” to “help us deploy AI” reflects maturing understanding. Organisations know what they want to do; they need help actually doing it.

Australia vs Global Patterns

A few Australia-specific observations:

More conservative adoption. Australian enterprises are spending less on AI as a percentage of IT budget than US counterparts. The gap is narrowing but still significant.

Stronger focus on proven use cases. Less experimentation, more following established patterns. Australian CIOs want to see something work elsewhere before committing budget.

Data sovereignty concerns shape choices. More investment in on-premises or Australian-hosted solutions than you’d see in the US. This adds cost but addresses compliance requirements.

Skills investment lagging. Training and upskilling spend is lower than it should be. Enterprises are buying tools without adequately preparing people to use them.

What the Numbers Don’t Show

Spending data doesn’t capture everything important:

Opportunity cost of slow adoption. Enterprises that haven’t invested in AI may be losing competitive ground in ways that don’t show up in budgets.

Hidden AI spend. Shadow IT spending on AI tools (personal ChatGPT subscriptions, departmental purchases) isn’t captured in official numbers.

Productivity gains reinvested. Some enterprises are funding AI expansion from productivity gains rather than new budget – making the official spending look smaller than the actual activity.

Predictions for Q2-Q4

Based on current patterns, I expect:

• Productivity tool spending to continue growing as Copilot rollouts expand
• Data infrastructure spend to hold steady – the foundational work takes time
• Custom development to remain flat unless specific high-value use cases emerge
• Consulting spend to shift further toward implementation support

The overall AI budget envelope will likely increase 15-20% through 2025, driven primarily by productivity tools and data platform costs.

What This Means for Your Planning

If you’re planning AI investments:

1. Don’t underestimate productivity tools. Copilot and similar may be the highest-ROI AI investment you can make, even if it’s not the most interesting.

2. Invest in data foundations. Without good data, everything else is compromised. Budget for the unsexy work.

3. Be realistic about custom development. Build custom only when you have a specific, differentiated use case that platforms can’t address.

4. Budget for implementation support. Internal teams are often stretched. External expertise accelerates time-to-value.

5. Track actual spend. Unofficial AI spend can be significant. You can’t manage what you don’t measure.

The AI spending conversation has matured from “how much should we spend” to “what should we spend it on.” That’s progress. Now comes the harder work of actually delivering value from those investments.