After more than two decades helping commercialize transformative computing technologies at Arm, I began noticing the same pattern repeated across every major technology shift. Innovation moves quickly. Organizations rarely do. That observation explains why I now focus on AI governance and assurance. Enterprise AI is following many of the same dynamics that reshaped the semiconductor industry decades ago. This article explores those parallels and why leadership, accountability, technical excellence, and disciplined execution, not AI capability alone, will determine which organizations create lasting competitive advantage in the age of enterprise AI.
My two decades at ARM offered a front row seat to one of the most consequential periods in computing history, the transition from vertically integrated semiconductor companies to a global ecosystem built on standardized processor architectures, an intellectual property licensing model, and specialized manufacturing. Working with companies across the entire ecosystem exposed a recurring pattern. Innovation created opportunity, but commercial success depended on how quickly entire industries reorganized around it.
Artificial intelligence is following the same path. The technology is advancing rapidly, but the organizations realizing measurable returns are those preparing themselves to change with it.
Technology rarely determines the winners of an industry transition. Organizational adaptation does. Every major computing shift, from semiconductor IP to mobile and cloud, rewarded organizations that aligned their structures, processes, and success measures ahead of widespread adoption.
That transition reshaped far more than the semiconductor industry. It changed mobile computing, consumer electronics, automotive systems, industrial automation, cloud infrastructure, and eventually nearly every software application that runs on modern computing platforms.
The experience also provided an opportunity to observe how industries respond to technological change. While each wave of innovation introduced different technologies, the underlying pattern remained remarkably consistent.
AI is outpacing previous enterprise technology cycles. New capabilities continue to emerge, new products appear almost daily, and organizations across every sector are racing to identify competitive advantages.
The similarities extend well beyond the technology itself. The largest barriers for many industries are unlikely to be algorithms or computing power. They are organizational adaptation, operational readiness, governance, and the ability to build trust at scale.
This article is the first in a series examining lessons from ARM’s role in the transformation of the semiconductor industry, and why those lessons remain directly relevant to enterprise AI adoption.
Most successful technology disruptions share common characteristics.
A new technology gains commercial traction when it changes the economic equation for adoption. That usually happens in one of three ways:
• It delivers a material capability improvement at the same or lower cost.
• It justifies higher cost through measurable return on investment, new business opportunities, or durable competitive advantage.
• It creates novel capabilities that were previously inaccessible, impractical, or economically unviable.
Clayton Christensen’s disruptive innovation work remains useful here. Disruptive technologies often begin as simpler, cheaper, or more accessible alternatives that serve overlooked markets before improving enough to challenge incumbents.
Looking backward is, therefore, useful. Previous technology transitions provide practical guidance for evaluating current ones.
For those who don’t know the story, when ARM was founded in 1990, the semiconductor industry looked very different.
Most companies developed proprietary processor architectures. Software portability was limited. Chip development required enormous investment. Manufacturing was largely owned by the companies designing the silicon.
ARM introduced a different model.
Rather than manufacturing chips, ARM licensed CPU designs as intellectual property. This enabled semiconductor companies to license standardized CPU implementations sharing a common instruction set architecture and focus their investment on system innovation instead of repeatedly developing proprietary processor cores.
Several industry shifts reinforced one another.
Standardizing around the ARM architecture reduced engineering cost and increased software portability. That in-house CPU design team could then be redirected to system development, increasing capacity to produce more products.
Processor performance improved while maintaining low power consumption and small silicon area, allowing entirely new categories of size, weight and energy-constrained devices to become commercially practical.
A rapidly expanding ecosystem of operating systems, software tools, middleware, and development environments emerged, increasing competition and accelerating innovation. As more companies standardized on the ARM architecture, each additional participant strengthened the value of the ecosystem. Replacing the CPU architecture eventually meant redesigning silicon, rebuilding software, retraining engineering teams, and disrupting supply chains, making switching costs increasingly difficult to justify.
At the same time, the semiconductor industry itself reorganized.
“Real men have fabs” – AMD founder Jerry Sanders’ famous line captures the traditional vertically integrated manufacturing mindset of the 1980s and 1990s. This gradually gave way to independent device manufacturers, fabless semiconductor companies such as Qualcomm and Nvidia, and contract manufacturing leaders such as Taiwan Semiconductor Manufacturing Company (TSMC). Within a decade, the economics of semiconductor development had fundamentally changed.
Moore’s Law amplified every one of these trends. The steady increase in transistor density allowed processors to deliver exponentially greater capability while reducing cost per computation. By the time the first iPhone landed in 2007, many of the technical and commercial foundations had already been established.
The mobile revolution was not created by one breakthrough. It resulted from multiple technological, business, and ecosystem changes reinforcing each other over many years. The broader lesson is that the architecture eventually became less important than the ecosystem that formed around it. The organizations that recognized this early positioned themselves to benefit long after the underlying technology became widely available.
Artificial intelligence has experienced its own sequence of reinforcing advances.
Machine learning and deep learning have steadily improved prediction, computer vision, speech recognition, and language processing over many years.
The transformer architecture represented another major inflection point by allowing models to scale in capability far beyond previous approaches.
Cloud computing supplied the expansive computational infrastructure required to train and deploy increasingly complex models without requiring organizations to build massive computing environments themselves.
Then ChatGPT changed the market.
It demonstrated sophisticated AI capabilities through an interface that almost anyone could use immediately and at practically zero cost. The result was the fastest consumer application adoption recorded to date.
The underlying research had existed for years. Cloud infrastructure had matured. Large language models had become practical. ChatGPT combined those ingredients into a product that made the technology instantly accessible to millions of users.
Just as the smartphone became the visible platform of decades of semiconductor and system innovation, ChatGPT became the visible expression of decades of AI and cloud infrastructure research.
Technology attracts attention because it is visible and attracts billions in investment capital.
Organizational change is where most implementation effort occurs.
The semiconductor industry illustrates this clearly.
Semiconductor intellectual property required new engineering workflows, verification methodologies, interoperability standards, software ecosystems, and business models.
Contract manufacturing required companies to redesign supply chains, procurement processes, quality management, and commercial relationships.
The standardized CPU architecture enabled software reuse across multiple product categories, fundamentally changing software development economics.
None of those changes were purely technical.
Each required organizations to modify structures, processes, responsibilities, supplier relationships, and investment priorities.
For most enterprises, AI presents a similar challenge.
Cloud infrastructure removed much of the capital expenditure previously required for advanced computing.
Foundation models continue to grow in complexity, much as transistor density increased during earlier generations of semiconductor development.
Increasing model capability creates opportunities across more business functions. At the same time, organizations inherit additional governance requirements, validation activities, operational controls, and accountability obligations. Frameworks such as the NIST Artificial Intelligence Risk Management Framework (NIST AI RMF 1.0) recognize that effective AI adoption depends on integrating governance, measurement, testing, evaluation, verification, and validation throughout the AI lifecycle, rather than treating deployment as a purely technical exercise.
This balance between technical progress and organizational readiness increasingly determines whether AI initiatives scale successfully, as recent research consistently exposes – MIT found 95% of generative AI pilots fail to improve P&L, largely due to integration and governance gaps rather than model quality; McKinsey found CEO-level AI governance is one of the factors most correlated with bottom-line impact; and BCG’s analysis of AI value creation attributes 70% to organizational and people factors, with just 10% to the algorithms themselves.
Several observations from previous computing transitions remain highly relevant.
First, organizational foundations usually change before technology adoption accelerates. Organizations should evaluate operating models, processes, infrastructure, and supporting governance practices before investing heavily in AI. While this exercise may not exclude trial and error, it will mitigate blind ambition.
Second, invest in systems rather than isolated applications. Standardized platforms consistently outperform collections of disconnected point solutions as adoption expands.
Third, proprietary data deserves as much attention as models. Foundation models are increasingly available to everyone. Unique organizational data is far more difficult to replicate and often provides longer-term competitive differentiation.
Fourth, target operational friction. The strongest early AI use cases reduce repetitive work and improve decision support, allowing organizations to redirect skilled employees toward activities where human judgment and context have the greatest impact.
Fifth, pursue incremental success before broad deployment. Organizations that demonstrate measurable value from controlled implementations are better positioned to scale responsibly. Sandbox demos often fail under real-world conditions; control the controllables.
Finally, evaluate AI investments using the same discipline applied to any major technology initiative. Expected benefits must comfortably exceed implementation costs, operational disruption, governance overhead, and ongoing management requirements. When that equation no longer holds, redirect investment elsewhere.
The headlines often emphasize model capability or benchmarks. History suggests a different conclusion.
New technology creates opportunity.
Organizations determine whether that opportunity becomes measurable business value.
In the next article, I will examine another lesson from semiconductor development that has become increasingly relevant to enterprise AI. As processor and system complexity increased, testing, verification, and validation consumed a growing proportion of engineering effort. Enterprise AI is beginning to exhibit many of the same characteristics, with implications for governance, risk management, and long-term return on investment.