From Semiconductor Demand to AI Agents: What Comes Next for Technology
- Joel Polanco, Senior Hardware Product Manager in the Data Center & AI Computing Group at Intel
- AI and Automation, Content Strategy and Marketing, Customer Experience (CX), Digital Transformation, Managing Org Transformation, Managing Your Career, Martech Stacks, Other Martech, Strategy and Trends
When Joel Polanco, Senior Hardware Product Manager in the Data Center & AI Computing Group at Intel, began his career, manufacturing was not considered the most desirable path in technology.
After completing his undergraduate and graduate studies at Arizona State University, Joel had already finished three internships, including one at Intel in Chandler, Arizona. That experience gave him an early look at tech manufacturing and helped shape a career decision that, two decades later, feels especially relevant.
At the time, Joel said, manufacturing was not seen as desirable as software, but that was exactly what drew him to Intel. “I wanted the opportunity to be in manufacturing because I thought it might be going away,” he said.
Now, the opposite is happening. Manufacturing is returning to the United States, semiconductor demand is surging, and Intel is positioning itself at the center of one of the most important technology transformations in decades. “What’s amazing here, twenty years later, is the manufacturing’s coming back and so is Intel,” Joel said, calling this “one of the most exciting points in my career.”
After completing his undergraduate and graduate studies at Arizona State University, Joel had already finished three internships, including one at Intel in Chandler, Arizona. That experience gave him an early look at tech manufacturing and helped shape a career decision that, two decades later, feels especially relevant.
At the time, Joel said, manufacturing was not seen as desirable as software, but that was exactly what drew him to Intel. “I wanted the opportunity to be in manufacturing because I thought it might be going away,” he said.
Now, the opposite is happening. Manufacturing is returning to the United States, semiconductor demand is surging, and Intel is positioning itself at the center of one of the most important technology transformations in decades. “What’s amazing here, twenty years later, is the manufacturing’s coming back and so is Intel,” Joel said, calling this “one of the most exciting points in my career.”
Watch the full interview with Intel’s Joel Polanco to learn why semiconductor manufacturing is booming again in the U.S., and why chipmakers can’t keep up with AI demand. Discover what the shift from generative AI to agentic AI really means, and why smarter infrastructure and more reliable AI systems are becoming essential for the next era of computing.
Why semiconductor manufacturing matters again
The semiconductor industry has become a focal point for business, government, and technology leaders because chips are now foundational to nearly every major innovation, from AI infrastructure and autonomous systems to retail automation and data center growth. Joel explained that Intel has been navigating a major transition in its business model. Historically, Intel was known as an integrated device manufacturer, or IDM, meaning the company designed and manufactured chips for itself. The company is now shifting toward a foundry model, where it manufactures chips for other companies.
In the semiconductor world, that transition is complex. The scale of that investment is enormous. Joel noted that each modern fabrication facility can cost “upwards of ten billion dollars to get up and running.”
The stakes became even clearer during the COVID era, when supply chain disruptions exposed how dependent the global economy had become on concentrated manufacturing capacity. “What became clear was that there are choke points, and there could be severe shortages and limitations if too much manufacturing or supply chain capacity is centralized in one company or one geographic area,” Joel said. That realization has helped bring semiconductor manufacturing back into national and industrial strategy conversations. At the same time, the explosion of AI has created a demand curve that the industry is still trying to meet.
In the semiconductor world, that transition is complex. The scale of that investment is enormous. Joel noted that each modern fabrication facility can cost “upwards of ten billion dollars to get up and running.”
The stakes became even clearer during the COVID era, when supply chain disruptions exposed how dependent the global economy had become on concentrated manufacturing capacity. “What became clear was that there are choke points, and there could be severe shortages and limitations if too much manufacturing or supply chain capacity is centralized in one company or one geographic area,” Joel said. That realization has helped bring semiconductor manufacturing back into national and industrial strategy conversations. At the same time, the explosion of AI has created a demand curve that the industry is still trying to meet.
AI demand is reshaping the data center
The release of ChatGPT marked a turning point. As generative AI tools became widely adopted, they created massive demand for chips, networking equipment, memory, data center capacity, and energy.
“What happened at that time was OpenAI released ChatGPT, which unlocked a whole new paradigm of AI that we’re all starting to use now,” Joel said. “With generative AI, you can query these LLMs or models and get back answers. At first, it felt almost magical. Now it’s starting to become normal, but at the time, it was magical.”
That “magical” experience comes with major infrastructure requirements. Behind every prompt, response, agent, workflow, and AI-powered application is a growing need for compute. Joel said ChatGPT helped drive “explosive demand” for chips, networking equipment, and memory. “Memory has gone crazy,” he said. “There’s just this massive need for computing right now.”
The biggest challenge, according to Joel, is supply. “The reality is, we’re constrained,” he said. “We cannot make enough chips for all the demand that is out there right now.” For semiconductor companies, solving that problem requires a careful balance. Building new factories creates more capacity, but it also introduces risk. Historically, the semiconductor industry has moved in cycles. Demand rises, companies build capacity, and then demand can fall before those investments fully pay off.
This time, Joel said, the scale of demand is unlike anything the industry has seen before. “It is truly unprecedented,” he said.
“What happened at that time was OpenAI released ChatGPT, which unlocked a whole new paradigm of AI that we’re all starting to use now,” Joel said. “With generative AI, you can query these LLMs or models and get back answers. At first, it felt almost magical. Now it’s starting to become normal, but at the time, it was magical.”
That “magical” experience comes with major infrastructure requirements. Behind every prompt, response, agent, workflow, and AI-powered application is a growing need for compute. Joel said ChatGPT helped drive “explosive demand” for chips, networking equipment, and memory. “Memory has gone crazy,” he said. “There’s just this massive need for computing right now.”
The biggest challenge, according to Joel, is supply. “The reality is, we’re constrained,” he said. “We cannot make enough chips for all the demand that is out there right now.” For semiconductor companies, solving that problem requires a careful balance. Building new factories creates more capacity, but it also introduces risk. Historically, the semiconductor industry has moved in cycles. Demand rises, companies build capacity, and then demand can fall before those investments fully pay off.
This time, Joel said, the scale of demand is unlike anything the industry has seen before. “It is truly unprecedented,” he said.
The scalability question
One of the recurring themes in the conversation was the gap between what technology can demonstrate and what it can deliver at scale.
Joel applied this lens to augmented reality, virtual reality, humanoid robots, autonomous driving, 3D printing, and AI agents. Many of these technologies are already real in some form, but the harder question is when they will become cost-effective, widely available, and operationally reliable.
To explain that gap, Joel referenced Amara’s Law, the idea that people tend to overestimate the impact of new technologies in the short term and underestimate them in the long term. He believes the same pattern will apply to AI. The early excitement may run ahead of what companies can scale today, but the long-term impact could be even bigger than many people expect.
He pointed to autonomous driving as an example. Waymo is already operating in cities like Phoenix, but autonomous driving has taken far longer to scale than many early predictions suggested. “Autonomous driving has been around for 20 years, and it still hasn’t scaled out completely,” Joel said.
Joel’s point is that adoption rarely depends on the technology alone. Even when a breakthrough is technically possible, companies still have to determine whether it can be delivered at the right cost, supported by the right infrastructure, trusted to perform reliably, and tied to a clear customer ROI. Until those conditions are in place, even promising technologies can remain stuck in pilots, demos, or limited deployments.
Joel applied this lens to augmented reality, virtual reality, humanoid robots, autonomous driving, 3D printing, and AI agents. Many of these technologies are already real in some form, but the harder question is when they will become cost-effective, widely available, and operationally reliable.
To explain that gap, Joel referenced Amara’s Law, the idea that people tend to overestimate the impact of new technologies in the short term and underestimate them in the long term. He believes the same pattern will apply to AI. The early excitement may run ahead of what companies can scale today, but the long-term impact could be even bigger than many people expect.
He pointed to autonomous driving as an example. Waymo is already operating in cities like Phoenix, but autonomous driving has taken far longer to scale than many early predictions suggested. “Autonomous driving has been around for 20 years, and it still hasn’t scaled out completely,” Joel said.
Joel’s point is that adoption rarely depends on the technology alone. Even when a breakthrough is technically possible, companies still have to determine whether it can be delivered at the right cost, supported by the right infrastructure, trusted to perform reliably, and tied to a clear customer ROI. Until those conditions are in place, even promising technologies can remain stuck in pilots, demos, or limited deployments.
From edge computing to data center AI
Before moving into his current role in Intel’s Data Center & AI Computing Group, Joel worked in Intel’s Edge Group, focusing on point-of-sale and self-checkout use cases.
He explained that many people associate Intel processors with laptops, desktops, and servers, but Intel technology is also embedded in a wide range of devices across retail, gaming, healthcare, industrial systems, kiosks, and other environments. “People sometimes don’t realize this, but Intel processors go into a lot of things that are not a server and are not like a laptop or a desktop computer,” Joel said.
In retail, his focus was on point-of-sale and self-checkout technology, including use cases tied to loss prevention and customer experience. “Point of sale is arguably the heart of a retail operation and it’s vital and important that it’s always up and running,” Joel said.
The next wave of retail transformation, he explained, is not only about checkout. It is about creating more seamless store experiences where transactions can be automated and associates can focus on helping customers rather than scanning items. “Retailers would like to have a frictionless environment with a great customer experience where you still interact with a retail associate, but it’s not in a transactional way,” Joel said. That shift requires cameras, sensors, computing power, and AI-enabled systems, all of which tie back to the broader demand for edge and data center infrastructure.
In his new role, Joel works on Intel’s Xeon product line, the CPUs (central processing units) that sit at the heart of servers. As agentic AI evolves, the CPU is increasingly important as the orchestrator of complex tasks—responsible for organizing and running the operations these systems depend on.
He explained that many people associate Intel processors with laptops, desktops, and servers, but Intel technology is also embedded in a wide range of devices across retail, gaming, healthcare, industrial systems, kiosks, and other environments. “People sometimes don’t realize this, but Intel processors go into a lot of things that are not a server and are not like a laptop or a desktop computer,” Joel said.
In retail, his focus was on point-of-sale and self-checkout technology, including use cases tied to loss prevention and customer experience. “Point of sale is arguably the heart of a retail operation and it’s vital and important that it’s always up and running,” Joel said.
The next wave of retail transformation, he explained, is not only about checkout. It is about creating more seamless store experiences where transactions can be automated and associates can focus on helping customers rather than scanning items. “Retailers would like to have a frictionless environment with a great customer experience where you still interact with a retail associate, but it’s not in a transactional way,” Joel said. That shift requires cameras, sensors, computing power, and AI-enabled systems, all of which tie back to the broader demand for edge and data center infrastructure.
In his new role, Joel works on Intel’s Xeon product line, the CPUs (central processing units) that sit at the heart of servers. As agentic AI evolves, the CPU is increasingly important as the orchestrator of complex tasks—responsible for organizing and running the operations these systems depend on.
What agentic AI really means
Generative AI began with humans prompting machines through a chat box. Agentic AI moves the interaction forward. Instead of only human-to-computer interaction, computers can increasingly interact with other computers, systems, tools, and workflows.
“Now what’s happening and what you’re hitting on is the computer-to-computer interaction,” Joel said. “And that’s where generative AI then transforms into agentic.” Joel described agents as software programs defined by a person or company to complete a task or achieve a goal. That could include building a SaaS demo, answering customer calls, negotiating with a service provider, routing leads, managing workflows, or performing repetitive business processes.
He also noted a larger internet shift. “The internet traffic has now switched from being a majority of humans to being a majority of bots or computers,” Joel said.
That change introduces enormous opportunity, but also new costs and new risks. Joel raised the issue of “unmanaged intelligence,” where AI agents continue running in the background, consuming compute, energy, memory, and tokens. Today, many AI tools can feel inexpensive or free because costs are often subsidized. Over time, companies will need to understand the cost of completed workflows and the economics of running agents at scale.
He believes a major transition is coming from cloud-based AI to hybrid AI, where some agentic workflows run locally on laptops or edge devices using open-source models. “That capability hasn’t scaled yet, but it’s coming,” Joel said.
“Now what’s happening and what you’re hitting on is the computer-to-computer interaction,” Joel said. “And that’s where generative AI then transforms into agentic.” Joel described agents as software programs defined by a person or company to complete a task or achieve a goal. That could include building a SaaS demo, answering customer calls, negotiating with a service provider, routing leads, managing workflows, or performing repetitive business processes.
He also noted a larger internet shift. “The internet traffic has now switched from being a majority of humans to being a majority of bots or computers,” Joel said.
That change introduces enormous opportunity, but also new costs and new risks. Joel raised the issue of “unmanaged intelligence,” where AI agents continue running in the background, consuming compute, energy, memory, and tokens. Today, many AI tools can feel inexpensive or free because costs are often subsidized. Over time, companies will need to understand the cost of completed workflows and the economics of running agents at scale.
He believes a major transition is coming from cloud-based AI to hybrid AI, where some agentic workflows run locally on laptops or edge devices using open-source models. “That capability hasn’t scaled yet, but it’s coming,” Joel said.
The business opportunity for smaller companies
Agentic AI will not only reshape enterprise software. It could also change how small businesses operate. For example, Joel discussed how plumbers, electricians, HVAC professionals, and other service providers could use AI agents to answer calls, respond to customers, and capture opportunities around the clock. “Plumbers and electricians that own their small business can hire an agent to answer the phone 24/7 and respond to every call,” Joel said. “Something that they couldn’t necessarily do in the past.”
He does not necessarily expect small business owners to build these agents themselves. Instead, he expects consultants, SaaS companies, and AI providers to package these capabilities into more accessible tools.
The decision for a business owner may come down to simple economics. “What would it cost me to say hire a virtual assistant in a low-cost geography versus what’s the cost of hiring a computer agent,” Joel said.
He also sees long-term potential in physical applications, such as 3D printing parts for tradespeople who need hard-to-find components. The capability exists today in certain industries, including dental, but broader adoption will depend on cost and availability.
He does not necessarily expect small business owners to build these agents themselves. Instead, he expects consultants, SaaS companies, and AI providers to package these capabilities into more accessible tools.
The decision for a business owner may come down to simple economics. “What would it cost me to say hire a virtual assistant in a low-cost geography versus what’s the cost of hiring a computer agent,” Joel said.
He also sees long-term potential in physical applications, such as 3D printing parts for tradespeople who need hard-to-find components. The capability exists today in certain industries, including dental, but broader adoption will depend on cost and availability.
The reliability challenge
While the opportunity is significant, agentic AI still faces an important reliability problem. In business workflows, a small error can create a large downstream impact. A lead routing system may send incomplete information. An AI-generated report may analyze only part of a dataset. A customer service agent may misunderstand availability. In these cases, the output may sound confident but still be wrong.
Joel explained the difference between probabilistic AI systems and deterministic software. “Probabilistic being the AI that we use today,” Joel noted. “And most of the time, the answers come back fantastic and great, but there are some errors.”
Deterministic software, by contrast, follows a defined process the same way each time. “This is the process. This is how it runs. And every time it needs to run exactly this one way,” Joel said.
As AI becomes more integrated into business workflows, Joel believes companies will need to focus on quality, cost, and completed outcomes, not just speed or intelligence benchmarks. “You need to manage the intelligence, and we need to move into a place where we’re actually calculating the cost per completed workflow,” Joel suggested.
For vendors using AI to deliver leads or automated services, that may mean pricing models need to evolve. If a lead is incomplete or inaccurate, the vendor may need to build in dispute processes, refunds, or payment structures tied to completed work. “The natural economics of the situation are going to work out where those folks that are focusing on the quality and driving down the cost of delivering a completed high quality lead, those are going to be the winners,” Joel said.
Joel explained the difference between probabilistic AI systems and deterministic software. “Probabilistic being the AI that we use today,” Joel noted. “And most of the time, the answers come back fantastic and great, but there are some errors.”
Deterministic software, by contrast, follows a defined process the same way each time. “This is the process. This is how it runs. And every time it needs to run exactly this one way,” Joel said.
As AI becomes more integrated into business workflows, Joel believes companies will need to focus on quality, cost, and completed outcomes, not just speed or intelligence benchmarks. “You need to manage the intelligence, and we need to move into a place where we’re actually calculating the cost per completed workflow,” Joel suggested.
For vendors using AI to deliver leads or automated services, that may mean pricing models need to evolve. If a lead is incomplete or inaccurate, the vendor may need to build in dispute processes, refunds, or payment structures tied to completed work. “The natural economics of the situation are going to work out where those folks that are focusing on the quality and driving down the cost of delivering a completed high quality lead, those are going to be the winners,” Joel said.
The next big shift
Joel’s thesis is that agentic AI will become one of the largest shifts in technology since the rise of the internet and his advice is to start learning now. “If you’re not familiar with agentic AI, I would spend a little bit of time just trying to understand it because it’s definitely here to stay,” Joel said. “It’s going to keep growing, and it can be very helpful for you.”
He offered one final thought experiment. “Imagine if I went to you today and said, here’s 100 employees, what are you going to go do?” Joel said. “Think about what your answer to that question might be.”
That question captures the promise and challenge of the next AI era. As agents become easier to deploy, the limiting factor may not be access to automation. It may be knowing what problems are worth solving, how to manage AI responsibly, and how to build systems that create measurable value.
He offered one final thought experiment. “Imagine if I went to you today and said, here’s 100 employees, what are you going to go do?” Joel said. “Think about what your answer to that question might be.”
That question captures the promise and challenge of the next AI era. As agents become easier to deploy, the limiting factor may not be access to automation. It may be knowing what problems are worth solving, how to manage AI responsibly, and how to build systems that create measurable value.
Related resources
Want more like this resource? Find helpful links below.
