Over the past weekend, I immersed myself in the session videos from Build 2026, focusing particularly on those that pertain to Windows 11. As I navigated through the content, several themes emerged, but one in particular captured my attention.
The sessions I explored can be broadly categorized into three distinct areas:
- Windows app development. Although this was the smallest segment, it marked a significant resurgence in Microsoft’s focus on Windows app development, a topic that has seen sporadic attention in recent years.
- Developer productivity. Microsoft has long championed Windows as the premier platform for developers, regardless of whether their projects are web-based, cloud-oriented, or cross-platform. Build 2026 showcased a remarkable array of advancements aimed at enhancing developer productivity.
- Agentic AI. This was undoubtedly the most prominent theme at Build this year, particularly within the context of Windows. The convergence of various factors, including rising costs of cloud-based AI, the continuous improvement of local AI, and the necessity to harmonize AI across both realms, made 2026 a pivotal year for agentic AI.
In my view, the essence of Build 2026 revolved around this last category. While the introduction of agents to Windows 11 was anticipated, the revelation of hybrid agents—those equipped with local AI sub-agents—was unexpected. Microsoft’s broadened interpretation of hybrid AI exceeded our initial expectations.
This development is intriguing for numerous reasons, particularly as it addresses the “pain points” that users have experienced this year. However, the groundwork for this transformation was laid years ago, predating the emergence of Copilot and when Microsoft’s AI aspirations were confined to basic machine learning capabilities on underwhelming Snapdragon-based Windows 10 devices.
From Always Connected PCs to Copilot+ PCs
Reflecting on Microsoft’s marketing strategies, which often seem paradoxical, it’s worth revisiting the initial announcement of the partnership between the Windows team and Qualcomm to launch Windows 10 on Arm-based PCs. These devices were branded as Always Connected PCs, highlighting their integrated cellular modems—a hallmark of Arm-based mobile devices. Intel soon followed suit, extending the Always Connected PC label to select x86-based models.
Fast forward to recent years, when Microsoft and Qualcomm introduced a new initiative featuring Nuvia-powered chips for Windows 11 on Arm. This time, the focus shifted to Copilot+ PCs, emphasizing the powerful neural processing unit (NPU) embedded in these new system-on-chips (SoCs) and its potential to deliver exceptional on-device AI experiences. Intel once again joined the fray, adopting the Copilot+ PC branding for its x86 offerings.
In both instances, Microsoft seemed to obscure the key advantages of Arm-based PCs over traditional x86 counterparts. Initially, the appeal centered on efficiency and battery life, which evolved to encompass reliability, performance, and compatibility as we transitioned into the Snapdragon X era. The first Snapdragon-based PCs faced significant limitations, but the introduction of Snapdragon X in 2024 marked a transformative leap forward.
Consider the underlying intent behind these initiatives. Terry Myerson, who spearheaded the Windows organization during the Windows 10 era, once expressed a desire to eliminate the omnipresent Intel stickers from PCs. While this statement was memorable, it encapsulated a deeper frustration: Intel’s longstanding neglect of Microsoft’s urgent need for more efficient mobile processors for laptops. If the partnership with Qualcomm merely compelled Intel to acknowledge the shifting market dynamics, Myerson deemed it a success. Ultimately, his goal was to enhance the PC experience by integrating the best aspects of mobile computing—instant-on capabilities, reliable performance, and exceptional battery life.
In essence, these efforts represented Microsoft’s ongoing commitment to improving the PC landscape. When Windows 10 on Arm was first announced, it promised to deliver a familiar Windows experience, complete with essential applications and enterprise functionalities, on a truly mobile and power-efficient platform. Unlike its predecessor, Windows RT, Windows 10 on Arm could run a wide array of applications, including Adobe Photoshop and Microsoft Office, and later expanded to support x64 applications. This hybrid approach aimed to merge the strengths of both PC and mobile technologies.
Such initiatives have the potential to elevate the entire industry. The Always Connected PC and Copilot+ PC movements encouraged PC manufacturers to adopt new chipsets, leading to the creation of sleek, lightweight laptops that appealed to consumers. Component manufacturers were also prompted to develop compatible drivers and utilities. Microsoft actively engaged with developers to facilitate the transition of their applications to native Arm formats. While these efforts directly competed with established x86 chipsets from Intel and AMD, Microsoft’s overarching hope was to inspire these companies to innovate and improve their offerings.
From NPUs to hybrid AI
The launch of Copilot+ PCs in 2024 marked a significant milestone in the journey toward hybrid AI capabilities, but it was not an isolated event. Microsoft had been advocating for machine learning and AI for years prior, introducing the Windows Studio Effects suite for NPU-equipped Snapdragon PCs during the Windows 10 era. Qualcomm, too, began promoting hybrid AI on PCs in May 2023, a year ahead of the Copilot+ PC launch and the arrival of the groundbreaking Snapdragon X chipsets. The term “AI PC” had already been circulating in the industry to describe any PC equipped with an NPU, even those that predated the Copilot+ era.
However, the Copilot+ PC, with its requirement for a 40 TOPS NPU, was ahead of its time. While it promised transformative capabilities, the reality was a series of incremental enhancements that many users noted could function adequately—if less efficiently—on standard PCs utilizing a GPU or CPU. In essence, there was no standout application, and many improvements remained invisible to users.
This situation posed a challenge for marketing, exacerbated by the perception that the NPU requirement was an artificial barrier, seemingly compelling users to upgrade their PCs. This created a perplexing cycle where potential customers struggled to see the value in investing in a Copilot+ PC for local AI, while developers were slow to adopt these capabilities due to the limited audience of Copilot+ users compared to the broader Windows 11 community.
As we enter the second year of the Copilot+ PC era, second-generation Snapdragon X2-based PCs are gradually making their debut, albeit at a slower pace than their predecessors. PC manufacturers continue to favor hardware based on Intel and AMD chipsets, including Copilot+ PCs. Despite Microsoft’s efforts to expand the unique features available on this platform, including the notable Click to Do functionality, the overall offering remains a patchwork of capabilities that could arguably function on any modern PC.
Simultaneously, the landscape of AI in the cloud has undergone a dramatic transformation, not only in terms of functionality but also in cost. After years of subsidizing the expenses associated with cloud-based AI interactions and investing heavily in infrastructure, Big Tech is now confronting the reality of usage-based pricing. This shift has led to soaring costs, reminiscent of rising gasoline prices, leaving consumers grappling with the true financial implications of this technology.
Ironically, the local AI that Microsoft has championed could potentially alleviate this issue if it were to operate effectively. However, various factors—including the aforementioned cycle, the NPU requirement, and the absence of a killer app—have hindered its performance. But what does it mean for local AI to “work well”?
At its core, the challenge lies in Microsoft’s failure to meet customers where they are. While an NPU offers significant efficiency advantages over a CPU or GPU, it is essential to recognize that most users possess PCs equipped with capable CPUs and integrated GPUs, with many also featuring dedicated GPUs. With the advancements in modern Intel, AMD, and Qualcomm chipsets, as well as the forthcoming Nvidia Spark chipsets, these non-NPU components are more than capable of handling local AI workloads. Users with powerful PCs prioritize functionality over efficiency; they simply want their systems to perform effectively.
When discussing hybrid AI, we typically refer to a combination of local and cloud-based AI. This interaction can occur in both directions: a local AI solution may defer to cloud-based AI when necessary, while cloud AI can revert to local AI when usage limits are reached. However, I believe there is a deeper complexity at play. True hybrid AI requires intelligent orchestration of AI workload traffic, taking into account the capabilities of the device in use and the availability of both free and paid cloud-based AI resources. It must adeptly navigate between the CPU, GPU, and NPU of the device and the cloud-based options at hand.
Today, we have AI PCs equipped with various NPUs, as well as Copilot+ PCs boasting 40+ TOPS NPUs. Soon, we will witness the emergence of Nvidia-based AI PCs that also qualify as Copilot+ PCs, featuring powerful NPUs alongside robust integrated GPUs. Whether this new class of PCs will receive a distinct branding remains uncertain, but it is evident that the concept of Copilot+ PC is poised for evolution. It may either fade away or expand to encompass these new Nvidia-based systems and their formidable GPUs.
From Build 2026 to building a better foundation for Windows
The insights gleaned from Build 2026 were not merely a result of effective communication from Microsoft; rather, they emerged from the various announcements made during the conference. Among the highlights were:
- Microsoft confirmed that Windows 11 would integrate AI agents, including local AI agents that operate on-device rather than relying on the cloud. Additionally, the introduction of hybrid agents—those combining cloud and local components—was unveiled, featuring local models that power what Microsoft refers to as sub-agents.
- The capabilities of Microsoft’s local AI models, previously known as small language models (SLMs), are being enhanced. The transition from Microsoft Phi to the more powerful Aion family of SLMs will facilitate the deployment of local agents in Windows 11.
- Microsoft is broadening the Windows AI APIs for developers to encompass CPUs and GPUs, rather than limiting them to NPUs. This expansion will commence with features such as speech-to-text recognition, text intelligence, and video super-resolution, all of which will function on-device across a combination of CPU, GPU, and NPU without necessitating a “cloud round trip.”
- The newly announced Nvidia chipsets will be the first to enable true large language models (LLMs)—previously confined to the cloud—to run locally. These local models support up to 1 trillion parameters, a feat that was once unimaginable. If marketed by Apple, this capability would likely be touted as a “datacenter on your lap.” However, these advancements are expected to become accessible to mainstream PCs in the near future, given the rapid evolution of AI.
It may be somewhat ironic that Microsoft describes these local AI capabilities as “unmetered” intelligence, yet it resonates with anyone facing monthly AI usage overage fees. In this hybrid AI landscape, LLMs—now referred to by Microsoft as frontier models—will address complex challenges, while other tasks will be handled locally and at scale. Ultimately, the orchestration of these processes is paramount.
To elaborate, hybrid AI encompasses both local and cloud AI components, necessitating an orchestrator capable of assessing the CPU, GPU, and NPU capabilities of the device in use and directing AI workloads accordingly. Furthermore, this orchestrator must evaluate the capabilities of local AI models—whether installed or not—and align them with the appropriate CPU, GPU, and NPU resources, seamlessly transitioning to the cloud when necessary. This orchestration is crucial for ensuring that AI agents can efficiently navigate the local/cloud divide, ideally with the security provided by Microsoft Execution Containers.
One final observation: this intricate orchestration may explain why Microsoft is finally addressing the longstanding “pain points” that users have raised for years without resolution. My evolving understanding of Microsoft’s intentions suggests that these changes are driven by the need for a robust local foundation to support hybrid AI. The enshittification of Windows 11 had left its desktop platform in a precarious state, plagued by both tangible issues—such as erratic Windows Updates—and perceived problems, like the public relations fallout from Recall. The security, reliability, and resource management improvements announced over the past year are not merely for the benefit of individual users; they are fundamentally aimed at supporting the broader vision of hybrid AI, which holds significant importance for Microsoft.
While this remains a theory, it is grounded in observable truths. The most encouraging aspect is that regardless of the underlying motivations, Microsoft is actively working to enhance Windows 11, and ultimately, all users stand to gain from these efforts.
