We’re halfway through 2025, which makes it a good time to step back and assess what’s happening in enterprise AI. Some developments have matched expectations; others have surprised.

What Happened as Expected

Platform consolidation. The market has settled around three major platforms (Azure AI, Google Vertex, AWS Bedrock) as predicted. Small AI startups are either being acquired or struggling for enterprise traction.

Productivity tool adoption. Microsoft Copilot and similar tools achieved mainstream adoption in large enterprises. The adoption curve matched historical patterns for enterprise software.

Governance pressure increased. Regulatory requirements and enterprise policies for AI have tightened. Governance is no longer optional for serious deployments.

Model capability continued improving. GPT-5, Gemini 2.0, and Claude 4 represent meaningful capability improvements. The pace of advancement remains rapid.

Integration challenges persisted. Connecting AI to enterprise systems remains difficult. The promise of “just plug it in” hasn’t materialised.

What Surprised

Slower productivity gains than hoped. While adoption is high, measured productivity improvements are more modest than early projections suggested. The 40% productivity improvement Microsoft cited for Copilot hasn’t been replicated in most enterprise deployments.

Faster cost escalation. AI infrastructure costs have grown faster than many organisations budgeted. Token-based pricing creates unpredictable spend as usage scales.

More AI fatigue than expected. After initial enthusiasm, many users have settled into limited AI use rather than transformation. Sustained engagement is harder than initial adoption.

Less disruption to incumbents. The prediction that AI-native startups would disrupt established enterprise vendors hasn’t played out. Incumbents have integrated AI into existing products effectively.

Skills gap widened. Despite investment in AI training, the gap between what organisations need and what they have has grown. Demand increased faster than supply.

Market Dynamics

Pricing competition intensified. Google and AWS have aggressively priced AI services to compete with Microsoft. This benefits enterprise buyers but creates margin pressure throughout the ecosystem.

Vertical solutions emerged. Generic AI is giving way to industry-specific applications. Healthcare AI, legal AI, and financial services AI are becoming distinct categories.

Open source gained traction. Meta’s Llama and other open models are viable for enterprises willing to self-host. This creates options for organisations with data sovereignty concerns or cost sensitivity.

The consulting market exploded. AI consultants Melbourne have seen dramatic growth as enterprises realise they need help translating AI capability into business value.

What’s Working

Based on H1 observations, these approaches are delivering value:

Focused use cases. Organisations that picked specific, measurable problems are seeing better results than those attempting broad AI transformation.

Productivity tool deployment. Despite modest gains per user, the scalability of Copilot and similar tools delivers aggregate value in large organisations.

Document processing. AI for contract analysis, document summarisation, and content extraction is a proven category with clear ROI.

Code assistance. Development teams using GitHub Copilot and similar tools report meaningful productivity improvements.

What’s Not Working

These approaches are struggling:

AI-first transformation. Organisations that reorganised around AI as a strategy rather than a tool are mostly disappointed with results.

Autonomous agents. Despite vendor hype, autonomous AI agents handling complex workflows without human oversight aren’t reliable enough for production.

Generic chatbots. Customer-facing AI chatbots remain frustrating for users. The technology isn’t ready for uncontrolled customer interactions.

One-size-fits-all models. General-purpose AI applied without customisation underperforms domain-specific approaches.

Australian Market Specifics

The Australian enterprise AI market has distinctive characteristics:

Adoption lags global averages. Australian enterprises are about 6-9 months behind US counterparts in AI deployment maturity.

Data sovereignty remains a constraint. Regulatory and risk concerns about data leaving Australia continue to limit some AI applications.

Skills shortage is acute. The AI talent market in Australia is tighter than global averages, particularly outside Sydney and Melbourne.

Industry concentration matters. Financial services and mining drive a significant portion of Australian enterprise AI investment. These sectors have specific requirements.

Government influence is growing. Public sector AI adoption and policy is increasingly shaping the market environment.

H2 2025 Predictions

Looking ahead to the rest of the year:

Consolidation will accelerate. AI startups that haven’t achieved scale will face difficult choices. Acquisition activity will increase.

ROI scrutiny will intensify. After significant AI investment, executives will demand clearer evidence of returns. Projects that can’t demonstrate value will face cuts.

Governance will become table stakes. Enterprises without robust AI governance will face increasing risk. Governance won’t be a competitive advantage – it will be a requirement.

Vertical solutions will dominate. Industry-specific AI will outperform generic approaches. Vendors and enterprises will specialise.

The agent narrative will persist but disappoint. Autonomous agents will remain a priority for vendors but won’t achieve reliable production deployment this year.

What Enterprises Should Do

Based on H1 learnings:

1. Double down on what’s working. If productivity tools are delivering value, expand. If document processing has ROI, scale it.

2. Cut what isn’t. Projects without clear path to value aren’t going to magically improve. Reallocate resources.

3. Invest in skills. The skills gap is the biggest constraint on AI value. Training and hiring deserve significant budget.

4. Strengthen governance. If you haven’t established AI governance, do it now. Waiting increases risk without benefit.

5. Manage expectations. AI is useful technology, not magic transformation. Calibrate stakeholder expectations to reality.

Final Thought

H1 2025 has been a normalising period for enterprise AI. The initial hype has given way to practical reality. AI works, but not as easily or dramatically as hoped.

That’s not bad news – it’s maturation. The enterprises that will succeed in H2 and beyond are those that approach AI as serious technology adoption, not revolution.

The work continues. The opportunity remains. But realism is required.