NVIDIA Enters the AI PC Market with RTX Spark, Reshaping Industry Power Dynamics

The PC industry, with its 40-year history, is truly on the verge of a major transformation.

In early June, NVIDIA unveiled the new superchip RTX Spark for Windows-based personal computers at GTC Taipei 2026, officially entering the market for core PC processors. With Microsoft’s presence at this grand event aimed at redefining the AI PC, NVIDIA’s announcement carried the aura of official endorsement.

Meanwhile, Acer, Asus, Dell, Gigabyte, HP, Lenovo, and MSI—all major players in the PC terminal market—have collectively rallied behind the same chip.

Moreover, at the Microsoft Build 2026 conference two days later, Microsoft CEO Satya Nadella redefined Windows as “the native platform for local AI agents” and launched the Surface RTX Spark Dev Box—a desktop workstation capable of running 120B-parameter large models locally.

Jensen Huang stated in a video link-up that after more than four decades, personal computers are entering a transformative turning point, as AI agents are reshaping the PC industry. NVIDIA and Microsoft are “reinventing” the personal computer, enabling local PCs to possess independent AI agent capabilities, as PCs evolve from personal computers to personal AI.

He gave an example: when users are away, they can send a message to their PC, allowing the local agent to invoke tools, modify code, and advance design, then continue iterating with the user. He emphasized that the PC is no longer just a tool operated by humans—it is also becoming an AI assistant capable of running tasks continuously.

However, a fact that is easily overlooked is that the concept of AI PC was not first introduced by NVIDIA—in fact, Intel is the originator of the AI PC concept.

Back in January this year, Intel unveiled its brand-new third-generation Core Ultra processor platform at CES. For Intel itself, this marked the debut of its Intel 18A advanced process technology and is pivotal to its future development. For the PC industry, "Core Ultra" carries additional significance—it can be seen as a key anchor point in the emerging AI PC landscape.

However, after NVIDIA made a major move into the AI PC market, Intel has indeed appeared more passive.

Moreover, it cannot be ignored that other players are gradually entering this major industry shift in personal computing: Qualcomm continues to invest heavily in PC chips, AMD has successively launched new products with integrated AI computing power, and Apple has demonstrated the feasibility of running ARM architecture on personal computing devices with its M-series chips.

All these actions point to the same key technological trend: AI is unprecedentedly moving toward personal computing devices.

When it comes to the story of the PC industry, the Wintel alliance is certainly the first to mention—but it has never been limited to just Wintel.

In 1980, IBM was preparing to produce its own branded PC. At the time, IBM was a dominant force in the computer industry, while Intel, though somewhat successful, had limited influence. Among microprocessor competitors, Motorola was also present and overall stronger than Intel.

However, Don Estridge, who was in charge of IBM’s PC business at the time, made a decision that would shape the landscape for decades to come: he awarded the processor order to Intel and the operating system order to Microsoft.

At that time, Microsoft was not yet considered a giant in the software industry. However, the story that followed this partnership was undoubtedly a defining moment in the history of the PC industry. In the early 1990s, Microsoft and Intel jointly took control of the PC market away from IBM.

This is the "Wintel alliance"—a personal computer architecture composed of Microsoft’s Windows operating system and Intel’s CPUs. For over two decades thereafter, the Wintel alliance dominated the desktop market, leveraging Intel’s Moore’s Law and continuous upgrades to Microsoft’s Windows system to jointly control downstream PC manufacturers and generate massive profits.

Over the past two decades, the power structure in the PC industry has been such that Intel controlled the core processors, Microsoft controlled the operating system, and PC manufacturers could only compete on price within the rules set by the upstream players.

But to understand today’s landscape, looking only at Intel and Microsoft isn’t enough—a third name must also be included: NVIDIA.

But for four decades of Wintel dominance, NVIDIA’s position was very clear: an accessory provider.

When PC users buy a computer, they think, “This computer uses an Intel processor.” Graphics cards? Those are just add-ons for gaming or rendering. NVIDIA’s GPU is merely a component plugged into a PCIe slot; the core architecture of the PC is determined by the CPU, managed and coordinated by the operating system.

Over several decades, NVIDIA's role has become increasingly important, but it has not altered the underlying logic of PCs; strictly speaking, it has merely been a performance amplifier.

Until 2020, Apple announced it would phase out Intel chips in its Mac lineup and adopt its own custom-designed chips. The M1 chip demonstrated one thing: integrating the CPU, GPU, NPU, unified memory, and system scheduling all into one package truly delivers a different user experience. But that was within Apple’s own ecosystem—Windows’ landscape saw little change.

In 2024, Microsoft defined the Copilot+ PC, requiring NPU performance of over 40 TOPS. The Qualcomm Snapdragon X Elite, Intel Core Ultra, and AMD Ryzen 8000 series all debuted simultaneously. AI PC shipments surged from concept to over ten million units within a year, doubling their market penetration rate.

Canalys data shows that global PC shipments reached 262 million units in 2024, a 3.1% year-over-year increase—the first growth after two consecutive years of decline; global PC shipments are expected to reach 274 million units in 2025, representing a 4.1% year-over-year increase, signaling that the global PC industry has transitioned from a period of demand exhaustion to a phase of steady recovery.

But the market soon identified an issue: most AI capabilities still rely on the cloud, and local computing power lacks practical applications. Consumers who bought them at home found that AI PCs were not fundamentally different from regular PCs.

By 2025, more industry players are realizing that AI PCs cannot rely solely on raw computing power—they must address the question of “what local AI applications are available.” Canalys predicts that AI PC penetration in mainland China will reach 34% in 2025 and rise further to 52% in 2026. However, overall PC market growth remains modest—in fact, IDC and Gartner forecast a double-digit decline in global PC shipments by 2026. This shift is fundamentally a structural replacement driven by corporate refresh cycles and consumer upgrades, rather than the creation of an entirely new market space worth hundreds of millions of units.

In other words, the profit distribution logic of this market cycle is: those who secure critical positions in the BOM (bill of materials) upgrades and value chain shifts reap the biggest rewards, rather than all PC manufacturers sharing equally. For NVIDIA, this marks a transition from being a “component supplier” to becoming a “platform provider.”

If successful, it would rewrite not just one or two quarters of shipments, but the underlying power structure of the Wintel alliance over the past three decades.

For NVIDIA, it doesn’t need PCs to serve as its new growth driver—so why is Huang choosing to enter the AI PC market now?

This answer is also very clear.

In March 2026, at the annual GTC conference, NVIDIA, commemorating the 20th anniversary of CUDA, revealed a number that captivated the entire AI industry: 6 million developers.

Six million people write code in CUDA, running on NVIDIA GPUs, covering AI training, inference, scientific computing, graphics rendering, and video production. The entire AI industry's software stack is built on CUDA at its core.

What does 6 million mean?

There are approximately 30 million iOS developers and 7 million Android developers. The number of CUDA developers has reached one-third the scale of the major mobile platforms.

But CUDA's true power lies not in the numbers, but in the migration cost. Developers write AI code in CUDA → PyTorch and TensorFlow are optimized for CUDA by default → NVIDIA GPUs sell better → more developers continue choosing CUDA. This is NVIDIA’s version of an ecosystem flywheel, closely mirroring the developer ecosystem logic of Android.

A developer starting out with PyTorch typically defaults to the CUDA backend from the beginning; once a team has built up a codebase, toolchain, and engineering expertise on CUDA, migrating to ROCm (AMD’s equivalent platform) or another platform becomes challenging.

Theoretically, AMD's official migration tool claims less than 5% code changes, but any involvement of custom kernels, memory access optimizations, or deep dependencies on cuBLAS/cuDNN call chains will require far more than 5% effort.

This is why NVIDIA continues to maintain a high market share in the AI training market, even though AMD's MI300 series performs well in benchmarks.

Where are the 6 million past CUDA developers? In data centers, using GPUs that cost tens of thousands of dollars each. RTX Spark brings CUDA to laptops.

After all, RTX Spark is not a graphics card—it’s a complete SoC, integrating 20 ARM Grace CPU cores, 6,144 CUDA cores, fifth-generation Tensor Cores, and up to 128 GB of unified LPDDR5X memory. NVIDIA has announced AI performance of up to 1 petaflop, enabling local execution of large language models with up to 120 billion parameters.

In the future, the code written by these people can be run directly on a laptop without any modifications or recompilation—the architecture is compatible.

At the event, Jensen Huang also said: “We are going to reinvent humanity’s most important tool”—referring to the PC.

He also announced that the second- and third-generation Spark chips are already in planning, and future NVIDIA platform architectures will each include a Spark chip, with over 30 laptops and more than 10 desktops launching simultaneously.

Moreover, Huang Renxun envisioned an even longer-term future—from the current Blackwell, to the upcoming Rubin, and then Feynman—NVIDIA has laid out its chip roadmap for desktops, laptops, and workstations all the way to 2030.

However, whether CUDA can truly reach every terminal depends on a variable NVIDIA cannot control: price.

Global DRAM is currently in a supply-constrained cycle, leading to rising memory prices; the starting price for the first batch of laptop products will not be low; we aim for CUDA to serve not only heavy users but also a broader range of products, aligned with cost curves for manufacturing processes and memory.

NVIDIA chose this moment to make its move. Simply put, it saw an opportunity: demand for computing power is shifting from the cloud to the edge.

"Large and sparse": models with a large number of parameters but relatively few activated parameters; these models require higher memory capacity but not high computational power, making them more suitable for deployment on edge devices.

"Small and specialized" models, created through distillation and model acceleration techniques, deliver strong performance in their specific domains and are also well-suited for deployment on edge devices.

These two trends in large models are the foundation of the rise of edge-side AI.

As a key player in edge AI, Intel has consistently strengthened its edge computing capabilities over the past few years, increasing edge compute performance by 48 times in three years. Additionally, Microsoft has begun taking edge AI seriously; ARM architecture has gained its first substantial OEM support on Windows; and the CUDA developer base has reached a significant scale.

Entering the AI PC market at this stage is both a crucial step for NVIDIA to seize control of the edge ecosystem and an inevitable choice to ensure the long-term competitiveness of its CUDA ecosystem.

Currently, the PC industry is showing several key signals.

The first signal: PCs are shifting from a CPU-centric to an AI SoC-centric architecture.

Apple's M series has validated the feasibility of integrating CPU, GPU, NPU, unified memory, and system scheduling into a single package.

Intel’s Lunar Lake is also beginning to integrate memory into the package, and AMD’s Strix Halo is pursuing a large memory pool approach. Now, NVIDIA is entering the scene with Blackwell GPUs, Arm CPUs, unified memory, CUDA, and the RTX ecosystem—essentially layering its data center AI platform strategy onto personal computers.

It’s no longer just about adding a graphics card to a PC—it’s about becoming an integral part of the PC platform itself. CPU, GPU, AI computing power, unified memory, and software ecosystem—all bundled together—this isn’t “accessory thinking,” it’s “platform thinking.”

There are three benefits here.

First, NVIDIA has moved its GPU advantages down to the SoC level. While AI PCs have been hyping NPU TOPS, the real currency for running local large models, AI video, 3D creation, and gaming remains the GPU and memory pool. If RTX Spark can leverage unified memory to solve data movement and model loading issues, the user experience will be smoother than traditional “CPU + discrete GPU + separate memory” setups.

Second, NVIDIA continues to embed CUDA, RTX, DLSS, and TensorRT deeper into the PC stack. This is even more critical than the hardware itself. In the AI era, whoever controls the development frameworks, inference libraries, model optimization tools, and creator toolchains holds the platform power. Huang clearly understands that chips are merely the ticket in—the ecosystem is the moat.

Third, NVIDIA is now targeting the most lucrative component in the total bill of materials for complete systems. Previously, in a high-end Windows PC, the CPU budget went to Intel or AMD, and the discrete GPU budget went to NVIDIA. In the future, if NVIDIA’s AI SoC becomes the core of the entire system, it will capture not only the value of the graphics card but also the value of the CPU platform, the premium for AI experience, and control over developer ecosystem pricing.

The second signal: PCs are shifting from being tools operated by humans to platforms where humans collaborate with agents.

Jensen Huang paints a future where, while you're out, you can message your PC to have a local agent invoke tools, modify code, and advance designs—then return to continue iterating. Your PC is no longer just a tool to be operated, but is becoming an AI assistant capable of running tasks continuously.

Windows is undergoing the same shift—Microsoft has redefined Windows as the native platform for local AI agents and introduced secure execution containers and OpenClaw for Windows, enabling AI agents to safely execute multi-step tasks in a controlled environment. This means Windows is no longer just a container for applications, but a runtime for agents.

The third signal: six million CUDA developers worldwide have found a new hardware platform.

NVIDIA has brought CUDA to every laptop with RTX Spark. Behind this is a complete ecosystem flywheel: developers are familiar with CUDA → run natively on RTX Spark → optimize applications and models → attract more users to purchase → drive more developers to join.

The iteration cycle of GPUs is measured in years, while developer habit formation occurs across generations. Once this flywheel starts spinning on the PC side, latecomers have almost no chance of catching up.

However, the adoption rate and commercial success of RTX Spark depend on three key factors: first, whether the final pricing can reach a broader user base; second, whether the software ecosystem for Windows on ARM can address its critical gaps in the medium to short term; and third, whether Microsoft can truly drive local AI agents from concept to killer applications that compel users to upgrade their devices.

Looking back, in this industry transformation centered on AI PCs, it is less about NVIDIA entering the AI PC market to reshape the entire PC industry’s power dynamics, and more about AI technology itself seeking its optimal role within a 40-year-old industry—a technological trend that no player can resist.

And don't forget that Intel has not gone against this major trend.

In early 2026, Intel also targeted the local market with the third-generation Core Ultra processor (codenamed Panther Lake)—the world’s first consumer computing platform built on Intel 18A process technology, featuring RibbonFET full-gate-all-around transistors and PowerVia backside power delivery, with a maximum AI computing power of up to 180 TOPS.

To some extent, Intel is also moving in the same direction.

So, ultimately, whether it’s NVIDIA, Microsoft, or Intel, each is just one player in this technological transformation. The difference lies in who can identify this trend earlier, who can transform more decisively, and who acts faster—only then can they keep pace with technological advancements and benefit from them.

From this perspective, Microsoft’s role in the PC industry is even more “timeless.”

Regardless, one thing is certain: with NVIDIA’s entry, the new era of AI PCs has arrived, and the PC industry is genuinely being reinvented—now, it’s up to Apple to decide what historical move it will make on its own Mac turf.

This article is from the WeChat public account "Timeline Timelines," authored by Zhao Ming.