Artificial intelligence enters the second quarter of 2026 with a notably different centre of gravity than it had at the start of the year. The conversation has moved from raw benchmark results toward three more practical questions: who has the compute, who has the agentic tooling that actually works, and who can deliver sovereign deployments for governments and regulated industries.
The compute story remains a contest of scale. Hyperscalers, sovereign vehicles and specialised neoclouds continue to add capacity across the United States, the Gulf and East Asia, with announced commitments running into the hundreds of billions of dollars. GPU availability that was the binding constraint of 2024 has eased in segments, but bottlenecks have moved upstream into networking gear, power transmission and skilled site-build labour.
Agents move from demo to deployment
The bigger near-term shift is in agentic AI. Frontier labs and infrastructure vendors have spent the past quarter packaging long-running agents that handle multi-step workflows in code, finance and operations. Customer references are no longer limited to small pilots: large banks, telecoms and manufacturers are running production deployments at meaningful scale, often with humans in the loop only for exception handling.
The trade-off for buyers is clear. Faster cycles and lower per-task cost in exchange for accepting a new operational risk surface, including prompt-injection vulnerabilities, opaque decision chains and the cost of running evaluation harnesses across changing model versions. Several large enterprises have begun standing up internal AI assurance functions modelled on internal audit, rather than treating the technology purely as an IT project.
Sovereignty is the regional theme
For the GCC, the dominant theme of May has been sovereign AI. National-champion operators in the UAE and Saudi Arabia continue to deepen partnerships with US frontier labs and chip vendors, while also building local talent benches. The architecture of choice is increasingly hybrid: foundation models accessed through controlled cross-border arrangements, with fine-tuning, retrieval and inference kept inside national perimeters.
Regulators are catching up at uneven speed. The UAE has continued to push specialised AI authorities and sandbox programmes; Saudi Arabia has expanded its data-centre-friendly regulatory bandwidth. Europe and the US continue to negotiate the contours of AI export controls and national-security review, with consequences that ripple back into every regional deployment plan.
What to watch this month
Three things matter for the rest of May. First, the cadence of frontier model releases and their pricing structures. Second, the depth of enterprise distribution agreements being announced between AI labs and the world's largest professional-services firms, which determine how quickly the technology reaches Fortune 1000 buyers. Third, the trajectory of cyber and trust-and-safety announcements, as both vendors and regulators acknowledge that AI is now actively used by attackers.
