AI is no longer a technology that operates only inside massive data centers. It has begun working faster and more quietly inside smartphones, laptops, cars, home appliances, factory equipment, and medical devices. As the three demands of speed, cost, and privacy converge, on-device AI is becoming a core technology that opens the next stage of the artificial intelligence order.

***

When generative AI first appeared, people were amazed above all by the almost magical experience unfolding on the screen. Type in a question, and a long answer comes back. Give only a few lines of explanation, and an image is created. Summarizing a meeting, translating a text, or drafting a report suddenly became possible in an instant. Until then, artificial intelligence had felt like the language of research labs, large corporations, and technology experts, but generative AI broke down that boundary. An era opened in which anyone could speak to AI as if using a search box, and anyone could converse with AI as if using a messenger app.

But behind this convenient experience lies a heavy structure. When a user asks AI a question on a smartphone or laptop, the content is usually sent to a distant cloud server. The server performs the computation inside a massive data center and then sends the result back to the user’s screen. To the user, it looks as though sentences appear within a few seconds, but during that short time, electricity, cooling facilities, server equipment, and networks are all moving at once. Every AI response carries invisible costs and energy behind it.

The problem is that using AI is becoming less and less a special activity and more and more an everyday one. In the past, AI may have been used occasionally to ask about something curious or to draft a document, but in the future, AI will be embedded as a basic function across document programs, email, photo apps, meeting tools, customer service, search, translation, education, medicine, automobiles, and factory equipment. It will no longer be a tool used a few times a day, but an invisible assistant operating all day long. When that happens, sending every command and every piece of data to the cloud for processing becomes increasingly burdensome.

Companies face the same concern. At first, what matters is how much AI can raise productivity. But once it is deeply introduced into actual work, the question changes. If thousands of employees use AI every day, and if AI handles customer service, document writing, internal search, and data analysis, costs can grow like a snowball. The more AI is used, the more money it costs, the greater the dependence on data centers becomes, and the larger the concerns over network delays and security become as well. As convenience grows, structural fatigue grows with it.

On-device AI has emerged within this fatigue. It literally means processing part of the AI computation directly inside the user’s device. A smartphone recognizes the user’s voice, a laptop summarizes a meeting, a car detects a pedestrian, and factory equipment identifies defective products without necessarily relying on a distant server. AI has begun descending from the cloud into devices in people’s hands and into equipment in the field.

This change is not simply a matter of where processing takes place. It changes the economics of the AI industry, the standards for privacy protection, and even the competitive landscape of semiconductors, operating systems, smartphones, PCs, automobiles, and home appliances. Until now, the center of AI competition has been who can build a larger model and train it on more data. From now on, another important standard will be who can make AI run smaller, faster, and more safely. The next contest in artificial intelligence is not only a competition to build a giant brain, but also a competition over how close that brain can be placed to the user.

The easiest way to understand on-device AI is to think of a smartphone. Today’s smartphones already make many judgments on their own. When you take a photo, they distinguish faces from backgrounds. In dark places, they automatically correct brightness. The keyboard suggests the word you are likely to type next. Voice-to-text, reading text inside photos, and recognizing a face on the lock screen are also familiar functions. These are all examples of AI quietly operating inside the user’s device.

The recent change is that generative AI functions are beginning to enter this space as well. Earlier on-device AI was closer to relatively limited functions such as photo enhancement or voice recognition. Now, broader functions such as text summarization, real-time translation, image generation assistance, meeting note creation, and personalized recommendations are entering devices. Smartphones and laptops are no longer mere input devices. They are becoming small AI workstations that understand the user’s situation and perform some judgments directly.

The change in PCs is also symbolic. In the past, a PC was a tool for writing documents, storing files, and accessing the internet. A high-performance PC meant a machine that opened programs quickly, edited videos smoothly, and ran games well. But the arrival of AI PCs changes the meaning of the PC. Now the PC is moving beyond a tool for writing documents and becoming a tool that understands the context of those documents. It is moving toward summarizing voices during meetings, reducing background noise in video calls, interpreting information on the screen, and helping users with what they need to do next.

What matters in this process is an AI-dedicated chip. AI tasks can be processed with ordinary central processing units or graphics processing units, but there are limits in terms of power and efficiency. That is why recent smartphones and laptops include neural processing units specialized for AI computation. These chips are designed to process the calculations frequently performed by AI faster and with less power. Simply put, they create a separate AI-dedicated workbench inside the device.

In smartphones, this change is even more sensitive. A smartphone is the user’s most private device. Photos, contacts, location data, messages, schedules, payment information, and health data are all stored there. Having AI inside this device means powerful convenience, but it also immediately touches privacy issues. If AI that analyzes the user’s photos, understands conversations, and grasps schedules and habits has to send all information to an external server, people cannot help feeling uneasy. That is why the trend of processing as many AI tasks as possible inside the smartphone is growing.

The same is true of cars. When a car detects danger on the road, there is no time to hesitate. When the vehicle ahead suddenly brakes, when a pedestrian suddenly appears, or when the lane shifts, the car cannot ask a distant server and then make a decision. Sensors and chips inside the vehicle must immediately interpret the surrounding situation. In this field, on-device AI is not a convenience feature but a safety feature. How close AI is and how quickly it can judge can be connected to human life.

Home appliances and wearable devices are also stages for on-device AI. Robot vacuum cleaners understand the structure of the home and the rhythm of daily life. Air conditioners operate by reflecting indoor temperature and user habits. Smartwatches analyze heart rate and sleep patterns. All of this information is closely tied to personal life. If data can be processed inside the device without sending everything outside, users can gain an experience that is both more convenient and less unsettling.

In the end, on-device AI is not a technology that merely makes AI smaller. It is a technology that brings AI closer. It places AI where data is generated, where judgment is needed, and where the user actually acts. If past AI was a massive central brain, on-device AI is closer to distributing parts of that brain across countless devices and real-world sites.

When people talk about AI performance, they often think of accuracy. Is the answer correct? Are the sentences natural? Does it generate images well? Does it write code properly? These are certainly important. But in actual user experience, there is another standard: speed. No matter how smart AI is, if it responds slowly, people quickly feel frustrated.

A smartphone voice command makes this easy to understand. When a user says, “Find only the document in the photo I just took,” the experience feels natural if the device responds immediately. But if the device has to wait for a server connection every time and pauses for several seconds depending on the network condition, the user will stop using that function often. For AI to become an everyday tool, users must not feel that they are waiting for AI. Like swiping a screen with a finger, it should respond as soon as the user speaks.

The first advantage of on-device AI is that it reduces this latency. Cloud AI has to send the user’s request to a server, process it there, and receive the result back. If the network is fast, this process becomes shorter, but it does not disappear completely. By contrast, AI processed inside the device has a short travel distance. The user’s data is analyzed directly inside the device, and the result appears immediately on the screen. It may seem like a tiny difference, but as repeated use increases, this difference greatly changes the quality of the experience.

In factories, speed is money. Imagine a production line where a camera photographs the surface of each product as it passes, and AI determines whether it is defective. Sending every image to a cloud server for analysis and receiving the result back may seem possible, but in the field, latency and network instability become serious problems. If the judgment is delayed, the defective product has already moved on to the next process. If the timing for stopping the line is late, the loss grows. AI that judges immediately inside the equipment reduces this risk. In manufacturing, fast judgment is quality, and quality is cost reduction.

Speed is also important in medicine and care. If a wearable device detects signals such as heart rate abnormalities, falls, or breathing changes, and then sends all data to a server before making a judgment, it may be too late in an emergency. A structure in which the device itself first identifies danger signals and notifies a guardian or medical institution only when necessary is more suitable. As populations age, these real-time detection and warning functions will become more important. On-device AI can become a technology that protects health and safety even in everyday spaces outside hospitals.

Cars show the meaning of speed most dramatically. On the road, one second is a long time. If a vehicle delays judgment when the car ahead suddenly stops or a pedestrian runs into the road, it can lead to an accident. What is needed then is not a lengthy analysis from a massive server, but an immediate judgment made inside the vehicle. The car’s cameras, radar, and sensors must read the surroundings, the AI inside the vehicle must interpret the danger, and braking or warning must be executed immediately.

When speed increases, expectations for AI also change. People will no longer be satisfied with an experience in which they “ask AI and wait.” When a document is opened, a summary is already prepared. When a meeting ends, the minutes are already organized. When a photo is taken, the necessary information is automatically classified. Cars naturally detect danger in advance. On-device AI is the foundation that changes AI from a special feature into a basic experience.

AI is drawing attention as a technology that raises productivity, but it is also a technology that creates costs. When companies first introduce generative AI, expectations are high. They see the advantages of reducing repetitive work, helping employees write reports and emails faster, automating customer consultations, and shortening the time needed to create marketing content. But once usage becomes serious, the calculation changes. The more AI is used, the more computation it requires, and computation eventually returns as cost.

At first, only a few departments use AI experimentally. At this stage, costs may not look large. But when AI spreads across the entire organization, the situation changes. If every employee summarizes documents several times a day, creates meeting notes, analyzes customer data, and requests draft reports, usage explodes. If AI is attached to customer service, internal search, or sales support systems, there are more users and requests become more frequent. As AI becomes a basic infrastructure for work, cloud costs become an item that corporate management cannot ignore.

On-device AI is a way to redesign this cost structure. Instead of assigning every task to the cloud, light and repetitive tasks are processed on the device. Simple sentence correction, first-stage recognition of meeting audio, photo classification, basic translation, keyword recommendation, and organizing frequently used document formats may not necessarily require a massive server. Processing these tasks inside the device can reduce server usage and data transmission.

Of course, not every task can be handled on the device. Writing a long report based on a vast amount of material, analyzing complex legal documents, or developing a strategy by integrating data from multiple systems still requires the powerful computing capabilities of the cloud. What matters is role division. Light tasks are handled nearby, and heavy tasks are sent to large servers. In this way, users gain a faster experience, and companies can manage costs more efficiently.

This change also affects corporate IT strategy. In the past, cloud contracts and server infrastructure were central to AI adoption. In the future, the laptops, smartphones, and work devices used by employees themselves become part of the AI strategy. What devices should be provided, what level of AI computing capability those devices should have, which tasks should be processed inside devices, and which tasks should be sent to the cloud all become important decisions.

Public institutions, financial companies, hospitals, law firms, and manufacturing companies are especially sensitive to this issue. They handle sensitive data. Customer financial records, patient medical information, contracts, research and development materials, and factory production data cannot easily be sent to external servers. For organizations that want to use AI but are burdened by data movement, combining on-device AI with internal systems is a more realistic approach. The more an organization must consider productivity and security at the same time, the greater the significance of on-device AI becomes.

The cost of the AI era is not only a matter of money. It also includes electricity and carbon, networks and security, server expansion and operating personnel. A method that pushes every AI task into central data centers is bound to face limits at some point. On-device AI is a technology that distributes that burden. To use AI more widely, AI must instead run closer, smaller, and more efficiently.

Another force pushing the spread of on-device AI is privacy. For AI to become smarter, it must know the user better. The more it understands the user’s speech patterns and habits, photos and schedules, location and preferences, work documents and conversation context, the more personalized help it can provide. But for that very reason, AI can also become an unsettling technology. If people have to give up too much information in exchange for convenience, they will not easily trust it.

Think of a smartphone. Inside a smartphone, almost an entire life is stored. Family photos, conversations with friends, banking apps, hospital appointments, travel routes, search records, memos, and work files are all contained there. If AI uses this information well, it can provide astonishing convenience. But if this data keeps being uploaded to external servers and users cannot know where and how it is processed, anxiety grows. Users want smarter AI, but they do not want their entire lives exposed to servers.

On-device AI reduces this contradiction. It keeps information that can be processed inside the device inside the device. For example, if a smartphone finds documents in photos, a laptop summarizes meeting content, or a keyboard suggests sentences without sending all original data outside, the user’s burden decreases. A structure becomes possible in which the device first makes judgments internally and uses external computation only in limited cases when truly necessary.

Privacy is now not an additional feature, but a product advantage. In the past, people chose smartphones mainly based on camera quality, screen size, battery, and storage capacity. In the future, “where my data is processed” will also become an important criterion. Whether a photo is uploaded to a server when it is analyzed, whether audio is stored when meeting content is summarized, and whether an AI assistant connects my schedule information to an external company’s system can determine consumer trust.

For companies, this issue is even more direct. A structure in which company meeting notes, contracts, customer information, financial data, and technical documents flow into external AI servers is a major burden. No matter how convenient AI is, it is difficult to use in actual work if it cannot pass security policies. That is why companies consider not only whether to use AI, but also which data must never be sent outside. On-device AI provides a safer option for such organizations.

In education, medicine, finance, defense, and public administration, privacy demands are even stronger. Student learning records, patient health information, personal credit information, and materials related to national security are highly sensitive. For AI to spread in these fields, it is not enough to claim that performance is excellent. Trust is needed regarding where data is processed, who can access it, and whether records remain or are deleted. On-device AI is one way to design that trust.

Of course, processing data inside a device does not make every problem disappear. The device itself can be hacked, and if users do not know what data is being processed by AI, another risk arises. On-device AI also needs transparent settings, clear permission management, security updates, and responsible design. But in that it reduces data movement and expands the option of not sending sensitive information to external servers, on-device AI is a very important technological direction in the age of privacy.

For on-device AI to become possible, AI models themselves must also change. Ultra-large models operating in massive data centers cannot simply be placed unchanged inside smartphones or laptops. Devices are smaller than servers, run on batteries, and face constraints in heat, storage, and space. Therefore, the core of on-device AI is not merely moving AI into devices, but reducing AI, making it lighter, and redesigning it efficiently for devices.

This is why model lightweighting has become important in the AI field. Leaving aside difficult technical terms, the essence is to maintain performance as much as possible while reducing size and computation. It is similar to making a small notebook that neatly organizes frequently used knowledge and functions instead of carrying around an entire encyclopedia. It may not answer every question perfectly, but it creates AI that is fast and useful enough for everyday and repetitive tasks.

In fact, most everyday tasks do not always require the largest AI. Slightly polishing a sentence, briefly summarizing a meeting, distinguishing objects in a photo, performing basic translation, and predicting the user’s repeated actions may be better suited to small and fast models. Large models are powerful, but they can be expensive and slow. Small models do not know everything, but they respond quickly from nearby. In the era of on-device AI, the value of these small models grows.

This change is also linked to semiconductor competition. If AI operates only inside data centers, high-performance server semiconductors become central. But if AI enters smartphones, PCs, cars, home appliances, and industrial equipment, the stage of competition becomes much broader. Each device has different conditions. Smartphones require battery efficiency and heat control. Cars require stability and real-time response. Factory equipment requires durability and accuracy. The AI chips and software optimization needed for each device differ.

For device manufacturers, on-device AI is also a new means of differentiation. The smartphone market is already mature. Screens are sufficiently sharp, cameras are already excellent, and designs do not differ greatly. The same is true of laptops and home appliances. It is now difficult to differentiate through hardware alone. In the future, what AI functions are provided, how naturally those functions operate, and how safely user data is handled will shape the impression of a product.

From the user’s point of view, good on-device AI does not show off. It does not create many buttons or loudly emphasize feature names. It works naturally at the moment it is needed. When a meeting ends, the summary is ready. When the user tries to find a photo, the desired scene appears immediately. When the user gets stuck while writing, the sentence continues smoothly. The user feels not that “I ran AI,” but that “the device helps me better.” This naturalness is the real competitiveness of on-device AI.

The age of small models is not a trend opposite to the age of large models. Rather, the two move together. Large models handle complex knowledge and high-level reasoning, while small models take charge of everyday and immediate tasks. It is a structure in which a giant brain and small brains divide their roles. On-device AI is expanding the market for these small brains.

Even as on-device AI becomes important, cloud AI will not disappear. Rather, the key going forward is balance between the two methods. AI processed inside the device can be fast, private, and low-cost. But it is not powerful enough to solve every problem. Cloud AI, by contrast, can use much larger models and computing power, but it carries the burdens of latency, cost, and data movement. Instead of one completely replacing the other, a structure that uses their different strengths together is more realistic.

When a user requests simple sentence correction, photo classification, or short translation, processing it immediately inside the device is appropriate. It is fast, sufficiently accurate, and data does not leave the device. But tasks such as analyzing dozens of documents to create a strategic report, solving complex scientific problems, or generating high-resolution video are still better handled by the cloud. Large brains handle tasks that require large calculations, while small brains handle immediate everyday tasks.

In the future, users are likely to be almost unaware of this division of labor. The device first judges the user’s request. It considers whether the task can be processed inside the device, whether the cloud is needed, whether sensitive data is included, what the network condition is, and whether the battery is sufficient, then selects the most suitable method. The user simply asks a question and receives the result. Behind the scenes, on-device AI and cloud AI divide their roles.

This structure resembles a hybrid car. The driver does not deeply think at every moment about whether the electric motor is moving or the engine has intervened. The car selects the efficient method according to the situation. AI can become similar. Some tasks are quietly processed inside the device, some move to the cloud’s large model, and some data is protected without going outside. A good AI service does not reveal this transition to the user in a complicated way.

In this process, the role of operating systems and platforms grows larger. A layer is needed to decide where to process the user’s request, which apps can access which data, and how AI-processed results should be stored and deleted. Smartphone operating systems, PC operating systems, automotive software, and enterprise security platforms become new gateways in the AI era. As much as the AI functions themselves, the ability to manage AI so that it works safely inside devices becomes important.

Hybrid structures are also important in corporate work environments. Sensitive materials can be processed on company devices or internal servers, while public materials or complex analysis can use cloud models. Employees feel as though they are using a single AI tool, but behind the scenes, the processing location changes depending on data sensitivity and task difficulty. This is how organizations can balance AI use and security.

Therefore, the future of on-device AI is not so much “AI leaving the cloud” as “AI no longer relying only on the cloud.” AI becomes more distributed, more context-aware, and more personalized in structure. An era opens in which central giant brains and small brains at the edge move together.

When on-device AI spreads, the first things to change will be the devices we touch every day. A smartphone becomes not merely a tool for running apps, but an assistant that understands personal context. It understands photos the user often searches for, sentences the user repeatedly writes, places the user often visits, and patterns in the user’s schedule, then offers help at the right moment. What matters is that these functions do not reveal themselves grandly. Even without separate commands from the user, the device prepares more and more tasks in advance.

PCs also change. When working on a document, the PC finds related materials. When reading a long text, it summarizes the key points. During a meeting, it organizes remarks. When creating a presentation, it suggests flow and wording. In the past, using AI required opening a separate website or app. In the future, AI naturally enters document programs, email, video meeting tools, and operating systems. AI no longer works like a guest that must be specially summoned, but like a colleague always by one’s side.

Home appliances also change quietly. An air conditioner is no longer just a device that adjusts temperature. It operates while considering the time of day, electricity rates, indoor humidity, and the user’s lifestyle patterns. Washing machines better understand fabric and the degree of contamination. Refrigerators understand the condition of ingredients and consumption habits. Robot vacuum cleaners can consider not only the structure of the home, but also the times when family members are usually active. If a significant portion of these functions is processed inside the device, users gain a more convenient and safer living experience.

On-device AI can also play an important role in cities and public infrastructure. If traffic signals, CCTV systems, energy management equipment, and disaster detection sensors judge directly in the field, city operations can become faster and more efficient. Sending all video to a central server for analysis creates high costs and major privacy controversy. By contrast, a structure in which field equipment selects only dangerous situations and sends alerts can reduce unnecessary data movement while increasing response speed.

There is also potential in education. If a student’s learning device can understand individual comprehension levels and repeated mistakes, and provide basic feedback even in environments with unstable internet connections, the learning experience can become more personalized. Of course, protection of learning data and algorithmic bias must be handled together. Still, on-device AI shows the possibility that educational AI will not depend only on massive platforms, but will work more closely and continuously inside the student’s device.

In medicine and care, on-device AI creates quiet but powerful change. Sensors that detect the movement of older adults, wearables that analyze heart rate and sleep, and devices that monitor the lifestyle patterns of people with chronic diseases can first detect danger signals in the field. During the time when a person is not at the hospital and is not being seen by a doctor, devices monitor the user’s condition. This does not mean devices replace every judgment made by medical professionals. It means their auxiliary role in detecting danger sooner and connecting necessary help becomes larger.

This change may not be felt as a huge revolution all at once. The world does not suddenly transform one day. Instead, the devices people use every day become a little faster and a little smarter. Photos become easier to find, writing becomes easier, meetings become lighter, housework becomes a little more automated, and health abnormalities are detected a little sooner. The real power of on-device AI lies less in flashy demonstrations and more in this quiet accumulation. When technology becomes so natural that it is no longer visible, changes in daily life become deepest.

On-device AI also sends an important signal to Korean industry. Korea has strengths in smartphones, memory semiconductors, displays, home appliances, automobiles, batteries, and telecommunications infrastructure. If AI remains only inside data centers, Korea’s strengths can be used only in limited ways. But if AI descends into devices and the field, the story changes. The manufacturing and hardware domains where Korea has excelled can once again move to the center of AI competition.

Smartphones and home appliances are the most direct stage. Korean companies already have strong product competitiveness in the global market. But future competition will not be decided only by camera pixels, screen size, or design. What AI experience is provided in the user’s daily life, how natural and trustworthy that function is, and how smoothly multiple devices are connected become important. Simply advertising that “AI functions are included” is not enough. Users must actually feel that their lives have become more convenient.

Semiconductors are also central. As on-device AI spreads, the importance of chips and memory that process AI inside devices grows. To run AI quickly inside a small device, not only computing performance but also the ability to efficiently load, store, and process data is important. For Korea, a powerhouse in memory semiconductors, this is clearly an opportunity. However, simply making large quantities and supplying them cheaply is not enough. High-value semiconductors, packaging, and software optimization suited to AI devices must move together.

The automobile industry is also deeply connected to on-device AI. Electric vehicles, autonomous driving, in-vehicle voice assistants, driver condition detection, and cabin personalization are all related to AI operating inside the vehicle. A car is no longer merely a moving machine. It is becoming closer to a mobile computer that combines software, sensors, and AI. Whether Korea’s automobile industry can move beyond hardware manufacturing and successfully combine AI experience with vehicle software may determine its future competitiveness.

There are also opportunities for small and medium-sized enterprises and manufacturing sites. Not every company can build a massive cloud AI system. But putting defect-detection AI into factory equipment, adding demand forecasting functions to logistics terminals, placing customer-service AI into store kiosks, and installing simple risk-detection functions in medical assistance devices are more realistic directions. On-device AI can open new markets not only for large companies, but also for small and medium-sized companies that create field-oriented solutions.

But if Korea is to seize this opportunity, hardware alone is not enough. On-device AI is a field in which chips, devices, operating systems, models, security, and app ecosystems move together. As much as the ability to make good components, the software capability to connect them into an experience that users can feel is important. If the device is excellent but the AI experience is awkward, competitiveness remains incomplete. Conversely, if AI functions are natural and trustworthy, the value of the entire product rises.

The task for Korean industry is clear. It must move from being a country that makes devices well to a country that designs device experiences in which AI operates. It must combine hardware completeness, software sensibility, trust in data protection, and service ecosystems. On-device AI is a signal to Korea that it must move beyond familiar manufacturing competition into a new competition over experience.

The expansion of on-device AI changes the standards of the AI industry. Until now, much of the discussion has focused on the size and performance of models. Larger models, more data, higher accuracy, and better generative capability have received attention. This competition will continue. But as AI enters everyday life and industrial fields more deeply, other questions become important. How fast is it? How inexpensive is it? How safe is it? How private is it? Does it work even when the network is unstable? Can users trust it?

On-device AI is one answer to these questions. Processing inside the device reduces latency. Reducing cloud usage can also lower cost burdens. Not sending sensitive data outside can increase privacy trust. Basic functions can be maintained even in unstable network environments. As these advantages come together, on-device AI is becoming not simply a technical option, but a necessary condition for the spread of AI.

Of course, on-device AI is not a universal key that solves every problem. Small models can sometimes be inaccurate, and differences in device performance can widen the gap in user experience. Security issues also remain. The fact that AI operates inside a device does not eliminate responsibility. Standards are needed regarding what data is processed, whether users can control it, and who is responsible when errors occur. As technology becomes closer, responsibility must also become closer.

Even so, the direction is clear. AI will not remain only a more centralized technology. It will evolve into a structure in which massive models in data centers and small models in the hand move together. The user’s device becomes not merely an input device, but an intelligent terminal responsible for part of the judgment. Corporate sites will have equipment and systems that judge for themselves rather than relying only on the cloud. Cities can move not through a structure in which a central server controls everything, but through a structure in which devices in the field each judge and connect.

The core of this change is not the size of the technology, but its distance. More important than how huge AI is will be how close it operates at the moment it is needed. AI that users do not have to wait for, that does not unnecessarily send data outside, and that does not make people hesitate because of cost burdens will spread more widely. On-device AI is a direction that makes artificial intelligence faster, more private, and more everyday.

In the future, people may no longer say they “use” AI. That is because the act of opening a smartphone, turning on a laptop, getting into a car, and using home appliances will itself already be an act of being with AI. If the age of the cloud was the time when AI showed itself to the world, the age of on-device AI is the time when AI seeps into everyday life. Artificial intelligence is no longer only a grand technology shining above the clouds. It is becoming an infrastructure of daily life that quietly moves in the hand.