Deloitte: AI Drives 166% Hardware Spending Surge, Reshapes Data Centers

A massive AI hardware spending surge is reshaping the tech industry, driven by demand for AI servers and semiconductors. According to Deloitte's 2026 Global Outlook, spending on AI compute and storage hardware soared 166% year-over-year to US$82 billion in Q2 2025, forcing a fundamental rethink of everything from chip design to data center architecture.

This momentum is now steering decisions from chip roadmaps to consumer-device pricing, creating an industry that prizes power density, trusted data use, and affordable innovation in a K-shaped economy.

AI hardware goes into hypergrowth

This hypergrowth is primarily fueled by the massive computational needs of AI models, particularly for GPUs, high-bandwidth memory, and specialized AI accelerators. With generative AI chips alone projected to capture half the semiconductor market, server makers and component suppliers are seeing unprecedented demand for more powerful hardware.

Semiconductor revenue is on track to climb from an estimated US$772 billion in 2025 to US$975 billion in 2026, driven by GPUs, high-bandwidth memory, and specialized AI accelerators according to Deloitte's outlook on semiconductors. Generative-AI chips alone could approach US$500 billion next year, roughly half of the total chip market.

Server makers are feeling the pull, with AI servers now representing 20% to 70% of revenue for leading vendors after a market-wide shipment decline in 2023. Related components are also exploding in value: power supplies for AI servers may rise from US$1.5 billion in 2024 to more than US$31 billion by 2028, while liquid-cooling gear could leap from US$300 million to US$30 billion in the same window.

A quick scan of breakout segments:
- AI server networking: US$8 billion in 2023, heading for US$34 billion by 2028
- Inference workloads: projected to consume two-thirds of AI compute cycles by 2026
- Chiplets and co-packaged optics: moving from pilot to mainstream to ease bandwidth bottlenecks

Data centers adopt dense and hybrid blueprints

The insatiable power and bandwidth demands of AI models are forcing a physical transformation of data centers. Facilities are rapidly shifting from traditional air-cooled rooms to dense, liquid-cooled racks connected by high-speed fiber. Deloitte predicts nearly half a trillion dollars will be invested in new AI-optimized data centers by the end of 2026, with operators building gigawatt-scale campuses to cluster GPUs.

However, enterprises are not moving all workloads to the cloud. Instead, the report highlights a strategic pivot to hybrid, multitier architectures that blend on-premises GPU clusters, regional edge locations, and public clouds (Deloitte). This approach allows IT teams to balance cost, latency, and data sovereignty by orchestrating workloads across a complex mix of CPUs, GPUs, and other processors. This shift is also creating a significant talent crunch for specialists in high-bandwidth networking, liquid cooling, and large-scale GPU management.

Consumer tech meets trusted innovation

In the consumer market, the AI hardware boom translates into a demand for products that are personal, private, and practical. Deloitte analysts identify a market shift away from flashy gadgets toward "trusted innovation" - devices that solve tangible problems while respecting user data and budgets. This trend is evident in products like advanced robot vacuums and sophisticated emotion-sensing wearables.

In today's K-shaped economy, this plays out in two ways: premium consumers invest in high-tech wellness devices, while value-focused buyers seek affordable essentials. For both segments, data responsibility is paramount. Brands that build loyalty are those embedding transparent privacy controls and clear sustainability goals directly into their AI-powered features.

What to watch through 2026

As hardware reclaims its central role in technology, three key trends will define the landscape through 2026:
1. Cooling Technology Investment: Capital spending on advanced cooling is seeing exponential growth, signaling intense competition to establish new industry standards for managing heat in dense AI systems.
2. Hybrid Orchestration Platforms: The success of enterprise AI will depend on software that can intelligently and automatically route tasks to the most cost-effective and efficient compute node, whether on-premises, at the edge, or in the cloud.
3. Consumer Trust Metrics: Privacy and data control features will become critical selling points, with "trust dashboards" potentially becoming as standard on consumer devices as battery life indicators.

Deloitte's 2026 analysis concludes that AI's evolution is now fundamentally a story of physical infrastructure. From the silicon chip to the smart device in your home, the next wave of value will be captured by those who master the hardware.

---

What is driving the 166% surge in AI hardware spending?

AI inference workloads - not training - will consume two-thirds of all AI compute by 2026, forcing enterprises to buy power-dense servers, liquid-cooling kits and optical switches at record rates.
Q2 2025 enterprise spend on AI compute and storage hit $82 billion, up 166% year-over-year, while companion markets are exploding even faster: AI server power supplies leap from $1.5 B in 2024 to $31 B in 2028 and liquid cooling from $0.3 B to $30 B over the same span.

How are data centers being physically rebuilt for AI?

New facilities are gigawatt-scale, GPU-heavy and fiber-rich: racks now mix two GPUs per CPU, use co-packaged optics inside switches, and draw so much power that liquid cooling and high-voltage bus-ways are mandatory.
Copper Ethernet is giving way to end-to-end optical networks to cut latency inside AI clusters, while chip-level tweaks like high-bandwidth-memory chiplets boost throughput without enlarging die size.

Why are enterprises choosing hybrid, multitier AI architectures?

Hybrid lets CIOs balance cost, latency and data sovereignty by placing each workload on the cheapest compliant tier - on-prem GPU clusters for sensitive inference, cloud bursts for training, and edge NPUs for ultra-low-latency tasks.
Deloitte expects open-source models inside sub-$1 million appliances to sit beside multi-rack GPU clusters, all orchestrated by software that routes jobs dynamically across cloud, metro-edge and on-prem gear.

How big will the AI infrastructure market be by 2029?

IDC projects total AI infrastructure spending - servers, storage and networking - will reach $758 billion in 2029, with accelerated servers alone capturing 94% of the total.
Within that figure, enterprise end-users are forecast to add $55.5 billion of new spend between 2024 and 2029, roughly a 30% share, as they race to deploy inference-dominant workloads.

What does "trusted innovation" mean for consumer tech in a K-shaped economy?

Trusted innovation equals AI that feels helpful, not intrusive: brands win when gadgets deliver clinical-grade wellness insights (e.g., WHOOP), protect privacy via robust data governance, and still hit mass-market price points through rapid prototyping and social-commerce channels.
The K-shape splits shoppers into premium wellness buyers and value-driven basics seekers, forcing vendors to hyper-segment offerings while keeping data responsibility and affordability front-and-center.