Hyperscalers Plan $660B in 2026 AI Capex, Widening Gap for Enterprise Buyers

Hyperscalers Double Down on Infrastructure Control

Microsoft, Amazon, Alphabet, Meta, and Oracle now plan to spend between $660 billion and $690 billion on AI infrastructure in 2026, according to Futurum Research. That is nearly double the roughly $380 billion spent in 2025. Amazon leads with $200 billion in projected 2026 capex, followed by Alphabet at $175 billion to $185 billion, Meta at $115 billion to $135 billion, Microsoft at $120 billion or more, and Oracle at $50 billion. Most of that capital goes to data centers, AI compute, networking, and power infrastructure.

For enterprise buyers, this creates two immediate effects. First, premium AI capacity remains scarce and pricing stays high because hyperscalers are building for their own needs and prioritizing internal workloads and strategic customers. Second, bundling intensifies. Vendors will tie AI credits, managed services, and platform commitments to infrastructure access as they work to monetize those investments. If you need guaranteed GPU allocation or reserved capacity, expect to negotiate multi-year platform commitments in return.

The spending gap also creates a structural disadvantage for smaller cloud providers and regional infrastructure vendors. Hyperscalers lock in advantages in GPU supply, power capacity, and data center locations that are difficult to replicate at smaller scale. If your procurement strategy depends on competitive tension between providers, that leverage narrows when five companies control the majority of global AI compute investment.

Sovereign and Regional Infrastructure Gains Momentum

Core42 secured $550 million in financing from HSBC to expand sovereign cloud and AI compute infrastructure across the U.S. and Europe. This strengthens the sovereign-cloud alternative to U.S. hyperscalers for regulated industries and government workloads. For buyers in financial services, healthcare, or public sector, this financing reduces risk around data residency and jurisdiction by adding a funded supplier to compare against AWS, Microsoft, Google Cloud, and regional colocation providers.

Nvidia and IREN are reportedly planning to deploy up to 5 gigawatts of AI infrastructure globally, with Sweetwater, Texas as a flagship site for Nvidia's DSX initiative, according to Data Center Knowledge. A 5 GW pipeline is a scale signal that tightens access to power and land while expanding Nvidia's ecosystem influence beyond chips into infrastructure deployment. For enterprise buyers, this means capacity availability is a strategic risk factor. Large committed power footprints often translate to longer lead times for new AI clusters and more dependence on vendors that can guarantee delivery at scale.

Thailand's Board of Investment approved TikTok's $29 billion investment plan to strengthen the country as a regional hub for data centers, cloud services, and AI infrastructure. This follows similar moves in Malaysia, Singapore, and Indonesia. Multinationals with APAC data residency or latency requirements may gain more deployment options, but they also face a more fragmented vendor landscape and more country-specific compliance reviews.

NextDC's KL1 facility in Kuala Lumpur, a 65 MW Tier IV data center designed for AI and hyperscale growth, has now topped out. The facility already achieved Uptime Institute Tier IV design certification and a Platinum Green Building Index rating. For buyers prioritizing uptime and energy profile, Tier IV-certified capacity lowers operational risk and simplifies vendor diligence, especially for inference and mission-critical AI workloads.

Platform Adoption and Governance Tooling Mature

Nearly 75% of Google Cloud customers now use its AI products, with Gemini processing more than 16 billion tokens per minute via direct API use, according to June 2026 data captured in the AI Edge roundup. High token throughput suggests maturity and scale, which reduces perceived platform risk for buyers evaluating model availability, latency, and production readiness. If accurate, this reinforces Google Cloud's position as a serious enterprise AI platform contender versus Microsoft Azure and AWS, especially on model consumption and developer adoption.

Red Hat released a Model Context Protocol server for OpenShift in technology preview. The MCP server gives AI agents a safer way to inspect and interact with Kubernetes and OpenShift clusters. For buyers running hybrid or on-prem AI infrastructure, this creates a clearer path to agent governance and least-privilege controls, which can reduce security review friction for production deployments. This is a governance-oriented move against infrastructure platforms that still lack native agent controls.

What to Watch

Track whether hyperscaler capex translates to broader enterprise capacity availability or remains concentrated in managed services and internal workloads. Monitor sovereign and regional infrastructure buildouts for credible alternatives to U.S. hyperscalers, especially in regulated industries. Evaluate platform lock-in risk as vendors tie infrastructure access to multi-year commitments. And assess agent governance tooling maturity if you plan to deploy agentic AI in production environments.