Text/Yang Jianyong

Since the introduction of ChatGPT at the close of 2022, there has been a surge in large-scale model services. Notably, last year's DeepSeek and this year's Seedance video generation large model have captivated global audiences. It's crucial to recognize that these large models demand substantial computational power, necessitating the establishment of extensive AI infrastructure to bolster their competitiveness.

The development of AI infrastructure and the ongoing refinement of large model technologies have spurred breakthroughs in AI applications, propelling the global semiconductor market toward the trillion-dollar milestone with remarkable growth. According to Omdia, the global semiconductor market is projected to exceed the $1 trillion mark by 2026.

Simultaneously, the global semiconductor landscape is transitioning from GPU computing to GPU-accelerated computing. Amidst this shift, GPU chip companies are garnering significant capital interest. Among them, Moore Threads, MetaX, and Biren Technology, as domestic GPU chip producers, have successively entered the capital markets, attracting considerable attention.

Among these, Moore Threads has emerged as a standout. Dubbed the first domestic GPU stock, it soared to a peak market cap exceeding 440 billion yuan, fueled by capital inflows, and became a highly coveted AI chip company.

However, amidst the excitement over capital's influence, concerns about a potential bubble have surfaced. Regrettably, compared to its peak market cap of 442.3 billion yuan, Moore Threads' valuation has diminished by approximately 176 billion yuan, currently hovering at 265.7 billion yuan.

Firstly, its revenue scale appears insufficient to sustain such a lofty market cap, and the company continues to endure substantial losses.

Revenue for 2025 is anticipated to range between 1.45 billion and 1.52 billion yuan, marking a year-over-year surge of 230% to 246%. Losses are projected to be between 950 million and 1.06 billion yuan, with a narrowing loss margin of 34.5% to 41.3%.

From a revenue standpoint, in the era of generative AI, Moore Threads is experiencing rapid expansion. Its compound annual growth rate from 2022 to 2024 surpasses 200%, with revenue skyrocketing from 46 million yuan in 2022 to an estimated 1.5 billion yuan in 2025. Over three years, its revenue has multiplied 32-fold, underscoring the vast potential of commercializing AI large model technologies.

However, the chip industry is inherently capital-intensive, necessitating significant and sustained R&D investment to maintain market competitiveness. Moore Threads' cumulative R&D expenditure from 2022 to 2024 amounted to 3.81 billion yuan, accounting for roughly 626% of its cumulative revenue over the same period.

Owing to the substantial R&D investment in chips, revenue falls short of covering R&D costs, resulting in significant and prolonged losses. From 2022 to 2025, Moore Threads' cumulative losses reached 6.22 billion yuan.

Despite the thriving AI large model sector, capital speculation has inflated the market caps of companies with mere billion-yuan revenues and substantial losses to hundreds of billions of yuan, inevitably fueling bubble formation.

Secondly, Moore Threads confronts numerous challenges, including technology iteration, software-hardware ecosystems, and international competition, which cast doubt on its lofty valuation and are likely to induce severe market volatility.

Established in 2020, Moore Threads is a GPU chip company that faces competitive disadvantages compared to industry behemoths like NVIDIA in the GPU arena.

In particular, NVIDIA commands the AI infrastructure market, with an estimated market share exceeding 80% and potentially surpassing 90% in the AI training chip market. In the global AI chip market, NVIDIA enjoys a near-monopolistic position, primarily due to its CUDA ecosystem, which offers a comprehensive development toolchain.

To bolster its competitiveness in the GPU market, Moore Threads secured 8 billion yuan through a listing on the Science and Technology Innovation Board. The funds will be allocated to R&D projects, including AI training and inference chips, graphics chips, AI SoC chips, and supplementary working capital. It also unveiled a new-generation GPU architecture, Huagang, which enhances computing density by 50% and energy efficiency tenfold.

Concurrently, based on the Huagang architecture, it introduced two chips, "Huashan" and "Lushan," tailored for computing power and graphics rendering scenarios, respectively. As a GPU chip company established in 2020, achieving a diversified computing acceleration product matrix encompassing AI intelligent computing, high-performance computing, graphics rendering, computing virtualization, intelligent media, and applications for personal entertainment and productivity tools within just five years is no mean feat.

Finally, amidst the wave of domestic substitution, local chip manufacturers like Moore Threads are entering a golden era of development.

In recent years, efforts to restrict domestic companies from procuring high-performance NVIDIA chips have intensified, aiming to stifle China's AI industry development. This external pressure has further expedited the substitution process.

Despite grappling with challenges such as technology iteration and software-hardware ecosystems, local chip companies have amassed substantial capital through the market to continually invest in core technology R&D and ecosystem development. The collective acceleration of technology iteration among local AI chip companies enhances their competitiveness against international chip giants.

Moreover, the market harbors high expectations for local chip companies. In under six months, four local AI chip companies—Cambricon, Moore Threads, MetaX, and Biren Technology—raised over 20 billion yuan in total, signaling that China's semiconductor industry is poised for leapfrog development through a dual-drive strategy of "capital + technology."

With the capital raised from the market and sustained R&D investment, the penetration rate of local AI chips continues to climb. Cambricon, in particular, has demonstrated outstanding performance, achieving substantial revenue growth and transitioning from prolonged losses to profitability. Its full-year revenue for 2025 is projected to range between 6 billion and 7 billion yuan, representing a year-over-year increase of 410% to 496%. Net profit is expected to be between 1.85 billion and 2.15 billion yuan, compared to a loss of 452 million yuan the previous year, marking a successful turnaround.

In the era of generative AI, the robust demand for AI chips, coupled with the trend of domestic substitution aimed at breaking free from reliance on high-end chips, has generated high market expectations for technological breakthroughs and commercialization of local chips.

Against the backdrop of sustained enthusiasm for AI chips, the long-term trend of substitution is irreversible. While reaping the benefits of substitution, it is imperative to rationally assess long-term development potential. It is also hoped that local manufacturers will bolster their technological capabilities while enhancing software-hardware ecosystems to support the growth of domestic GPU chips.

Yang Jianyong is a contributor to Forbes China, and the views expressed here are his own. He is committed to in-depth analysis of cutting-edge technologies such as AI large models, artificial intelligence, the Internet of Things, cloud computing, and smart hardware.