Each quarter I analyse enterprise AI spending patterns. Q2 2025 shows some meaningful shifts from earlier in the year. Here’s what I’m seeing.

The Headline Numbers

Based on conversations with finance and technology leaders across Australian enterprises:

Productivity tools: 45% of AI spend (up from 40% in Q1) Data infrastructure: 22% of AI spend (down from 25% in Q1) Custom AI development: 15% of AI spend (down from 20% in Q1) AI consulting and strategy: 18% of AI spend (up from 15% in Q1)

Total enterprise AI spend is up approximately 12% quarter-over-quarter.

Productivity Tools Continue Dominating

Microsoft Copilot spending continues to grow as rollouts expand from pilots to broader deployment. The additions in Q2:

More seats. Organisations that started with 20-30% of knowledge workers are expanding to 50-60%.

More products. Beyond M365 Copilot, organisations are adding GitHub Copilot, Copilot for Dynamics, and Copilot for Security.

Google catching up. Gemini for Workspace is gaining traction among Google-native organisations, though volumes remain smaller than Copilot.

The spending increase isn’t enthusiasm – it’s inertia. Once committed to productivity AI, organisations expand rather than evaluate alternatives.

Data Infrastructure Spending Plateaus

The foundational data work that dominated early 2025 budgets is stabilising:

Initial investments completing. Many data platform upgrades and governance implementations are finishing their first phases.

Maintenance mode emerging. Ongoing costs are lower than project costs. Spending shifts from capital to operational.

Value questions emerging. Some organisations are asking whether data infrastructure investments are delivering promised AI enablement. The connection isn’t always clear.

This doesn’t mean data work is done – most enterprises still have significant gaps. It means urgency has faded.

Custom Development Retreating

The most notable shift: organisations are building less custom AI.

What’s driving retreat:

• Platform capabilities have improved (why build what you can buy?)
• Early custom projects disappointed
• Skills constraints limit what’s achievable
• Maintenance burden of custom AI became clear

What’s still being built:

• Truly differentiated applications (industry-specific, proprietary data)
• Integrations and customisations on top of platforms
• Internal tools where off-the-shelf doesn’t fit

The “we should build our own” impulse has been replaced by “can we use what exists?”

Consulting Spend Grows

More money flowing to external help:

Implementation support. Organisations have tools; they need help deploying them effectively.

Change management. Realising that technology deployment isn’t the hard part, adoption is.

Strategy refinement. Year two of AI initiatives requires different strategy than year one.

Governance establishment. External expertise helping build sustainable governance.

This shift is healthy. It recognises that internal capability alone isn’t sufficient for most organisations.

Category-Level Details

Productivity AI

Spending profile:

• Copilot M365: $30-45 AUD per user per month
• GitHub Copilot: $25-30 AUD per developer per month
• Gemini for Workspace: $25-35 AUD per user per month

Deployment patterns:

• Typically 40-60% of eligible employees
• Higher adoption in professional services, lower in operations
• Power user programs emerging to drive engagement

ROI claims:

• Organisations claiming 10-20% productivity improvement
• Hard to verify – measurement remains challenging
• Some scaling back licenses for under-users

Data Infrastructure

Spending profile:

• Snowflake/Databricks: Variable, often $200k-$2M annually for large enterprises
• Data quality tools: $50-150k annually
• Data governance platforms: $100-300k annually

Investment patterns:

• Consolidation around major platforms continuing
• Data quality investment often underweight
• Integration costs often exceeding platform costs

Custom AI

Spending profile:

• Internal team costs: Variable, typically $500k-$2M annually for meaningful capability
• External development: $200k-$1M per significant project
• Platform costs: Variable based on usage

Investment patterns:

• Shifting from model development to application development
• More integration work, less algorithm work
• RAG implementations most common pattern

Consulting and Strategy

Spending profile:

• Strategy engagements: $150-500k depending on scope
• Implementation support: Variable, often $200k-$1M
• Governance establishment: $100-250k

Investment patterns:

• Shift from “what should we do” to “help us do it”
• More focused engagements, fewer broad assessments
• Growing demand for specialised AI implementation expertise from AI consultants Sydney

What’s Being Cut

Budget reallocation means something is losing funding:

Experimental projects. The “let’s try AI for X” initiatives without clear business cases.

Unused licenses. Organisations are getting serious about cutting licenses that aren’t being used.

Redundant tools. Consolidation is reducing spending on overlapping AI capabilities.

Generic AI training. “AI awareness” programs being replaced by role-specific capability building.

This reallocation is appropriate. Spending is becoming more disciplined.

Implications for Planning

Based on Q2 patterns:

1. Budget for productivity AI expansion. If you started pilots, plan for broader rollout budgets.

2. Don’t abandon data work prematurely. Spending may plateau, but gaps remain. Maintain investment.

3. Evaluate build vs. buy honestly. Custom development should be exception, not default.

4. Plan for external support. Internal teams need augmentation. Budget for it.

5. Cut the underperformers. Not every AI initiative deserves continued funding. Be willing to stop.

H2 Outlook

Looking ahead to the rest of 2025:

• Productivity tool spending will continue growing
• Custom development will further consolidate around platforms
• Governance and security spending will increase as regulatory pressure mounts
• Skills investment will grow as the constraint becomes clearer

Total AI spending will likely grow another 15-20% in H2, driven primarily by productivity tools and governance requirements.

The wild card: economic conditions. If belt-tightening becomes necessary, AI spending isn’t immune. ROI demonstration will determine what survives cuts.