PwC’s Dallas Dolan highlights the massive investment and shifting profitability from software to hardware in AI infrastructure, while cautioning about supply chain constraints and the need for multi-cloud strategies to optimize AI adoption and cost management. He emphasizes the growing role of autonomous AI agents in automating enterprise tasks and encourages embracing AI-driven transformation as an opportunity for innovation and workforce evolution.
In this insightful conversation at the Google Cloud Next conference, PwC TMT leader Dallas Dolan discusses the massive trillion-dollar investment underway in AI infrastructure, particularly data centers and telecom. While the scale of spending is enormous, Dolan cautions that not all investments will pay off due to supply chain constraints, including chip shortages and labor shortages for building facilities. He draws parallels to historical infrastructure buildouts like fiber and early internet, emphasizing that while the opportunity is vast, investors must be mindful of risks and bottlenecks.
Dolan highlights a significant shift in the AI value chain, noting that profitability is moving back from software to hardware, especially chips, reversing a long-standing trend where software dominated margins. AI is compressing traditional software moats by enabling enterprises to build capabilities themselves more quickly and cheaply, reducing reliance on large software vendors. However, this shift also brings challenges such as trade bottlenecks and supply chain issues that limit hardware expansion and margin growth. Despite these challenges, there is strong willingness among users to pay premium prices for AI services, which bodes well for frontier model builders like OpenAI, Google, and others.
On the ground, Dolan explains that enterprises are not yet hitting compute or token limits in their AI usage but are instead managing costs carefully to ensure return on investment. PwC itself uses gamification to encourage AI adoption among its 300,000 employees while maintaining strict security controls. When it comes to choosing AI ecosystems, Dolan advocates a multi-cloud and multi-provider approach, leveraging the strengths of different hyperscalers and frontier model providers based on client needs, security, and market demand. He also notes that structural limitations like data center capacity and power availability will eventually constrain growth, but the timeline for these limits is still uncertain.
Dolan offers a practical definition of agentic AI—autonomous AI agents given authority and skills to perform tasks on behalf of users within controlled environments. He shares examples from his own workflow, where multiple AI agents automate tasks such as news aggregation and report generation, orchestrated through Microsoft’s ecosystem. In enterprise, the most promising agentic AI use cases are in back-office finance functions, front-office marketing and sales campaigns, and legal contract review, where agents can significantly reduce manual effort and improve efficiency.
Finally, Dolan reflects on the broader societal and workforce implications of AI automation. He acknowledges the disruption but draws historical parallels to previous technological shifts, emphasizing human adaptability and the creation of new jobs. He celebrates the “shadow AI” innovators within organizations who push boundaries and build impactful solutions despite constraints, viewing them as key drivers of progress. Dolan concludes on an optimistic note, encouraging enterprises to embrace AI challenges as opportunities for growth and transformation rather than threats.