Embracing AI PCs means embracing the “Swift Moment” of efficiency innovation.
Cover Source｜Visual China
The summer of generative AI has arrived too early, and before entertainment can take its bow, various industries are already ready for a productivity revolution.
Large models with billions of parameters are packaged into processing units, placed on both sides of the “future productivity” landscape. The focus battle among tech giants constantly stimulates our imagination of how quickly new things can land: AI PCs are not a consensus strategy created. Localized AI application scenarios not only exist in large numbers but may also be put to work and production sooner.
In this era of “hardware discourse power,” the alliances of AI PCs no longer belong solely to PC brands and operating systems. As a processing manufacturer that has recently gained momentum and has fully entered the era of artificial intelligence hardware, Intel is also not to be outdone.
Canalys predicts that by 2027, 60% of PCs shipped will become true AI PCs. The upcoming “replacement wave” is merely the pace of history, while hardware manufacturers are more concerned with the seamless integration of performance and application under the technological wave.
AI PC: An Excellent Carrier of Technological Change
The prosperity of generative AI is undoubtedly a watershed moment for modern productivity technology. Commercial computers, as core tools for enterprises and end-users to handle business scenarios and professional-level demands, will also be completely iterated.
Throughout history, the PC’s irreplaceable ability to digest and carry transformative emerging technologies is not only reflected in “larger devices can do more things.” More importantly, the PC environment, due to its natural fit with native application development, is always the first to land technology scenarios. Scenarios break through layer by layer from top to bottom, from development to application, from tool attributes to consumer entertainment. Whether it is hardware innovation or operating systems, they are all moving in this direction, hence the upgrade of enterprise productivity always happens first on the PC side.
Undoubtedly, whether from the technical or market side, the current “fervor” for AI stems from future application value. In the business field, as long as the value of applications remains to improve productivity, the AI PC transformation across various industries will be an inevitable path.
At this year’s AI PC product launch, Intel announced collaborations with over 10 OEMs and more than 35 local ISV partners to utilize its fully AI-enabled Core Ultra processors in the commercial AI PC field.
Image Source: Visual China
If the success of software ecosystems like Office and Adobe decades ago resulted from the ever-increasing performance of commercial PCs pushing many industries to define “digital standardization,” then what Intel is doing today with Core Ultra is nothing less than using more detailed and controllable PC performance to force industries and enterprises to define “productivity scenarios.”
After all, no matter how good the large model is, some enterprise end-users still find it “tasteless” in industry applications. This perception mainly stems from the large model technology’s “unrealistic greed” regarding hardware load, cost, and speed.
What users naturally hope for is results that are faster, more accurate, and more diversified than what humans can create. Under such expectations, AI PCs should not be a “winged horse”; they should be a low-flying car that can fly and run: combining the essence of industrial manufacturing with actual demand scenarios to form a lower-threshold innovative application benefit.
While developers are still figuring out whether NPU or GPU is the first choice for large model training, AI PCs have already validated the processor’s AI functionality adaptability for users’ mainstream application scenarios. In Intel’s workload design for AI PCs, NPU, CPU, and GPU each play their roles in different application scenarios and different tasks within the same scenario: NPU provides faster computing power than the cloud for low-intensity application instructions such as image and audio processing, freeing up CPU and GPU for heavier large model tasks.
Business restructuring often starts from the last mile of the terminal. Intel’s current dominance in the processor market is due to its ability to position itself early in each generation of PC revolutions through architectural innovation. Actively promoting new technological markets based on the results of infrastructure allows one to stand at the forefront of technological applications.
Commercial AI PCs face harsher tests than consumer-grade AI PCs. After all, commercial groups are more rational, and their pursuit of stability far exceeds the desire for novelty. Enterprises themselves do not care about parameters, nor do they target computing power and low-power inference engines like large model manufacturers, nor will they pay for unnecessary features; the experience of business efficiency outweighs everything else. Therefore, whether it is Intel or other hardware manufacturers, the processor is the foundation of a towering tree, while ISVs, developers, and solutions are the new sprouts in the front.
Doing well in the application ecosystem reflects in two aspects: one is to explore the deep waters of business thoroughly, and the other is to simplify a large number of low-threshold, highly interactive lightweight daily application scenarios. However, without the redesign of processor architecture and the deep integration of releasing computing power, many scenarios positioned as “optimizations” will cease to exist. Therefore, another trend in the AI PC era is that core hardware manufacturers like Intel are also strengthening their investment in software ecosystems comprehensively, transforming products like Core Ultra processors and vPro platforms into targeted commercial solutions.
Battle of the Hundreds of Models: The Three Laws of AI Applications
Starting from user operation habits over the past two years, the localized approach represented by AI PCs may seem “counterintuitive.” After all, the mainstream migration of digital production from local to cloud has only taken a few years. Under the long-term education of cloud computing, enterprises and users may have concerns about localization.
So why has the return of large AI models from the cloud to local become a prevailing trend?
At the CES 2024 conference, Intel CEO Pat Gelsinger summarized three major laws regarding the future trends of AI.
First, according to economic laws, considering the high costs of cloud services, future AI data should be processed locally to reduce AI service costs.
Secondly, there are physical laws. The response speed may be affected when transmitting AI data between the cloud and local, so improving the efficiency of physical layer data transmission is crucial for the timeliness and accuracy of AI services. Only through multi-faceted optimization and integration of hardware, operating systems, and application technologies can the bottleneck issues behind AI operational efficiency be resolved. A simple example is that even without a network, a large model with 20 billion parameters can run entirely on Intel’s AI PC.
Third, the data confidentiality law further highlights the significance of localizing AI services. Many large model manufacturers’ own data sources are “unspeakable”; in an era where AI and big data run parallel, the security and privacy of local data far exceed those of cloud services.
Moreover, Intel’s new architecture has made us realize that the improvement of processor performance and functionality can directly reflect better local performance capabilities of large models. Under the localization of AI PCs, enterprises can develop their businesses in a relatively autonomous and controllable manner while keeping pace with the continuous evolution of technology. Conversely, if the workflows of enterprises and end-users rely heavily on cloud services, their intuitive experience of hardware changes will also be limited by the pricing and business strategies of the services themselves.
After a long golden age of SaaS, the quantifiable structure of productivity for enterprises is also being reshaped. One of the controversial directions is how AI PCs in 2024 will avoid the Occam’s razor problem.
Deconstructing and reconstructing may seem like a subtraction about cost, but it actually needs to awaken the market again. For example, in the current generation of commercial computers that we have begun to refer to as “traditional PCs,” voice digital assistants like Siri and Cortana are widely integrated, yet the commercial value they can generate is meager. This has made many wary of the “AI assistants” on AI PCs.
In fact, one major advantage of AI PCs is that with the enhancement of core hardware performance, coupled with pre-installed optimized large models, the capabilities of AI will avoid being “useless due to generality” and instead become more “tailored.” It can serve as a simplified tool for complex tasks, a foundational infrastructure and imagination behind concepts, or a native technology stack that drives reasoning and innovation across the entire industry.
With Intel’s Meteor Lake achieving the most significant transformation in CPU architecture in nearly forty years, the application directions of the “battle of hundreds of models” are also being refined continuously, and the application threshold of AI in PCs has been significantly lowered. The first batch of application scenarios and industry applications in commercial AI PCs have already landed steadily, suggesting that embracing a fully AI-enabled production method will be the primary theme of digital transformation for most industries and enterprises within the next five years.
In this process, Intel’s promotion of AI PC application scenarios is crucial. Because no single product or industry application developed by any OEM or ISV can “define” AI PCs and integrate AI productivity, the persuasive power of AI PCs fundamentally comes from hardware manufacturers and operating systems. In other words, most users’ understanding of PCs has already entered deep waters; parameters can no longer “fool” people. Before the key architecture such as processors is disrupted and costs become affordable, the “critical point of qualitative change” for local AI to achieve industry-level applications is almost impossible to reach.
Future of AI PCs: Embracing Ecosystem Partners and Exploring Key Scenarios
At the AI PC launch, Intel identified six key application scenarios. These include chatbots that are vertically optimized for specific industries and fields. Besides the scenarios themselves, Intel also emphasized the importance of building an AI PC ecosystem. Localization serves as a security lock for digital assets, and as a new phenomenon, growing together with developers and technology partners who have been deeply engaged in the enterprise service field for many years is the “key” to unlocking sustainable value for AI PCs.
Image Source: Visual China
In the currently most mainstream chatbot scenario based on large models, the AI PC powered by Core Ultra introduces more professional AI Q&A capabilities for specific industries and fields. Supported by Intel hardware, the AIGC assistant of Driver Life, which can pre-load localized large models like Baichuan 2-7B, is a good ecological example. Driver Life is both a product of the traditional PC era and a rapidly iterating transformation in the current AI scene through integrating large model capabilities, while maintaining the professionalism of local hardware perception and the intelligent experience of the AI era.
Document processing and other office scenarios are another major battlefield for AI PCs. The AI Office Assistant can embed third-party plugins into Office and other software to generate or optimize office experiences from a single point. Among them, the Kutools series tools from Extended Office, a vendor of Office extension applications, have also undergone in-depth optimization and AI capability upgrades for Intel’s AI Office Assistant environment, including data processing, AI document writing, etc. In a data environment that is seamlessly integrated, it can not only understand commands but also learn the office habits of end-users, thus providing customized support.
It is worth mentioning that Kutools is currently only achieving AI iteration under Intel’s AI PC ecosystem and has not yet collaborated with other platforms.
For knowledge-intensive fields such as finance and education, the AI PC’s local knowledge base is a “contributor” that enhances the professionalism of computers as productivity hardware. Under Intel’s ecosystem, there are already numerous RAG (Retrieval-Augmented Generation) applications, with ecosystem partners including Star Ring and Digital China. With the hardware support of AI PCs, RAG can now understand various professional document formats, achieve multi-turn dialogue and video summarization, helping individuals and enterprises share information content, while also allowing AI PCs to provide customized solutions for different fields from the perspective of knowledge accumulation.
In addition, in the continuously evolving AI video image processing, text-to-image, text-to-video, and text-to-3D model fields that require higher hardware performance, Intel continues to embrace technological development. Undoubtedly, the integration of low-power management, independent graphics architecture, and NPU has made Core Ultra perform exceptionally well in office scenarios. Compared to PCs from three years ago, performance has improved by as much as 47%.
The scenario changes brought about by high-performance processors are the horizontal coordinate, while the AI transformation across various industries is the infinite possibility on the vertical coordinate. Especially in the rich commercial application scenarios of the Chinese market, every commercial AI PC may represent a specific production environment. Grasping the industry ecosystem means grasping the maximum imagination of AI PCs.
This also indicates that in the deep waters of commercial computers, where industrial production innovation is involved, AI PCs must continuously deliver “answers” that match the value of their hardware performance. The AI pharmaceutical practice shared by InSilico at Intel’s press conference is a typical case.
It is well known that AI’s ability to reduce costs and increase efficiency in new drug development has received high attention from the global industry. InSilico, as a leading domestic AI pharmaceutical company, has been using a complete drug development platform covering everything from target discovery and small molecule generation to clinical outcome prediction. After the commercialization of AI PCs, it quickly adapted its most important platform, PandaOmics, to Intel’s AI PC. This application aims to provide more convenient target discovery and pharmacological and pathological mechanism research methods for various professionals including hospitals, university teachers, and pharmaceutical researchers. Since the use of such platforms involves processing sensitive genomic data, running them locally to protect user data is indispensable. This is precisely the value point where Intel AI PCs excel as a technological foundation.
Returning to the three laws of AI PCs, the ecological cycle of any emerging technology will not deviate from the track: based on a breakthrough and explosion of a single technology, providing fertile ground through core hardware foundation; then fusing key performance into computers, commercial PCs, and other production devices and environments to break through scenario bottlenecks; subsequently incorporating more economic models in the process of continuous evolution, gradually making technological innovation the mainstream production method. For enterprises themselves, AI PCs are currently both inclusive and secure AI terminals; embracing AI PCs means embracing efficiency innovation’s “Swift Moment.”
The strong upward trajectory of AI will eventually conquer the commercial computer field, which is slightly lengthy in the replacement and innovation cycle. Localized AI services will evidently bring the next growth curve to the PC field. Although a thousand people may have a thousand interpretations of “what is a true AI PC,” on this issue, Intel, which has already positioned its processor technology architecture and is continuously exploring ecosystems and application scenarios, believes that: the ability to apply AI capabilities locally and generate application value is what defines an AI PC.
