AI’s asset avalanche: Managing the hidden risks of the next tech revolution

If you were asked tomorrow to show the exact cost of your AI models, could you do it? And which hidden “Shadow AI” might already be draining your budget or harvesting your organisational data with no control?

AI has scaled from isolated pilots to the engine room of modern enterprises, driving an explosive rise in hardware, software, and cloud assets. If unmanaged, these multiplying assets carry financial, regulatory, and environmental consequences that can quickly escalate to board-level crises.

The mounting risks of AI: Uncontrolled costs, compliance and sustainability pressures

The AI boom has triggered an infrastructure buying spree. Spending on AI-specific servers and storage nearly doubled in the first half of 2024 to $47.4 billion (+97% YoY), with servers accounting for ~95% of AI infrastructure spend. The International Data Corporate (IDC) expects AI infrastructure to surpass $200 billion by 2028^1. In 2025, spending on AI-optimised servers is projected to reach $202 billion, more than twice the outlay on traditional servers². According to Gartner, overall data-centre systems spend will rise 23.2% to $405.5 billion in 2025³.

While generative AI adoption is accelerating, only around 42% of CFOs say their organisations are actively exploring GenAI, underscoring the fragmented, often informal nature of adoption⁴. Licensing terms are growing more complex: some proprietary vendors now claim rights over model outputs, while many open-source licences require derivative contributions. Regulators worldwide, from the UK’s Online Safety Act to HIPAA and the EU’s GDPR, are tightening personal data rules.

At the same time, AI workloads do not stop consuming resources once deployed. Inference can run continuously, driving up cloud and energy costs. IDC tracks rapid growth in AI server deployments, including cloud-hosted models with significant energy footprints¹. Investors and regulators are therefore increasingly demanding evidence that AI aligns with ESG commitments, making sustainability an inseparable part of compliance.

Why the stakes will only rise

Macro indicators point upward. Meta projects up to $72 billion in AI infrastructure spend in 2025⁵, while Nvidia’s CEO forecasts global data-centre buildouts will break the $1 trillion mark two years earlier than expected⁶. In parallel, Gartner forecasts data-centre systems spend at $405.5 billion (+23.2%) in 2025², and IDC projects overall AI spending to reach $632 billion by 2028⁷, with annual growth averaging ~32%⁸.

Against this backdrop, most organisations remain unprepared. Deloitte’s surveys and client engagements reveal the same pattern: boards increasingly approach us with urgent questions about tracking fast-growing AI estates and staying compliant amid tightening regulation⁹.

The findings of the most recent Deloitte IT Asset Management (ITAM) survey corroborate these observations¹⁰:

• Visibility crisis: a growing lack of visibility into cloud-based assets and consumption was identified as the top challenge faced by ITAM teams in managing assets within AI/GenAI and cloud ecosystems.
• AI compliance gaps: the growing need to ensure compliance, with increasingly complex licensing terms, for AI and cloud services was cited by 42% of organisations.
• Cost governance maturity: organisations are seeing measurable benefits through combining FinOps cos transparency with ITAM asset tracking, however, despite these benefits, 82% of respondents admit their organisation has low or underdeveloped synergy between ITAM and FinOps.

How can AI be managed?

“AI asset management” is not yet a clearly defined discipline, but the principles are familiar. It will require evolving several practices: FinOps, governance, security, and ITAM. AI should be treated as an asset class that needs to be managed across its entire lifecycle. One proven starting point is ITAM, which already provides structures for lifecycle control, compliance, and financial tracking.

While FinOps provides visibility into cloud costs, it requires the asset-level context to explain where those costs originate or whether they are compliant. Modern ITAM, when extended to AI, can help plug this gap by linking real-time telemetry, such as cost per 1,000 tokens or kilowatt-hours per training epoch, with lifecycle and licensing data. The result is a shift from static inventories to live dashboards that not only highlight runaway clusters early, but also connect spend to compliance and business value.

Key pillars include:

• AI asset lifecycle management: extending ITAM practices to encompass licensing, deployment, and decommissioning of AI assets.
• Performance monitoring & optimisation: tracking the efficiency and effectiveness of AI systems, e.g. monitoring latency in an AI-powered fraud detection system.
• Cost management: mapping licensing fees, cloud resources, and storage costs back to business value, making AI spend predictable and optimised.
• Sustainability integration: recording energy use per training epoch or inference workload, ensuring AI aligns with ESG commitments.

Building a practical framework

Building a framework for managing AI’s complexity and scale will require a layered approach that extends ITAM best practices and integrates them with FinOps and governance oversight.

Foundational pillars may include:

• Continuous discovery scanners that fingerprint every GPU, TPU, and AI-SaaS subscription across multi-cloud estates, enriched with lineage, classification tags, and energy profiles.
• Contract-intelligence engines that parse licence agreements, including token-based models, mapping them to ISO/IEC entitlements to prevent “Shadow AI” drains.
• Governance workflows that route anomalies, such as rogue model uploads or carbon hotspots, into remediation pipelines.
• A FinOps overlay that correlates compute, storage, and network telemetry with cost centres, exposing unit economics that resonate with CFOs and strategists.

The verdict

AI’s explosive growth demands a robust management framework that brings together ITAM, FinOps, governance, and sustainability. Organisations that treat AI as a managed asset can control costs, strengthen compliance, and align initiatives with business value.

Deloitte supports clients in shaping these frameworks, combining experience in ITAM with deep AI expertise to guide enterprises through this journey, not only with proven methods, but also with the confidence that comes from helping leading organisations manage technology at scale. Boards that ignore these issues today may face costly compliance failures tomorrow, so start by mapping where your AI workloads sit - and whether they’re covered by your existing ITAM, FinOps, and compliance structures.

_____________________________________________________________

References

1. IDC: Artificial Intelligence Infrastructure: 1H24 and 2028 Outlook (1H24 $47.4B, +97% YoY; servers 95%; infra > $200B by 2028).

2. Gartner - AI-Optimised Servers 2025 (~$202B in 2025; >2× traditional servers).

3. Gartner: Worldwide IT Spending Forecast 2025 (Data-centre systems $405.5B, +23.2% YoY in 2025).

4. Deloitte: CFO & Enterprise GenAI Signals (≈42% of firms actively exploring GenAI).

5. Meta Platforms, Inc.: Earnings call / CapEx guidance 2025 (CapEx up to $72B, driven largely by AI infra).

6. Nvidia Corporation: CEO statement, 2024/25 outlook (data-centre buildouts to $1T, two years sooner).

7. IDC: Worldwide AI Spending Guide (AI spending to $632B by 2028).

8. IDC: Agentic/AI Growth Outlook (AI spend CAGR ~31.9% through 2029).

9. Deloitte: State of Generative AI in the Enterprise (adoption & governance insights).

10. Deloitte ITAM survey 2025.