Enterprise AI Spending: Where the Money Is Going in 2026

Enterprise spending on AI infrastructure is accelerating faster than any technology adoption cycle in recent history. According to IDC's Worldwide AI Spending Guide, global enterprise AI spending is projected to reach $227 billion in 2026, up from $154 billion in 2024 — a 47% increase in two years. The question that matters is not how much is being spent, but what is actually delivering returns.

Where is the money going?

Enterprise AI budgets are flowing into four primary categories, each with different maturity levels and return profiles:

Infrastructure and compute absorbs the largest share at approximately 40% of total spending. This includes GPU clusters, cloud AI services (AWS Bedrock, Google Vertex AI, Azure AI), and the networking and storage infrastructure required to support large-scale model inference. NVIDIA reported record data centre revenue of $35.1 billion in Q4 2025, driven almost entirely by AI demand.

AI applications and platforms account for roughly 30% of spending. This includes both off-the-shelf AI SaaS products (Copilot, Jasper, Harvey) and internal platform investments for building custom AI solutions. The platform layer — tools like Amazon Bedrock, Google Vertex AI, and internal ML platforms — is growing fastest as enterprises move from experimentation to production deployment.

AI services and consulting represent about 20% of budgets. According to McKinsey's Global AI Survey, 72% of enterprises have engaged external consultants to support AI implementation, up from 50% in 2023. Demand is particularly high for AI strategy, data readiness assessment, and change management.

Data infrastructure — including data lakes, pipelines, governance tools, and quality systems — takes the remaining 10%. This is widely acknowledged as underinvested relative to its importance. Models are only as good as the data that feeds them, and most enterprises have significant data quality gaps.

What is actually delivering ROI?

The ROI picture is mixed. Boston Consulting Group's research found that only 26% of enterprises have moved generative AI projects beyond pilot stage, and fewer than 10% have achieved measurable ROI at scale.

The use cases delivering the clearest returns share common characteristics:

• Customer service automation. AI-powered contact centres are reducing cost-per-interaction by 30-50% while maintaining or improving satisfaction scores. Companies like Klarna have reported replacing hundreds of full-time equivalent roles with AI agents handling routine queries.
• Code generation and developer productivity. GitHub's research indicates that developers using AI coding assistants complete tasks 55% faster. At enterprise scale, this translates to significant productivity gains across engineering organisations.
• Document processing and summarisation. Legal, financial, and healthcare organisations are seeing strong returns from AI systems that process contracts, reports, and clinical notes. The ROI is straightforward — tasks that took hours now take minutes.
• Sales and marketing content. Personalised outreach, product descriptions, and marketing copy generated by AI is delivering measurable improvements in engagement and conversion.

What is not working?

Several high-profile investment categories are underdelivering:

General-purpose chatbots deployed without clear use cases are showing poor adoption rates internally. According to the Stanford AI Index 2025, enterprise chatbot projects have a 60% abandonment rate within 12 months of deployment. The pattern is consistent: broad deployments without specific workflow integration fail to change behaviour.

Autonomous decision-making systems in high-stakes domains (lending, hiring, medical diagnosis) face regulatory friction, liability concerns, and user trust barriers that limit deployment. The AI Act's requirements for human oversight in high-risk applications add compliance costs that many organisations have not yet budgeted for.

Custom model training by non-AI-native enterprises has largely failed to deliver advantages over fine-tuned foundation models. The cost of acquiring training data, maintaining training infrastructure, and iterating on model architecture exceeds the benefit for most use cases.

What does the spending trajectory look like?

Three trends will shape enterprise AI spending through 2027:

1. Consolidation around fewer, larger platforms. The early experimentation phase produced a proliferation of point solutions. Enterprises are now consolidating onto two or three core platforms to reduce integration complexity and vendor management overhead.

2. Shift from inference to agent infrastructure. As organisations move from single-query AI to multi-step agent workflows, spending will shift toward orchestration platforms, tool integration layers, and monitoring systems. The agent infrastructure market is projected to grow at over 100% annually through 2028.

3. Data investment catching up. The realisation that model capability is bottlenecked by data quality is driving increased investment in data infrastructure. Enterprises that invested early in data governance are seeing disproportionate AI returns — a pattern that is not lost on the laggards.

The enterprises generating real returns from AI share a pattern: they started with specific, measurable use cases, invested in data quality before model sophistication, and built internal capability rather than outsourcing everything to consultants. The technology works. The challenge is organisational, not technical.