By Lina Irawan 
Washington and Silicon Valley have found a new fight to pick over artificial intelligence in “pacing,” or the deliberate throttling of frontier model development until safety, alignment, and society at large can catch up, yet this contentious debate fundamentally misinterprets the true dynamics of AI integration and economic value creation. To its detractors, pacing is a dangerous act of unilateral disarmament in the high-stakes global race with China, potentially ceding technological superiority and future economic dominance. To its champions, however, pacing represents the only responsible and ethical path forward for a technology whose own pioneering creators issue stark warnings of catastrophic and even existential risks. This schism, while seemingly logical, is built upon a shaky foundation, revealing a deeper misunderstanding that could jeopardize the very future of AI’s beneficial deployment.
Both camps, despite their opposing viewpoints, have inadvertently fallen prey to what can be termed the “Compute-to-GDP Fallacy”—the mistaken and often overly optimistic belief that every incremental leap in AI model performance immediately and linearly translates into measurable macroeconomic output and productivity gains. This perspective ignores the historical precedent set by every prior general-purpose technology, from electricity to the internet, which consistently took decades to fully diffuse into the economy and generate significant, measurable productivity increases. While AI is undeniably accelerating this diffusion curve, there’s a pervasive sense that the fundamental laws of history or economics, which govern how new technologies are adopted and integrated, somehow do not apply this time around. This "AI exceptionalism" risks leading policy makers and industry leaders astray, diverting attention from the real challenges and opportunities at hand.
In stark reality, Corporate America finds itself years, if not a full generation, behind the bleeding edge of AI frontier research and development. The commercial fortunes of leading AI labs will ultimately be decided not by raw computational prowess or the sheer scale of their models, but by the far more challenging and nuanced tasks of earning trust and achieving widespread, meaningful adoption within existing enterprise structures. From this vantage point, a strategic period of pacing would cost the economy remarkably little in terms of immediate productivity loss. Conversely, racing ahead without adequately addressing fundamental safety, ethical alignment, and robust integration challenges could incur far greater costs—ranging from regulatory penalties and reputational damage to systemic risks and potential societal disruption. The current debate, fueled by either arrogance or a profound misdiagnosis of the situation, is therefore fundamentally misdirected.
The suspicions harbored by pacing skeptics are not entirely frivolous. Legitimate questions arise: Is "pacing" a genuine commitment to safety, or a shrewd marketing gambit employed by frontier labs and cybersecurity companies to polish their financials ahead of anticipated IPOs, capitalizing on growing public anxiety? Would a U.S. commitment to pacing truly cede its hard-won lead in AI to China, or could it foster a reciprocal commitment from Beijing to also pace its own development, leading to a more stable global landscape? These are critical geopolitical and economic considerations that demand careful analysis, but they must be viewed through a lens of pragmatic reality rather than speculative fear-mongering.
The Frontier Problem: Mismanaged Messaging and Public Trust
AI has undeniably reached a critical capability milestone. Warnings of catastrophic or even existential risk, once confined to science fiction or fringe academic circles, can no longer be dismissed outright, even if their near-term probability remains modest and subject to intense debate. However, by focusing almost exclusively on cutting-edge models and pushing the boundaries of raw capability, frontier labs have inadvertently mismanaged both their public messaging and, crucially, public trust. Extensive discussions, including over 100 recent conversations with CEOs, policy leaders, and AI scientists for an upcoming book, When Machines Act, reveal a consensus: the pacing debate has largely lost sight of first-principles thinking, becoming entangled in the hype cycle rather than grounded in practical implementation challenges. This disconnect between what AI can do and what enterprises need it to do is at the heart of the current fallacy.
The Alignment Problem: Enterprise Integration is a Slog
Since the watershed release of OpenAI’s ChatGPT in late 2022, corporate leadership worldwide has scrambled to respond with a speed unmatched in modern commercial history. Executive suites have mobilized with unprecedented urgency, yet they face the immutable structural physics of enterprise architecture. This includes fragmented data silos, decades-old legacy Enterprise Resource Planning (ERP) systems, stringent compliance regimes, and often, a basic lack of data hygiene. These realities make true economic absorption and transformative integration of advanced AI an inherently slow and arduous slog. As corporate budget shocks from runaway “tokenmaxxing” demonstrated, many daily enterprise workflows require far simpler, more cost-effective models. Indeed, precious few tasks at the average Fortune 500 company genuinely demand a frontier system operating at the bleeding edge. Therefore, a period of pacing would neither significantly harm economic output nor choke off the labs’ commercial revenues, precisely because enterprises need considerable time simply to assimilate and responsibly deploy the capabilities already widely available.
The data underscores this challenge: Among high-performing companies, more than two-thirds identify data readiness and quality as the primary barrier to successfully implementing AI. This figure has proven remarkably stubborn, persisting even as the models themselves have made dramatic leaps forward in capability. A mere 7% of enterprises describe their data as “completely ready” for AI, according to a report from Cloudera and Harvard Business Review Analytic Services. Furthermore, fewer than a quarter of companies possess a comprehensive data strategy, and a staggering 63% either lack AI-suitable data management practices or are unsure if they have them. These foundational issues, not the latest model’s benchmark score, dictate the pace of real-world AI impact.
The Fallacy Problem: Economic Transformation Takes Time
As McKinsey Senior Partner Asutosh Padhi emphasized on air with Fareed Zakaria, technical availability is fundamentally different from economic transformation. General-purpose technologies (GPTs) have historically required decades to reorganize workflows, transform industries, and generate broad-based productivity gains. For example, electricity took approximately 75 years to lift productivity economy-wide after its invention. Computers required nearly 50 years to move beyond niche applications and substantially impact white-collar productivity, a period famously dubbed the "productivity paradox" in the 1980s. The internet and mobile devices, while faster, still demanded 25 years or more to reach their full transformative potential. The underlying AI models may be technically ready, but the systemic organizational restructuring, workforce reskilling, and cultural shifts they demand will inevitably take substantial time. When McKinsey surveyed the business community, the firm found that only 6% of companies reported a "significant" impact from AI, with only modest earnings attribution. This highlights the chasm between raw technological potential and realized economic value.
Companies are pragmatically concentrating their efforts on high-reward, low-risk automation tasks that models one or two generations old can already solve effectively and affordably. As one highly respected former Wall Street CEO noted, these systems will often run in parallel with existing legacy systems for years to confirm their correct operation and ensure no regulatory risk is unknowingly absorbed. This cautious, iterative approach is a natural and necessary part of enterprise adoption.
A parallel dynamic has emerged in the economics of silicon. Older-generation chips, initially cast aside in the frantic scramble for cutting-edge AI accelerators, are finding a valuable second life as workhorses for the practical inference tasks that dominate enterprise demand. As Growth Protocol founder and CEO Miro Dimitrov noted at a recent Yale CEO Caucus, deploying neuro-symbolic architectures has allowed his enterprise reasoning platform to slash inference costs by roughly 80-fold in live client deployments, largely by shifting workloads off ultra-expensive GPUs and onto more readily available and cost-efficient everyday enterprise CPUs. This demonstrates that optimal AI deployment often prioritizes efficiency and practicality over raw, cutting-edge compute.
The Three Phases of AI Adoption: A Roadmap to Trust
Corporate AI adoption is best understood not as a monolithic event, but in three distinct phases, each distinguished by how much work a company can responsibly hand over to AI, gated by data readiness and the trust the systems have earned.
- Assistance: This initial phase consists of off-the-shelf copilots and intelligent assistants that ride atop existing enterprise platforms such as Salesforce. These tools connect data across existing applications and enable employees to work faster and more efficiently with minimal re-architecting of core systems. Payback typically arrives quickly, and the risk remains modest, as a human is still performing much of the critical work and retaining ultimate oversight.
- Orchestration: The second phase covers more complex, agentic workflows that demand real investment. This involves structuring proprietary data, connecting far-flung data lakes never originally designed to interact, and building sophisticated automation pipelines. In this phase, a human remains "in the loop," approving each consequential step and providing critical judgment.
- Autonomy: The final and most advanced phase brings end-to-end agentic operations across seamlessly interconnected systems, with humans supervising by exception rather than actively participating in every step. This requires the highest levels of trust, robust safety protocols, and deeply integrated, self-correcting AI systems.
Reward compounds with each successive phase, but so do the inherent risk and the level of trust required from leadership and employees. This is why the average Fortune 500 CEO currently remains firmly in the first phase, making sizable but early investments to prepare their organizations for the eventual transition to the second. Furthermore, these phases do not advance in lockstep across an entire enterprise. Most companies will oversee a multi-phased portfolio, piloting orchestration in select forward-leaning departments as the rest of the organization becomes comfortable with basic assistance.
Underlying all three phases is the fundamental need for CEOs to trust that AI will perform as any other employee would—following the guidelines spelled out in the employee handbook, obeying the rule of law, and maintaining a foundational layer of human values and judgment. So, when markets, media commentators, or investors fret that pacing for AI alignment will hinder progress in frontier models, they misdiagnose how enterprise value is truly created. This concern often confuses the vague promises and threats of "AGI" with the sublime opportunities and genuine catastrophic risks of reaching the "singularity." If AI technologies are meant to automate human tasks, they should be held to the same stringent standards of values, judgment, and moral alignment as any current or prospective human employee. If the most advanced frontier models cannot meet those standards in a controlled testing environment, they are simply not ready for widespread release—which is precisely why the U.S. has always maintained robust laws protecting consumers against dangerous and defective products.
As former FTC Chair Lina Khan reminded the public on X: “There is an extensive set of laws that govern dangerous and defective products… releasing unvetted AI models or agents can violate consumer protection laws. Shipping flawed AI tools without implementing adequate measures to detect and stop rogue or defective AI agents can be an ‘unfair or deceptive’ act or practice under the FTC Act (and analogous state laws).” These established laws hold companies strictly responsible for harm done to consumers, employees, investors, patients, competitors, and the critical markets and financial systems on which they all depend. The legal and ethical imperative for safe AI is not a new concept; it is an extension of existing product liability and consumer protection frameworks.
America vs. China: Speed vs. Trust in the Global AI Race
Wherever one lands on the complex China distillation debate, the undeniable fact remains that China now fields AI models rivaling the frontier systems on the market today. Indeed, the Chinese Communist Party (CCP) has signaled a strategic pivot, turning its energies more towards diffusing AI throughout its economy and society rather than solely continuing to push the absolute frontier of model capabilities. The leading frontier labs in the West must recognize that two distinct races are underway simultaneously: one to reach "AGI" or "superintelligence," and the other for market share and "share of wallet" within the vast enterprise landscape. As their most established customers, largely Fortune 500 enterprises, can attest, recapturing a customer after a decisive adoption decision has been made is extraordinarily difficult. China understands this dynamic well, a lesson powerfully demonstrated by its victory in the global telecommunications race, which the U.S. largely lost.
The CCP has also expressed deep concern over AI alignment, albeit with a different ideological lens. Beijing’s domestic alignment priorities center on state control, party orthodoxy, and narrative consistency. In contrast, Western alignment primarily focuses on fiduciary reliability, consumer safety, and product liability. Yet, beneath this significant ideological gulf lies an identical commercial and operational reality in both systems: unpredictable, "hallucinating" agents that fail to adhere to organizational rules, judgment, and institutional guardrails simply cannot be trusted to run mission-critical workflows or drive durable economic growth. Both nations recognize that unreliable AI is ultimately useless, or worse, dangerous.
The Trump administration should still actively pursue avenues to collaborate and coordinate with President Xi to ensure that no mass destruction or catastrophe, intentional or accidental, issues from advanced AI. Whether Trump will raise the matter directly is another question entirely. At a recent CEO Caucus, 93 of the roughly 100 CEOs surveyed did not believe the president was correct to classify warnings about AI’s dangers as a "hoax." Almost 90% said the president should press the need for joint AI-safety guidelines with China during the state visit, yet nearly three-quarters did not expect him to do so. Based on preparatory discussions between Treasury Secretary Bessent and his Chinese counterpart, however, those 75 business leaders may soon be gladly proven wrong, suggesting a potential for pragmatic cooperation on this critical issue.
A pacing interval is far from a passive holiday or an economic ceasefire; it must be conceived as an active defensive hardening window. Both Washington and Beijing urgently need intentional breathing room to allow their critical infrastructure—across financial, healthcare, and education systems—to build resilient defenses against sophisticated autonomous agentic exploits before the next generation of frontier capabilities is unlocked. And instilling human alignment is only half the battle. The immediate priority during such a period must be fortifying the institutions that serve as the bedrock of civilization. These vital systems should be rigorously probed for vulnerabilities by the most advanced AI models, much like Anthropic demonstrated through its restricted deployment of Mythos under Project Glasswing. These models may breach defenses through impressive engineering ingenuity, but their exploits succeed only because of systemic weaknesses and vulnerabilities in existing corporate digital infrastructure.
The frontier labs, driven by their ambitious vision, often believe they are running a single, singular race toward superintelligence. However, the race that will ultimately decide their fortunes—earning the enduring trust of the enterprises, regulators, and citizens who must live with what they build—is a far slower and more complex endeavor. Speed may win headlines and early investment, but trust earns sustainable market share and lasting value. This race is unequivocally a marathon, not a sprint. As past runners ourselves, we understand that the first half of any long-distance race is dedicated to strategic pacing, and the second half is for making decisive passes. In every technological revolution, from the expansion of the railroad networks to the advent of the internet, the greatest and most enduring fortunes were amassed by those who profoundly understood that a new frontier is ultimately worthless until the settlers, the users, and the wider economy truly arrive and integrate its potential.
