Estimated reading time: 6 minutes

SMT Perspectives & Prospects: 12 Predictions for Using AI in 2026

At a committee meeting in Washington, D.C., someone asked me if I think AI is a bubble. Over the past several months, this has become a prevalent and pervasive question in both business circles and broader society. Yet, despite its ubiquity, no one had ever asked me that question directly, and especially with such immediacy in front of a committee. My instinctive reply was, “It depends on how you define the bubble. It is certainly not an internet bubble.” The querier lamented, almost disappointed, “Well, you are in the minority.”

The question has lingered with me, prompting deeper reflection. Is AI a bubble? Could it become one? If not, why not?

In the broader arc of AI, we are in the early stages of what it can accomplish. AI is not merely technology; it is shaping an era, and like any powerful force, it will be as benign or malignant as its creators and users. As Winston Churchill observed, “The farther backward you can look, the farther forward you can see.” Since AI captured worldwide attention three years ago and has become a global sensation, the pace of advancement has been transcendent. Investment has surged from big tech companies and start-ups to a variety of business sectors and governments worldwide.

This acceleration has led to large-scale deployments across a wide range of industries, with meaningful productivity gains starting to materialize. AI’s speed and scale are nothing short of astonishing. Yet if we define a bubble as a situation where some investments don’t generate an ROI, where many start-ups will not survive, and certain stock valuations are irrationally exuberant and will inevitably adjust, then yes, aspects of that will occur. That is the nature of innovation cycles. But does that make AI itself a bubble?

Here are the top 12 things about AI we can expect in 2026, in order of increasing scope and challenge.

1. Viable AI use cases will grow, and deployment will expand, albeit not distributed evenly. AI technology is being used in real-world applications, affecting diverse industries and sectors. Productivity gains will kick in.

2. LLMs will advance, yet SLMs will proliferate. 2025 was the year of LLMs. In 2026, SLMs will grow. A vast number of parameters that serve as the model’s knowledge bank characterize LLMs. They require high computing power and parallel computing (GPU). SLMs reduce the expense of inference by requiring a lower number of parameters, concentrating on a specific target task to add value. SLMs also facilitate deployment at the edge. They can run on CPU, inference-centric chips (such as LPU), AI-specialized ASICs, or customized chips.

3. Multimodality models will become mainstream. With the release of Google’s Gemini 3 on Oct. 22, 2025, we see the model for advancements in reasoning, deep thinking, and processing and integrating multiple forms of data simultaneously, including text, images, audio, and video. These allow for a more contextual and human-like comprehension. Advancements in these capabilities will continue to gain momentum.

4. Generative AI tools will be ubiquitous. In 2025, we enjoyed the convergence of search and chat. With fierce competition and robust investment, the market is expected to see a proliferation of multimodal AI tools, which will enhance end-user interaction and data processing by enabling the simultaneous handling of multiple input types (e.g., text, images, charts, graphs, and documents such as PDFs). This will lead to improved responses and has broad applications across various industries, including manufacturing, engineering, and document analysis (e.g., interpreting complex technical manuals or financial reports).

5. Connectivity and data quality will be crucial to building a “private” model at the edge. To be cost-effective and time-efficient, the “distillation process” and other emerging techniques are taking good, large models to make small models smarter at domain-specific tasks. To leverage AI at the edge, robust 5G connectivity and data management are crucial.

6. There will be a demand for tools to reduce hallucinations. Blending text generation with information retrieval using Retrieval-Augmented Generation (RAG), such as the ChatGPT Retrieval Plugin and other emerging techniques, will enhance the accuracy and relevance of AI-generated content.

7. AI Agent(s) will be deployed toward an agentic AI system. Achieving “Think, Plan, Reason, Adapt, and Act” autonomously while retaining memory by understanding the task, breaking it into steps, choosing the right tools, and executing and learning from feedback will be the key to success. This means, “Take real-world actions, not just answers.” There have been several developments of AI agents and agentic AI workflow tools for simple or complex workflows, respectively, for example: (Table 1)

8. Electronics manufacturing will adopt and flourish in running AI workloads outside data centers. These will be closer to manufacturing sites by integrating network, edge computing, and AI algorithms. “Agent” platforms distribute heavy workloads across a network of linked computers while reducing latency by feeding real-time data directly back into AI models at the source. Once we train a model, we can deploy subsets closer to the generation of new data. Synchronized workflow in robots, IoT, and AI will see more deployment. For example, the tasks for warehouse applications include:

• Running an AI-powered computer vision system allows the warehouse to identify and handle thousands of packages and inventory items
• Aiding warehouse workers in picking orders
• AI driving warehouse robots running onsite
• Improving overall worker safety

9. 2026 will be the year of advanced packaging and co-packaging. Bandwidth-hungry AI applications are driving the need for high-speed data transfer, leveraging the speed of light to deliver greater bandwidth, lower latency, and reduced power consumption. Hardware components are connected via copper interconnects, while the connections between the racks in data centers often use optical fiber. Advanced packaging and co-packaging incorporate optical interconnects at the CPU and GPU levels, utilizing optical signals, which drives the need for near-packaged optics with high-performance PCB substrates (or non-PCB interposers) on the host board, as well as photonic integrated circuits (PICs). Co-packaged optics (CPO)—a single package integration of electronic and photonic dies to enable connectivity within data centers at the speed of light through optics—complements existing short-reach electrical wires.

10. The strategy of bundling chips (vis-à-vis cutting-edge chips) will shape the competitive landscape. Whether bundling many chips can compete with the leading chips will create additional competitive pathways in the AI race.

11. The role of computing power, and thus, electric power, will intensify. The need for data centers and high computing power leads to an ever-higher demand for electric power and increasingly fierce competition between the U.S. and China. According to the International Energy Agency/Energy Information Administration, as of 2025, China has the largest power grid in the world, with 3.75 terawatts of power generation capacity, more than double the U.S. capacity. It commands a low electricity cost of $0.03/kilowatt-hour versus the U.S. cost of $0.07–0.09 kilowatt-hour.

12. Breakthroughs in architectural pathways beyond LLMs may emerge. Examples include leveraging quantum advantage for specific AI optimization tasks or exploring other architectures to advance AI abilities closer to human intelligence.

In the current landscape, AI cannot learn and act like humans. Holistically, AI is not just a technology; it’s a force that will affect every aspect of human life and work.

Well, am I in the minority?

Appearances
On March 15, Dr. Jennie Hwang will deliver Professional Development Courses on “Artificial Intelligence & AI-Powered Electronics Manufacturing” and “High Reliability Electronics for Harsh Environments,” at APEX EXPO 2026. She will also deliver a Professional Development webinar on “Reliability of Electronics: The Role of Solder Joint Voids,” Feb. 17 and 19, 2026, for the Global Electronics Association.

This column originally appeared in the January 2026 issue of SMT007 Magazine.