The AI supercycle is entering a new chapter. At Mizuho’s 2026 Technology Conference, leaders from across the technology ecosystem — including Lumentum, Salesforce, IBM, and others — described a market shifting from building AI capability to deploying it at scale across enterprises, data centers, and the infrastructure that connects them.

Just a few years ago, AI adoption was defined by pilots and productivity tools. Today, companies are embedding AI into workflows, reallocating budgets around deployment, and confronting the operational, economic, and physical constraints of increasingly autonomous systems. The result is a new phase of AI adoption defined less by possibility and more by integration.

AI Becomes the Operating Layer

In many vertical markets, AI is becoming part of the operating system of the business itself.

Take the payments ecosystem, for example. In restaurants, AI-powered devices can listen to customer orders and populate point-of-sale systems automatically, while AI agents embedded within merchant platforms help business owners run analytics, marketing, and reputation management through natural-language interactions.

Knowledge work is changing in similar ways. Work that once required teams of dozens can be completed by much smaller groups aided by digital workers and agents. Consulting was cited as a prime example, with AI reducing staffing for engagements while lowering delivery costs.

However, lower costs do not necessarily imply less work. As the cost of knowledge work falls, demand for it may rise — a version of Jevons paradox. One speaker estimated that AI could displace roughly 10% of jobs over the next decade while creating more than enough new work to offset it, provided workers are reskilled.

As AI becomes embedded in everyday work, it’s also changing the economics of software. Because agents can continue operating long after employees log off, AI spending is becoming harder to forecast. Enterprises often don’t yet know how many workflows they will automate or what those workloads will ultimately cost. That uncertainty is also challenging traditional pricing models. Unlike seat-based licensing where revenue is tied to the number of users, usage-based models expand as customers build more agents and workflows. That creates a compounding effect while raising new questions around budgeting and ROI.

As AI moves into mission-critical workflows, reliability becomes just as important as intelligence. While consumers may tolerate occasional inaccuracies, enterprise applications often cannot. In fields such as semiconductor design, engineering, and financial services, precision matters more than creativity. As one executive observed, "you can't vibe-code your way into a chip." That shift is driving greater emphasis on governance, security, and trust.

The Next AI Bottleneck is Physics

While graphics processors remain essential, discussions across the semiconductor and networking ecosystem highlighted a broader reality: many of AI's next challenges are no longer computational, but physical.

One frequently discussed topic was the industry's growing "memory wall." Today's AI systems are constrained by both memory capacity and how far memory can physically reside from compute. New architectures designed to disaggregate memory from processors could expand memory pools from hundreds of gigabytes to multiple terabytes while improving the economics of large-scale inference.

The challenge also extends beyond memory, as the bottleneck increasingly becomes the movement of data itself as AI systems scale. Solving that problem is driving innovation across networking, interconnects, and optical technologies designed to move larger volumes of data with greater speed and efficiency.

Photonics illustrates that shift. The industry is entering what one executive called a "photonic supercycle," driven by the growing need to replace traditional electrical connections with optical ones as copper reaches its physical limits over meaningful distances. And future performance gains may depend as much on breakthroughs in memory, interconnects, and photonics as on advances in the models themselves.

What's Next in Tech?

Innovation is also abundant in the tech world outside of the headline-grabbing AI advances.

Digital assets remain a source of optimism and excitement as the core of emerging digital infrastructures. Quantum computing is moving beyond theory, with advances in scientific modeling — including protein modeling that has progressed from a handful of atoms to tens of thousands — suggesting quantum systems may eventually solve problems that are uneconomic or infeasible for classical computers. Rather than replacing AI, CPUs, or GPUs, quantum may become a specialized computing layer for the hardest scientific and industrial challenges, which helps explain growing government interest in quantum infrastructure.

Despite ongoing questions around costs, regulation, and workforce transformation, companies appear unlikely to slow investment. In a fast-moving technology cycle, standing still can mean ceding customers, market share, and future growth. As one executive put it, “the greatest risk for businesses is not taking one.”

‍