For the past two years, the conversation about generative AI has been dominated by one piece of hardware: the GPU.
GPUs provided the parallel computing needed to train large language models, and their scarcity quickly became an indicator of AI readiness.
But that shorthand is now incomplete.
Managing partner of Cardiff.
The next phase of enterprise AI won't be defined by accelerators alone.
It will consist of CPUs, memory bandwidth, cloud capacity, networks, and workflow systems that enable AI to move from casual experimentation to daily business operations.
The true economic impact of AI will not come from access to models; It will depend on whether companies can turn AI into a reliable and profitable operational capability.
AI is becoming an infrastructure problem
The first wave of generative AI adoption was largely experimental. Employees used standalone tools to compose emails, summarize documents, or write code. These ad hoc use cases were useful, but they did not require companies to redesign the way work is actually done.
The next wave is different. As AI moves deeper into business workflows, IT infrastructure requirements become exponentially more complex.
A customer service tool that composes a response is simple. An AI system that reads account history, checks policy, updates a CRM, records interaction, and triggers a follow-up task is a completely different beast. This system not only needs a powerful model; requires IT orchestration, secure data access, software integrations, permissions, audit trails, and fallback logic.
This is where the GPU-centric view fails. While GPUs remain critical for heavy inference, CPUs coordinate how these workloads interact with databases, APIs, security layers, and operating systems. As a result, memory bandwidth, latency, and power availability are becoming true strategic constraints.
The high cost of using unstructured AI
The first business playbook was simple: give employees access to powerful tools and see what happens. While this accelerated learning, it also exposed enormous financial vulnerability. Individual, unstructured prompts are expensive, difficult to measure, and difficult to link to tangible business results.
We are seeing a major corrective shift among the tech giants. Microsoft recently began retiring internal licenses for Anthropic's Claude Code, which cost between $500 and $2,000 per engineer monthly due to high token consumption, and is forcing its Experiences and Devices division to transition to GitHub Copilot CLI before its fiscal year end on June 30.
Similarly, Uber completely exhausted its budget for AI coding tools in just four months. The ride-hailing giant deployed Claude Code to approximately 5,000 engineers and aggressively encouraged adoption through internal leaderboards. The experiment was incredibly effective (assisted systems generated nearly 70% of the committed code), but token usage increased faster than anyone anticipated, forcing Uber leaders to publicly question the net return on investment.
Consequently, the future of enterprise AI will shift from fragmented indications to a central intelligence model. Instead of thousands of disconnected interactions, companies will rely on shared layers of intelligence: centralized systems that understand corporate data, apply consistent business rules, route tasks between applications, and track performance.
This model is inherently more efficient because the same intelligence is reused across workflows rather than being recreated from scratch by individual users.
From responses to workflows
The most critical change in enterprise technology is the transition from tools that answer questions to systems that do work.
Traditional software is deterministic: a user clicks a button and a system performs a known action. AI workflows are more dynamic. An agent workflow can retrieve data in real time, reason through a multi-step process, interact with third-party software, and connect to a human for approval.
This puts immense pressure on the entire technology. To unlock real productivity gains, businesses need clean data infrastructure, disciplined governance, and strong integrations. Advanced models are useless if they are layered on top of fragmented and disconnected corporate systems.
Unprecedented change management and “AI-native” workforce
As these agent systems mature, the impact on global employment will trigger a corporate change management crisis on a scale never seen before. AI will fundamentally alter hiring patterns and role requirements long before it eliminates large-scale workforces.
Historically, headcount was the default lever for expanding capacity; more customers required more support staff. AI breaks that linear relationship. Instead of asking how many people are needed to handle an influx of volume, leaders will increasingly ask how much of a process can be handled by automated systems.
This environment will aggressively reward adaptability. Professionals who stay on the cutting edge of technology, learn to design AI-based workflows, and manage systemic exceptions will benefit disproportionately.
In contrast, the risk of displacement is greater for those who rely solely on inherited industry experience. Traditional technical and management paradigms are being disrupted by a new cohort of AI-native developers, product managers, and team members. These professionals not only use AI as an assistant; They build, manage, and think in terms of automated, model-based systems.
Those who fail to transition from traditional operators to native AI orchestrators risk being replaced by those who do.
AI infrastructure is economic infrastructure
The broader economic impact of AI will be determined by how deeply it can be integrated into the core systems that run global companies.
GPUs, CPUs, networks, and data centers form the physical foundation. Agent orchestration, security and observability form the operational foundation. Together, they determine whether AI remains a novelty or becomes a scalable business capability.
The GPU race was simply the opening chapter of the rise of AI. The next chapter will be defined by the holistic compute, data, and workflow systems that enable AI to do real work at scale. That is the moment when AI stops being a tool and truly becomes infrastructure.
Check out our list of the best IT automation software.
This article was prepared as part of TechRadar Career Insightsour channel to feature the best and brightest minds in today's tech industry.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing, find out more here:
