The deployment race for industrial artificial intelligence (AI) in the energy sector is on, and the United States has roughly twenty-four months to set the standards that will create technological and commercial dependencies for the next decade or longer.
AI architectural decisions are already yielding real world results, according to energy companies.* In Saudi Arabia’s Khurais oil field, artificial intelligence is processing data from more than 40,000 sensors in real time and adjusting production without a human in the loop. The same architecture is running on Shell’s deepwater platforms, at SLB’s autonomous drilling operations offshore from Brazil, and across ADNOC’s refineries in Abu Dhabi.
This architecture lives in the interfaces that connect the layers of AI—and whoever defines how those three layers talk to each other will set the standard that many in the global energy industry will follow.
From isolated tasks to integrated systems
What separates the operators capturing this value from the rest is not more AI pilots. For a decade, companies have run AI as a catalog of dozens of discrete “use cases.” The approach felt rigorous but has become the industry’s largest source of stranded AI capital: a 2025 study of enterprise AI found that 95 percent of pilots deliver no measurable return, diagnosing the failure as one of workflow integration, not technology. Leaders do the opposite: they reinvent entire workflows as integrated systems, from the drilling decision loop, to pipeline integrity, to grid dispatch. The unit of transformation is the workflow, not the task.
That architecture integrates three layers of AI: generative, which reads incoming data; agentic, which takes a situational picture and decides what to do; and edge, which executes the decision at the wellhead, the substation, and the compressor station. The architecture that integrates these three layers will determine who leads in AI, and it has not yet entered the policy framework in Washington.
Policy must pivot accordingly
Currently, the AI policy debate in the United States is almost entirely about chips: export controls on advanced semiconductors to China, domestic fabrication subsidies, and data center buildout. While important, chip-focused policy is not what will determine who wins the race for AI dominance across the energy sector. As an analogy, this would be like the auto industry focusing just on engines when industry leadership is determined by system-wide thinking about how all of a vehicle’s components work together as a cohesive whole.
Similarly, to ensure AI leadership, US policymakers need to focus on the architecture that runs on top of the chips. The good news is that the United States has the tools to set the standards for this approach. It is not yet using them, but that can change if key mechanisms are reoriented to do so.
A technical primer: From Cloud 1.0 to Cloud 3.0
In the early days, industry ran its computing on its own servers, then increasingly on rented capacity in the data centers operated by providers like Amazon, Microsoft, and Google. This shift changed what energy companies could do with their data. The first wave—call it Cloud 1.0—was infrastructure migration. This was mostly back-office workloads moving to hyperscale cloud capacity. The second wave, Cloud 2.0, added an analytics layer that produced useful insight, but the insight stayed in the office.
Cloud 3.0 is categorically different from the earlier waves. It deploys AI to allow models to act. This action enables decision-making at the asset itself: a wellhead that adjusts its own production settings the moment sensor data shifts, or a pipeline compressor that detects an anomaly and throttles down before a human operator has even seen the alert.
This operational edge is where the competitive advantage resides. Run the three layers separately, in three different vendor stacks, and the loop never closes. Integrate them into a single architecture, and the asset begins running itself within the envelope the operator defines.
For operators already deploying this approach, its significance is clear. Saudi Aramco’s Khurais field has reported a 15 percent production increase through AI-enabled optimization across its sensor network. SLB’s autonomous drilling deployments, with Equinor* offshore Brazil have increased drilling speed by 60 percent on targeted well programs. Shell runs production-surveillance edge AI on-premise across its deepwater footprint. ADNOC’s downstream program with Honeywell anchors the refining application.
Four levers for US policy
The global leader in AI deployment will be whoever integrates the AI layers at scale, not whoever invents the protocols that connect them. The procurement cycles closing in the next twenty-four months are setting the pace of that integration, and what every subsequent operator, vendor, and regulator inherits.
A useful analogy is the standard that has been set for the USB-C connector, the plug that has become the universal device for transmitting data, video, and power. Regardless of who makes it, the value of the USB-C connector is in its status as the standard. The same dynamic is playing out across industrial AI. The leader will be whoever drives the adoption and integration of these protocols across the industry.
A shift in US policy can better position Washington to set the standards for industrial AI across the energy sector. It must start with the recognition that a new ecosystem of energy-transition startups, grid capacity projects, and autonomous drilling platforms is forming around Cloud 3.0, and that this ecosystem is fundamentally changing how the energy industry is incorporating AI. Rather than using AI to focus on individual tasks, operators are leveraging the technology to reinvent horizontal workflows, including the drilling decision loop, the integrity-management cycle, and dispatch and balancing. This transformation requires new operating models for decision making and the policy environment to accelerate their deployment.
The interfaces that connect the AI layers already exist. Open protocols such as the Model Context Protocol, now adopted across the major AI providers, define how AI systems exchange data and instructions. This is the USB-C lesson: the value was never in inventing the plug; it was in everyone adopting it. What is missing is not the protocol but the integration: the operating models inside energy companies, and the ecosystem of service providers, systems integrators, and equipment manufacturers that assemble the layers into working systems. Policy cannot legislate that integration, but it can accelerate it, and it can stop rewarding the use-case portfolio that current frameworks incentivize. Four levers can do this work, each through existing executive-branch authority, with no new legislation.
Federal procurement on the FedRAMP model.
In 2011, the Office of Management and Budget created the Federal Risk and Authorization Management Program (FedRAMP) to set the standards that a cloud provider had to meet before any government agency could buy from it. Within a decade, it reshaped how every major provider built its security architecture. The same mechanism can accelerate integration today: requiring that AI systems purchased for federal energy infrastructure run on open, published interfaces and demonstrate integrated operation across the generative, agentic, and edge layers. Federal purchasing would then reward proven integration rather than another pilot.
Certification frameworks through the National Institute of Standards and Technology (NIST) and federal energy regulators.
Today, every utility, pipeline company, and grid operator must convince its regulator that its AI system is safe and reliable under a wide variety of regulatory frameworks, which is slowing deployment of AI systems. A uniform certification framework would reduce unnecessary regulatory burdens by setting a common national standard for energy operators to meet and against which regulators can benchmark. The building blocks to achieve this uniform framework already exist. NIST’s AI Risk Management Framework, a voluntary guidance for managing risks across the AI lifecycle that can be applied to any industry, can be formally adopted to certify integrated industrial systems, including the safe handoff of decisions between the AI layers. The Federal Energy Regulatory Commission and North American Electric Reliability Corporation, which set reliability standards for the power grid, can do the same for AI-enabled grid assets, and the Department of Energy (DOE) can lead for energy facilities. Certifying the integrated system, and the integrators who assemble it, means a system certified once can be trusted everywhere.
Capital conditioning through the DOE Loan Programs Office (LPO) and tax-credit guidance.
Today, federal capital flows into energy projects with no regard for the AI architecture those projects are built on. Conditioning would fix that. The LPO, using existing authority, can require that AI-enabled projects it backs meet specific architectural conditions: interoperable interfaces between the AI layers, inference that can move between vendors, and data the operator owns and can take with it, and that they demonstrate workflow-level integration in the project design rather than a catalog of pilots. Section 45 tax-credit guidance can attach the same conditions to the projects that claim it.
US-led international standards leadership.
The certification that a vendor is required to pass today will determine what every competitor must match by 2050. The question is whose certification that will be. Europe largely regulates AI that it does not build. China builds AI that Western regulators will not certify. The de facto interfaces of industrial AI are already American, as the cloud standards were before them, and the deployments that would anchor the American case are already running in Saudi Arabia, Abu Dhabi, and offshore Brazil. The task is to carry these de facto standards into the international bodies and bilateral frameworks before others codify alternatives. Only the United States can anchor an international standard in demonstrated industrial performance.
US policy is not wrong to focus on chips. But the integrated-stack standard for industrial AI is not determined by who has the fastest chip. It is determined by federal procurement requirements, the certification frameworks that define what a compliant industrial AI system looks like, the capital conditions attached to federal energy loans and tax incentives, and the standards bodies where the long-cycle norms get set.
In the next twenty-four months, operator capital budgets, vendor contract renewals, and the first regulatory frameworks will lock in. Within this period, the United States has the opportunity to pull existing levers to shape global standards for industrial AI that will cascade throughout the energy sector. What is needed is the policy focus above the silicon layer, where industrial AI is actually decided.
Arno van den Haak spent more than two decades at Royal Dutch Shell in senior engineering and commercial roles across global upstream operations and is the former head of global business development for energy at Amazon Web Services.
*Aramco, Shell, and Equinor are partners of the Global Energy Center.
Stay connected
Sign up for PowerPlay, the Atlantic Council’s bimonthly newsletter keeping you up to date on all facets of the energy transition
related work
our work
The Global Energy Center develops and promotes pragmatic and nonpartisan policy solutions designed to advance global energy security, drive economic opportunity, and foster a sustainable energy future.
Image: Off Shore Oil Rig, Huntington Beach CA (Arvind Vallabh, Unsplash, https://unsplash.com/photos/an-oil-rig-in-the-middle-of-the-ocean-kjk90_OSGZU).