2026 Computex Taipei’s official theme is "Connecting AI," but the biggest headline-grabber came from NVIDIA CEO Jensen Huang’s pre-show bombshell: the launch of RTX Spark, the world’s first Arm-based AI superchip designed specifically for Windows PCs. From an industry perspective, this was far more than a simple product launch. When 1 petaflop of AI compute, a 20-core Grace CPU, and up to 128 GB of unified memory are packed into a thin-and-light laptop chassis, it crystallizes a core question sweeping the entire AI industry: Is on-device local AI inference finally ready to rival cloud computing power? Is the relationship between cloud and local compute a zero-sum game, or a complementary industrial evolution? And for investors, behind these two pathways, how should AI investment priorities be identified, broken down, and allocated in 2026?
The True Coordinates of the 2026 On-Device AI Market
Discussions around on-device AI in 2024-2025 still carried a strong narrative flavor, but entering 2026, all the data points needed for judgment have a quantifiable foundation. According to the "Edge AI Hardware Market Report" from Mordor Intelligence, released in January 2026, the Edge AI hardware market was valued at approximately $26.17 billion in 2025 and is projected to grow to $30.74 billion in 2026, reaching an estimated $68.73 billion by 2031, with a compound annual growth rate (CAGR) of roughly 17.46% from 2026 to 2031. Looking at the more specific Edge AI chip market (excluding peripheral hardware), data from Research and Markets, released in February 2026, showed sales of approximately $7.05 billion in 2025, growing to $8.33 billion in 2026, representing a CAGR of about 18.2%. While different agencies report varying absolute figures due to different statistical scopes, the directional consensus is remarkably consistent: on-device AI is in a high-growth trajectory, with an expected annual growth rate of 15%-20% over the next five years.
Another set of more telling data comes from changes in AI PC penetration. According to consensus estimates from CITIC Securities and Soochow Securities, the AI chip penetration rate for smartphones and PCs is expected to reach 45% and 62% respectively in 2026. The global on-device AI market size is projected to jump from RMB 321.9 billion in 2025 to RMB 1.22 trillion by 2029, a staggering CAGR of 40%. In Q2 2026, the share of AI processing done on the device side globally reached 52% for the first time, signaling that the entire industry has crossed the critical inflection point from "cloud dependence" to "local offloading." The underlying drivers supporting this penetration logic include the tangible acceleration of the AI smartphone and AI PC upgrade cycle, the rigid demands of data privacy and security regulations favoring local processing, and the large-scale diffusion of generative AI from cloud-based applications to terminal devices.
This is the fundamental coordinate system for understanding the on-device AI investment thesis: the market space is large enough, the growth rate fast enough, and the driving forces already form a realistic foundation of policy, technology, and user demand converging, not just an industry narrative.
Three Industry Signals from RTX Spark
Against this market backdrop, the launch of RTX Spark gains even more penetrating significance. This SoC, jointly built by NVIDIA and MediaTek using TSMC’s advanced 3nm process, is not just an incremental iteration within Intel’s or AMD’s existing CPU product lines. It is a landmark product breaking new ground, from its underlying architecture to its market positioning. In terms of specs, the flagship RTX Spark variant (N1X series) features a 20-core NVIDIA Grace CPU, a Blackwell architecture GPU (6144 CUDA cores), delivering up to 1 petaflop of AI compute, paired with up to 128 GB of unified memory. The connection between GPU and CPU uses NVLink-C2C, offering bandwidth up to 600 GB/s, roughly 5 times that of traditional PCIe Gen5. A lighter version (N1 series) targets the thin-and-light, high-performance laptop segment with power consumption between 18W and 45W. The first batch of laptops featuring these chips is slated for release in Fall 2026 by brands including ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI.
The industry signals from RTX Spark are threefold. First, NVIDIA has formally expanded the AI computing competition from data centers and training clusters into consumer computing devices. Over the past few years, Intel held a significant incumbent advantage in the laptop CPU market, while AMD maintained its competitive edge with CPU+GPU combos and advanced process technology. The launch of RTX Spark essentially signals that NVIDIA is no longer content with its dominant position in AI training. It is now actively pushing on-device AI inference as a second major growth curve alongside its cloud AI ecosystem. Second, RTX Spark validates that the physical bottlenecks for deploying large models on-device are being systematically resolved. Previously, the core constraints hindering large-scale on-device AI deployment were concentrated on power consumption, heat dissipation, and memory bandwidth. RTX Spark achieves 1 petaflop-level AI compute within an SoC TDP of under 80W, while enabling long battery life in a slim chassis. When a consumer-grade laptop chip can locally run a 120-billion-parameter large language model, the compute gradient between on-device AI and cloud AI shifts from "cannot be replaced" to "partially replaceable" – a new phase for the industry. According to Yahoo Tech’s analysis, the core value of RTX Spark lies in bringing server-grade AI capabilities, previously found only in NVIDIA’s DGX Spark developer desktop products, down to the consumer laptop platform. Combined with native support for Windows on Arm, it provides the hardware foundation for Microsoft to rebuild the underlying OS capabilities for local AI agents.
Third, the launch of RTX Spark strengthens the certainty of the on-device AI investment thesis. When an industry titan like NVIDIA marks on-device as a key strategic direction, the signal it sends is structural. It indicates that the prospects for on-device AI have reached sufficient industrial scale and economic rationality to support a long-term product roadmap. This leads to a fundamental judgment: by 2026, the investment thesis for on-device AI is no longer science fiction; it is a reality reflected in industry data and product roadmaps.
The Enduring Dominance of Cloud Computing: The True Weight Revealed by Data
In assessing the on-device AI investment thesis, another set of data should not be ignored: the sheer volume of cloud computing power remains far larger than the on-device side. According to a Morgan Stanley report, the combined capital expenditure of the world’s top 10 cloud providers is expected to reach $632 billion in 2026. The compound annual growth rate for the cloud AI chip market from 2024 to 2029 is still expected to be over 36%. Just considering NVIDIA’s data center business alone, its cumulative revenue for 2026-2027 is projected to be in the hundreds of billions of dollars. In scenarios involving training large-scale foundation models, processing extremely long context reasoning tasks, or querying vast external knowledge bases, the scale and computational elasticity advantages of the cloud will not be replaced by the device side within the foreseeable future. Cloud training and on-device inference are not substitutes; they operate at different levels of market demand. The compute evolution NVIDIA is driving is growing simultaneously in both directions: cloud AI supports the continuous training of large models and new architecture iterations, while on-device AI drives the penetration of AI capabilities into everyday consumer devices and task offloading. These are the coordinated extensions of the same computing ecosystem across different market tiers, not opposing forces.
On this point, a research report from Soochow Securities offers an industry mainstream framework adopted by multiple institutions. "The endgame for on-device models is not to replace cloud large models," the report states, "but to form a collaborative architecture with clear division of labor with the cloud. High-frequency, lightweight, and privacy-sensitive tasks are prioritized for closed-loop local processing on the device. Heavy inference, long-generation, and high-compute tasks are packaged and scheduled on the device, then sent to the cloud for execution." This judgment provides the logical basis for allocating weight between the two investment pathways.
From Compute Roadmaps to Asset Allocation: The Gateway Value of Gate Stock Trading
Once the two investment pathways of cloud and on-device compute are logically broken down, the next inevitable step lands on a practical, executable question: For investors focused on AI industry trends, how can they efficiently allocate core assets from these two pathways within a single account system?
On June 1, 2026, Gate officially launched its real stock trading service, allowing users to directly trade stocks and ETFs from major U.S. securities markets using USDT within the platform. The core differentiator of this service lies in its product nature. Users buy real underlying assets that trade synchronously on NASDAQ and the NYSE. These assets are custodied by a member broker of SIPC (Securities Investor Protection Corporation), and users hold genuine ownership certificates. This is a fundamental distinction from common tokenized stocks or stock perpetual contracts, which are derivatives primarily designed for price tracking.
Gate’s real stock trading currently supports over 10,000 stocks and ETFs, covering major U.S. securities markets and liquidity networks including the New York Stock Exchange (NYSE), NASDAQ, NYSE Arca, NYSE American, and BATS. Users can manage both crypto assets and stocks in a unified manner within a single Gate account, allowing for flexible cross-asset allocation based on changing market conditions. For investors seeking to build a diversified portfolio across different directions within the AI compute space, the value of Gate’s stock service lies in providing an integrated trading gateway without the need to constantly switch back and forth between cryptocurrency and traditional financial markets.
The Logic of Allocating Both Pathways and the Gateway Value of the Gate Platform
Synthesizing all the above judgments, the investment logic for on-device AI and cloud computing power can be framed as follows.
From a structural perspective of compute growth, the center of gravity for AI workloads is shifting from the training side to the inference side. Inference involves both large-scale deployment in cloud data centers and localized inference on-device. The on-device pathway benefits from the upgrade cycle of AI PCs and AI phones, the rapidly increasing penetration of on-device AI chips, and the sinking of AI capabilities into various verticals from wearables to smart vehicles. The core drivers for the cloud pathway are the sustained high prosperity of data center capital expenditure, the economies of scale for training clusters, and the diffusion effect of AI infrastructure into industry applications.
In terms of selecting investment targets, the on-device AI direction could focus on AI chip design firms (NVIDIA, AMD, Qualcomm), key supporters of the ARM architecture ecosystem, as well as advanced process foundries (like TSMC) and high-bandwidth memory suppliers (like SK Hynix). The cloud AI direction could focus on the AI server supply chain, data center networking equipment providers, and leading companies in the cloud computing infrastructure space. On the Gate stock trading platform, the core listed companies from both pathways are already covered. Users don’t need to open multiple brokerage accounts; they can achieve integrated allocation within a single Gate account.
Beyond stock trading, Gate, as a comprehensive financial services platform in the crypto industry, continues to expand its ecosystem boundaries. The platform was founded in 2013 by its CEO, Dr. Han. Today, it has over 54 million registered users globally, with spot trading volume consistently ranking among the top three worldwide. The platform supports trading for over 4,700 crypto assets and more than 10,000 stock assets, dedicated to providing users with a one-stop multi-asset allocation experience. Gate was among the first to implement a 100% proof of reserves. As of March 16, 2026, the overall reserve ratio is approximately 122%, significantly exceeding the industry safety benchmark of 100%, covering nearly 500 different types of user assets.
Conclusion
The competition between on-device AI and cloud computing power is not a binary choice. In 2026, the AI industry is at a historical juncture where both the cloud and on-device pathways are accelerating simultaneously. On-device AI drives the penetration of AI capabilities down to everyday consumer devices, heralding a shift from "using AI well" to "AI everywhere." Cloud AI supports the continuous evolution of super-large models and the heavy lifting of large-scale inference tasks.
From an investment perspective, each pathway has its own growth potential and risk characteristics. The investment flexibility of the on-device route depends more on the deterministic increment from rising AI terminal penetration and the continuous advancement of on-device compute power. The investment logic of the cloud route is built on the sustained high prosperity of data center infrastructure investment. When NVIDIA’s RTX Spark compresses data-center-level compute into a thin-and-light laptop, and when on-device AI processing surpasses cloud processing for the first time, the direction pointed to by the data is clear. The dual logic of compute investment is not the end of market consensus, but the starting point for a new round of industry judgment.
Risk Warning: This content is for informational purposes only and does not constitute investment advice. Stock trading involves market risks, and the cryptocurrency market is highly volatile. Gate’s real stock trading service connects to the U.S. securities market via the compliant broker Alpaca; users purchase real underlying assets. Please make careful decisions based on your own risk tolerance.
