The year 2026 is poised to be a pivotal moment for artificial intelligence, marking its "takeoff year" and ushering in an era previously confined to science fiction. This transformation is driven by three interconnected trends: accelerating AI capabilities, intensifying geopolitical competition, and critical, often overlooked, constraints.

Accelerating Capabilities and Industrial Intelligence

AI's capability is not only accelerating but defying predictions of hitting a developmental wall. This advancement is characterized by a shift towards "industrial intelligence," a concept rooted in the large-scale, resource-intensive process of manufacturing non-biological intelligence. This process begins with raw materials like silicon, transformed through immense energy inputs into sophisticated AI systems.

Historically, AI skeptics predicted a halt in progress due to "human data exhaustion." However, researchers circumvented this by focusing on inference scaling, allowing models to "think longer" and test their logic, leading to significant improvements in reasoning. Further advancements include breakthroughs in autonomous architectures, enabling AI models to act independently for extended periods, giving rise to AI agents.

Despite these leaps, compute remains a constraining factor. Even cutting-edge models in 2026 are trained on existing infrastructure, with significant industrial buildouts of new compute and algorithmic efficiencies yet to be fully integrated into public models. This pipeline suggests continued exponential growth in AI capability.

Mythos: A Paradigm Shift in AI

The emergence of models like "Mythos" exemplifies this accelerating capability curve. Anthropic's decision not to release Mythos publicly underscores its emergent and potentially dangerous capabilities, particularly in cybersecurity. Mythos has demonstrated an astonishing ability to exploit zero-day vulnerabilities across various operating systems and browsers, many of which have remained unpatched for decades.

Anthropic's collaboration with "Project Glass Wing" highlights the perceived threat of Mythos's cyber capabilities. The dispersal of these capabilities across models necessitates a proactive approach to cybersecurity, focusing on AI-driven defenses to counter AI-driven threats. The development of AI to combat AI is becoming a fundamental requirement for a secure cyber ecosystem.

The Economics of Intelligence: Cost, Compute, and Constraints

The economics of non-biological intelligence present a stark contrast between cheap and valuable intelligence. While the cost of older AI model queries, such as GPT-3, has plummeted by 99.9% (from an estimated $60 to$0.06 per million tokens), the cost of frontier model intelligence remains exceptionally high. This is due to the extensive token usage required for models to validate their logic, enhance reasoning, and autonomously execute tasks. Developers are reportedly spending thousands of dollars daily on agentic experiments.

This "intelligence divide" between affordable baseline intelligence and extraordinarily expensive frontier intelligence is a critical economic factor. While the cost of individual tokens may decrease, the overall demand for compute and tokens is increasing, driven by the pursuit of more valuable intelligence. This trend, akin to Jevons paradox, suggests that as the cost per token falls, overall token consumption rises, potentially leading to severe shortages of compute and tokens for the most demanding tasks.

The challenge lies in making frontier-level intelligence, particularly with agent workloads, affordable and distributable. OpenAI's projected expenditure of $25 billion on training compute in 2026, rising to$121 billion by 2028, illustrates the immense cost of developing frontier intelligence. This development is largely subsidized by U.S. hyperscalers, labs, and venture capital. However, as enterprises increasingly demand access to this intelligence, a token and compute constraint is anticipated, potentially driving up costs and creating a supply crisis for frontier intelligence in the short to medium term.

Tokens are not merely an add-on but a requirement for productivity and discovery. The ability to leverage tokens for agentic experiments in design, discovery, and delivery represents a significant force multiplier. This is expected to accelerate token use across enterprises, leading to a constraint on both tokens and intelligence. Productivity will increasingly be measured by token availability rather than solely by human capital.

Anthropic's revenue growth exemplifies the enterprise demand for frontier intelligence. From $1 billion in revenue at the end of 2024, it projected$9 billion by December 2025 and $120 billion by December 2026. This surge, primarily from businesses, underscores the value of accelerating AI capabilities. While over a thousand customers spend more than$1 million annually with Anthropic, the addressable market is vast, encompassing every business. However, realizing this potential hinges on reducing the costs of frontier intelligence. Currently, less than 1% of businesses deploying AI have mature deployments, indicating a significant opportunity for growth as costs decrease and advantages compound.

Geopolitical Competition and the AI Superpower Race

The control and deployment of non-biological intelligence at scale are becoming the defining geopolitical competition of the 21st century. Vladimir Putin's 2017 assertion that "whoever controls AI will rule the world" accurately diagnosed this trend. While Russia has focused on military conflict, China and the United States have made AI dominance a central policy objective. China's 2017 AI development plan aimed for global leadership by 2030, while the U.S. has explicitly pursued AI dominance since 2025.

This competition is framed by a "five-layer pyramid" for becoming an AI superpower, encompassing:

1. Energy and Raw Resources: The foundational input for industrial intelligence, including manufacturing capabilities.
2. Hardware and Compute: Semiconductors, semiconductor manufacturing, and sovereign cloud infrastructure.
3. Intelligence Layer: The AI models themselves.
4. Applications: Deployment of intelligence in areas like drug development.
5. Transformational Integration: The apex of the pyramid, achieved through the synergy of the lower layers.

The critical insight is that AI superpower status cannot be achieved without a strong foundation across all layers. The West's historical offshoring of supply chains and industrial bases is now recognized as a strategic vulnerability. Vertical integration across this pyramid is essential, a feat achievable only by superpowers. The United States and China are the primary contenders for this vertical integration.

While other nations play crucial roles in specific layers—Taiwan in advanced semiconductor manufacturing, the Netherlands in EUV machines, and the Gulf in energy and compute—they may not achieve the comprehensive vertical integration required for AI superpower status.

China's Advantages

China possesses a distinct advantage in the foundational layer: raw resource production, energy generation, and its industrial and manufacturing base. As the "world's factory," China benefits from an established ecosystem for producing complex physical goods, including EVs, drones, batteries, and rockets. With an estimated 200 million skilled workers, China is not merely catching up but leapfrogging the West in critical industries like electric vehicles (BYD), batteries, and commercial drones (DJI). This industrial prowess, combined with elite AI researchers, positions China strongly for scaling industrial intelligence.

China's energy production capacity is also significant. In 2025, China added over 400 gigawatts of generating capacity, exceeding the U.S.'s total historical solar capacity addition in a single year. Its combined energy production capabilities surpass those of the U.S. and the EU, and it is actively building 38 nuclear reactors, contrasting with Germany's decommissioning of its nuclear facilities.

The Intelligence Layer and Open Source

While the U.S. is perceived to lead in the intelligence layer (models), China is rapidly closing the gap and leading in the open-source framework for models. This presents a potential scenario where U.S. frontier models remain closed, while global development relies on Chinese-developed open-source architectures.

Innovation at the frontier is further complicated by industrial-scale espionage targeting U.S. labs for model weight exfiltration. Additionally, the presence of illegal chips in China, despite export controls, facilitates training runs. Coupled with the fact that approximately 50% of the world's elite AI researchers are in China, the notion of frontier model innovation being exclusively a U.S. domain is inaccurate.

The Compute Dollar and Sovereignty

For the past 50 years, oil was the most valuable resource, shaping geopolitical dynamics. The 21st century's most critical resource is intelligence, and by extension, compute. The "compute dollar" is expected to set the terms for the geopolitical order, similar to how the petrodollar did.

National sovereignty in the 21st century is intrinsically linked to economic prosperity and security, both of which are downstream of AI and advanced technology. Control over intelligence production and secure supply chains are paramount for maintaining sovereignty.

Constraints: Physical and Human

As AI capability and competition accelerate, constraints will determine the trajectory of AI development and the outcome of the AI superpower race.

Physical Constraints

The physical foundations of intelligence production at scale present significant challenges. These include:

• Energy: The "energy wall" is a critical bottleneck. Cheap electricity is essential for AI inference and training. China's substantial energy production capacity, including over 300 gigawatts of solar capacity added in 2025, significantly outpaces the U.S. and EU. The U.S. is also increasing its energy production, with over 400 gigawatts requested for the Texas grid, 70% of which is for intelligence factories. The global energy landscape is shifting, with hyperscalers investing directly in energy production, including small modular reactors and co-location of power sources, to meet the immense demand. This has led to a new model where energy generation is treated as a proprietary component of the tech stack. The White House's "ratepayer protection pledge" with hyperscalers aims to prevent AI energy costs from burdening consumers by requiring hyperscalers to build or secure their own energy supplies.
• Semiconductors: A critical chokehold exists, with over 90% of the world's most advanced semiconductors produced in Taiwan. Efforts to re-industrialize the U.S. and reshore chip production are underway.
• High Bandwidth Memory (HBM): Over 80% of HBM is controlled by two South Korean companies, posing another supply chain constraint.
• Infrastructure Components: Delays in AI factory construction are occurring due to shortages of essential components like power transformers and switches needed for stable energy delivery.

While innovation, capital, and human will can overcome physical boundaries, these constraints are substantial and require significant investment and strategic planning.

Human Constraints: Talent and Trust

The "people" constraint is arguably the most underestimated. It encompasses two key areas:

• Talent Shortage: The development of industrial intelligence requires a vast workforce, extending beyond elite AI researchers. The U.S. may face a shortage of skilled tradespeople, such as electricians and plumbers, needed to build and maintain intelligence factories. An estimated 150,000 electricians are required for the projected 90 gigawatts of intelligence factory capacity coming online in the U.S. over the next two years. This highlights a potential "blue-collar flip," where these workers become critically valuable.
• Public Trust Deficit: In Western democracies, a significant public trust deficit exists regarding AI. Low public buy-in, particularly in the U.S. (around 30% support compared to 87% in China), poses a substantial risk. The narrative has often focused on AI's dangers, such as existential risk and job automation, and the perception of AI development as a pursuit of wealthy billionaires rather than a benefit for ordinary people. This has fueled organized, bipartisan political resistance, with over $100 billion in AI infrastructure projects reportedly blocked or delayed. Bills calling for a federal moratorium on intelligence factory and data center buildouts have been introduced. The perception that AI is not "by the people, for the people" is a significant hurdle. This opposition has, in some instances, escalated to violence, with incidents like a house being shot at for supporting data centers and a Molotov cocktail thrown at Sam Altman's house, followed by shots fired.

The lack of public buy-in in Western democracies, where AI is not perceived as a tool for economic prosperity, scientific discovery, or improved quality of life, could lead to the failure of democracies to harness AI's potential. This contrasts with China's clear mission to leverage technology to overcome its historical "century of humiliation." The West lacks a similarly unifying mission, and the risk lies in failing to rise to this transformative moment, rather than in the emergence of a superintelligence.

The Intelligence Divide and the Future of Democracies

The ultimate question for industrial intelligence is what happens when the price of intelligence approaches zero. While a utopian vision of democratized, utility-like frontier intelligence is possible, a significant and growing "intelligence divide" currently exists. The high cost of frontier intelligence, coupled with compute constraints, limits access to a select few enterprises.

Unless the cost of frontier intelligence is reduced and it is distributed as a public good, this divide will widen, leading to severe economic and political ramifications. The populist backlash and growing political movements against AI are partly fueled by the perception that godlike capabilities are accessible only to those who can afford exorbitant annual fees.

The greatest risk to Western democracies is their failure to adapt to this transformative moment. The ability to shape the future of AI development and leverage it for renewed sovereignty is within reach. However, succumbing to fear and viewing AI as inherently too dangerous could lead to missed opportunities. The challenge for Western democracies is to articulate a clear, unifying mission that embraces AI as a tool for prosperity, security, and the preservation of shared values, thereby securing their future for the next century.