It's not Jensen Huang who wants to transform the PC, but the PC that needs to revolutionize itself.

The PC industry, which has a history of 40 years, is really going to experience a significant change.

At the beginning of June, NVIDIA launched a brand - new super chip, RTX Spark, for Windows - based personal computers at GTC Taipei 2026, officially announcing its entry into the core processor market for personal computers. At this grand event aimed at redefining AI PCs, Microsoft's support gave NVIDIA's event an air of "officially endorsed".

Meanwhile, a group of manufacturers representing the PC terminal market, including Acer, Asus, Dell, Gigabyte, HP, Lenovo, and MSI, all stood behind the same chip.

Moreover, at the Microsoft Build 2026 conference two days later, Microsoft CEO Sayta Nadella redefined Windows as "the native operating platform for local AI agents" and introduced the Surface RTX Spark Dev Box equipped with RTX Spark - a desktop workstation capable of running a large - model with 120 billion parameters locally.

Jensen Huang said in a video call that after more than forty years of development, personal computers are approaching a new turning point. AI agents are reshaping the PC industry. NVIDIA and Microsoft are "reinventing" the personal computer, enabling local PCs to have independent AI intelligent agency capabilities. PCs are evolving from personal computers to personal AIs.

He gave an example: When users are out, they can send messages to their PCs, allowing the local agent to call tools, modify code, and advance design, and then continue to iterate with the users. He emphasized that the PC is no longer just a tool operated by humans but is also starting to become an AI assistant that can continuously run tasks.

However, an easily overlooked fact is that the concept of AI PC was not first introduced by "NVIDIA". In fact, Intel was the one who proposed the concept of AI PC.

As early as January this year, Intel launched a new third - generation Core Ultra processor platform at CES. For Intel itself, this was the debut of the Intel 18A advanced process technology and was also related to Intel's future development. For the PC industry, "Core Ultra" actually has another meaning. It can be regarded as a key anchor point in the emerging field of AI PCs.

However, after NVIDIA's large - scale entry into the AI PC market, Intel did seem a bit passive.

Moreover, it cannot be ignored that in this major transformation of the personal computing industry, other players are also gradually entering the market. For example, Qualcomm is continuously increasing its investment in PC chips, AMD has successively launched new products with integrated AI computing power, and Apple has proven the feasibility of the ARM architecture running on personal computing devices with its M - series chips.

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

Building high - rise buildings, hosting banquets, and then the buildings collapse

When talking about the story of the PC industry, of course, the Wintel Alliance must be mentioned first, but it is never just about the Wintel Alliance.

In 1980, IBM was preparing to produce its own brand of PCs. At that time, IBM was a prominent figure in the computer field, while Intel, although having achieved some success, had limited influence. Among the peers in the microprocessor field, Motorola had a stronger overall strength than Intel.

However, Don Estridge, who was in charge of the IBM PC business at that time, made a decision that would affect the pattern for the next few decades: the processor procurement order went to Intel, and the operating system order went to Microsoft.

At that time, Microsoft could not be considered a giant in the software industry. However, the subsequent story of this combination was undoubtedly a decisive stroke in the development history of the PC industry. In the early 1990s, Microsoft and Intel jointly snatched the dominance of PC computers from IBM.

This is the "Wintel Alliance" - a personal computer architecture composed of Microsoft's Windows operating system and Intel's CPU. In the following more than twenty years, the Wintel Alliance monopolized the desktop market. Relying on Intel's Moore's Law and the iterative upgrades of Microsoft's Windows system, the two companies jointly controlled downstream PC manufacturers and reaped huge profits.

During these more than twenty years, the power structure of the PC industry was as follows: Intel controlled the core processor, Microsoft controlled the operating system, and PC manufacturers could only compete on price within the rules set by the upstream.

But to understand the current situation, it is not enough to just look at Intel and Microsoft. A third name must also be included - that is NVIDIA.

During the forty - year hegemony of the Wintel Alliance, NVIDIA's positioning was very clear: a component supplier.

When PC users bought a computer, they thought about "this computer uses an Intel processor". What about the graphics card? It was an additional component for gaming and rendering. NVIDIA's GPU was just a component plugged into the PCIe slot. The core architecture of the PC was determined by the CPU and then managed and allocated by the operating system.

Over the decades, although NVIDIA's role has become increasingly important, it has not changed the underlying logic of the PC. Strictly speaking, it is just a performance amplifier.

It was not until 2020 that Apple announced that it would abandon Intel chips in its Mac series and adopt self - developed chips. The M1 chip proved one thing: when the CPU, GPU, NPU, unified memory, and system scheduling are all integrated, the user experience is indeed different. However, this was within Apple's own ecosystem, and the pattern of the Windows camp did not change much.

In 2024, Microsoft released the definition of Copilot + PC, requiring the NPU computing power to reach more than 40 TOPS. Qualcomm Snapdragon X Elite, Intel Core Ultra, and AMD Ryzen 8000 series all made their debuts. The shipment volume of AI PCs quickly exceeded 10 million units from the conceptual stage within a year, and the penetration rate doubled.

Canalys data shows that in 2024, the global PC shipment volume reached 262 million units, a year - on - year increase of 3.1%, the first positive growth after two consecutive years of decline. In 2025, the global PC shipment volume is expected to reach 274 million units, a year - on - year increase of 4.1%, indicating that the global PC industry has moved from the previous period of over - demand to a stable recovery period.

However, the market soon discovered a problem: most AI capabilities still rely on the cloud, and there is a lack of application scenarios for local computing power. Consumers found that there was no essential difference between an AI PC and an ordinary PC when they brought it home.

In 2025, more industry players began to realize that AI PCs cannot just pile up computing power. The problem of "what local AI applications are available" must be solved. Canalys predicts that the penetration rate of AI PCs in mainland China will reach 34% in 2025 and further rise to 52% in 2026. However, the growth of the global PC market is not very impressive. IDC and Gartner even predict that the PC shipment volume may experience a double - digit decline in 2026. In essence, it is a structural replacement of enterprise computer replacement and consumer upgrades, rather than a new market space of hundreds of millions of units emerging out of thin air.

In other words, the profit - distribution logic of this round of market is: whoever occupies a key position in the BOM (Bill of Materials) upgrade and value - chain transfer will get the benefits, rather than all PC manufacturers sharing equally. For NVIDIA, this time it has jumped from a "component supplier" to a "platform provider".

If successful, it will rewrite not just the shipment volume for one or two quarters, but the underlying power structure of the Wintel Alliance over the past thirty years.

Jensen Huang's focus on entering the market: still the ecosystem

For NVIDIA, it does not need the PC as its new growth point. Why does Jensen Huang choose to enter the AI PC market at this time?

The answer is actually quite clear.

In March 2026, at the annual GTC conference, while commemorating the 20th anniversary of CUDA, NVIDIA announced a figure that would catch the eye of the entire AI industry: 6 million developers.

These 6 million people write code using CUDA, which runs on NVIDIA's GPUs. It covers AI training, inference, scientific computing, graphic rendering, and video production. The entire software stack of the AI industry is based on CUDA at the bottom.

What does 6 million mean?

There are about 30 million Apple iOS developers and about 7 million Android developers. The scale of CUDA developers has reached one - third of that of mainstream mobile platforms.

However, the real power of CUDA lies not in the numbers but in the migration cost. Developers write AI code using CUDA → PyTorch and TensorFlow are optimized for CUDA by default → NVIDIA's GPUs sell better → more developers continue to choose CUDA. This is NVIDIA's version of the ecosystem flywheel, which is highly similar to the developer ecosystem logic of Android.

When a developer starts learning PyTorch from scratch, the framework defaults to use the CUDA backend. Once a team has accumulated a code library, toolchain, and engineering experience on CUDA, is it easy to migrate to ROCm (AMD's similar platform) or other platforms?

Theoretically, AMD's official migration tool claims that the code modification is less than 5%. However, whenever it involves custom kernels, memory access optimization, or a call chain deeply dependent on cuBLAS/cuDNN, the workload is definitely more than 5%.

This is why even though the performance evaluation of AMD's MI300 series is not bad, NVIDIA still maintains a high market share in the AI training market.

Where were the 6 million CUDA developers in the past? In data centers, using GPUs worth tens of thousands of dollars each. What RTX Spark does is bring CUDA to laptops.

After all, RTX Spark is not a graphics card; it is a complete SoC. It integrates a 20 - core ARM Grace CPU, 6144 CUDA cores, fifth - generation Tensor Cores, and up to 128GB of LPDDR5X unified memory. The AI computing power data announced by NVIDIA is as high as 1 Petaflop, supporting the local operation of a large - language model with 120 billion parameters.

In the future, the code written by these people can run directly on a laptop without modification or recompilation. The architecture is compatible.

Jensen Huang also said at the press conference: "We are going to reinvent the most important tool for humanity"; he was referring to the PC.

He also announced that the second - and third - generation chips after RTX Spark are already in planning. In the future, each generation of NVIDIA's platform architecture will include a Spark chip, and more than 30 laptops and more than 10 desktops will be launched simultaneously.

Moreover, Jensen Huang also envisioned a more distant future - from the current Blackwell, to the upcoming Rubin, and then to Feynman. NVIDIA has laid out the chip roadmap for desktops, laptops, and workstations all the way to 2030.

However, whether CUDA can truly be integrated into every terminal depends on a variable that NVIDIA cannot control: price.

The global DRAM market is currently in a period of tight supply, and memory prices are rising. The starting price of the first - batch laptop products will not be low. To make CUDA cover more than just heavy - users, more generations of products are needed, along with the cost curves of the manufacturing process and memory.

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

For "large and sparse" models with a large number of model parameters but relatively few activated parameters, such models require higher memory capacity rather than very high computing power and are more suitable for running on the edge; for "small and specialized" models formed through distillation and model acceleration technologies, which perform well in specific professional fields, they are also suitable for edge deployment.

These two major trends of large models are the root cause of the rise of edge AI.

As an important player in edge AI, Intel has also been continuously strengthening its edge computing power over the years. In three years, it has increased the edge computing power by 48 times. In addition, Microsoft has also started to seriously consider edge AI. The ARM architecture has received large - scale OEM support on Windows for the first time, and the number of CUDA developers is already large enough.

Entering the AI PC market at this time is not only a crucial step for NVIDIA to seize the edge ecosystem but also an inevitable choice for NVIDIA to ensure the long - term competitiveness of the CUDA ecosystem.

The self - revolution of the PC industry has begun

Currently, the PC industry has shown several key signals.

The first signal is that PCs are shifting from "CPU - centric" to "AI SoC - centric".

Apple's M - series has verified the feasibility of the direction of "integrating CPU + GPU + NPU + unified memory + system scheduling".

Intel's Lunar Lake has also started to integrate memory into the package, and AMD's Strix Halo is also following the path of a large - memory pool. Now, NVIDIA is entering the market with its Blackwell GPU, Arm CPU, unified memory, CUDA, and RTX ecosystem, which is equivalent to applying the AI platform approach used in data centers to personal computers.

It is no longer just adding a graphics card to the PC but directly becoming part of the PC's main platform. Packing the CPU, GPU, AI computing power, unified memory, and software ecosystem together is no longer the "component thinking" but the "platform thinking".

There are three benefits in this.

First, NVIDIA has advanced its GPU advantage to the bottom layer of the SoC. In the past, AI PCs talked about NPU TOPS, which sounded exciting. However, when it comes to running local large models, AI videos, 3D creation, and games, the GPU and memory pool are the real essentials. If RTX Spark can solve the problems of data transfer and model loading with unified memory, the experience will be smoother than the traditional "CPU + discrete graphics card + separate memory" configuration.

Second, NVIDIA continues to integrate CUDA, RTX, DLSS, TensorRT, and other technologies into the bottom layer of the PC. This is more crucial than the hardware. In the AI era, whoever controls the development framework, inference library, model optimization, and creator toolchain will have platform power. Jensen Huang clearly understands that the chip is just the ticket, and the ecosystem is the moat.

Third, NVIDIA is starting to grab the most profitable part of the PC's BOM. In the past, for a high - end Windows computer, the money for the CPU went to Intel or AMD, and the money for the discrete graphics card went to NVIDIA. In the future, if NVIDIA's AI SoC becomes the core of the whole machine, it will not only capture the value of the graphics card but also the value of the CPU platform, the premium of the AI experience, and the pricing power of the developer ecosystem.

The second signal is that PCs are shifting from "tools operated by humans" to "platforms for humans and agents to work together".

Jensen Huang described a future where when you are out, you can send messages to your PC, allowing the local agent to call tools, modify code, and advance design, and then come back and continue to iterate with you. The PC is no longer just a tool operated by humans but is also starting to become an AI assistant that can continuously run tasks.

The positioning of Windows is also undergoing the same shift. Microsoft has not only redefined Windows as the native operating platform for local AI agents but also introduced a secure execution container and OpenClaw for Windows, enabling AI agents to safely execute multi - step tasks in a controlled environment. This means that Windows is no longer just a container for applications but a runtime environment for agents.

The third signal is that 6 million global CUDA developers have found a new hardware carrier.

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 buy → drive more developers to join.