Chinese AI chip company Iluvatar CoreX executes large-scale procurement of 50,000 GPUs, emerging as major domestic AI infrastructure player.

On June 24th, Doubao Professional Edition was officially launched. Days prior, ByteDance announced a plan to procure no fewer than 50,000 AI inference chips from Chinese GPU company Iluvatar CoreX to support Doubao's terminal inference needs.

This large-scale procurement announcement is still reverberating through the market. If realized, it would represent a major milestone in Iluvatar CoreX's business trajectory, evidenced by the company's sharply rising stock price.

In the Chinese capital markets, Iluvatar CoreX is less well-known than Moore Threads and MoonlightX, which went public more recently. Yet it was established earlier and stands as the first Chinese company to mass-produce both training and inference GPUs, and the first to achieve 7nm mass production. Following founder Li Yunpeng's departure, the company cycled through two additional CEOs before assembling a management team pairing a capital-oriented chairman with executives holding technical backgrounds from AMD and NVIDIA. After listing on the Hong Kong Stock Exchange, Iluvatar CoreX joined Moore Threads, MoonlightX, and Biren Technology as the "Four Giants of Domestic GPUs" in the secondary market.

Facing explosive downstream demand and the blockade on high-end chip imports, Iluvatar CoreX abandoned a head-to-head competition with NVIDIA in high-performance training chips. Instead, it strategically developed across three directions: training, inference, and edge computing. This proved highly successful. Winning this ByteDance order represents a watershed moment in the company's history and has opened substantial space for domestic chip substitution in an already competitive market.

The domestic GPU breakthrough arrived through the inference channel.

According to Reuters, ByteDance's large-scale procurement stems directly from the enormous computational demand generated by Doubao. The application simply has too many users. Doubao is the market's leading AI-native application, with monthly active users reaching 368 million in May—a year-on-year increase of 182 percent. As of March 2026, Doubao's daily Token call volume exceeded 120 trillion. Facing continuous geometric growth and high-frequency requests, Doubao urgently needs to expand its inference capacity.

If completed, Iluvatar CoreX would become ByteDance's third-largest domestic GPU supplier, after Huawei Ascend and Cambricon. Reports from the same period indicate ByteDance is also evaluating Baidu Kunlun chips, while Alibaba is in contact with Iluvatar CoreX and likely to become its next customer.

Inference presents a far easier challenge than training. It operates in the deployment stage following model training, primarily executing inference tasks. It demands high real-time performance, fast response times, and low power consumption while maintaining accuracy. Training has an endpoint; inference never stops as long as a product remains live. Inference demand will rise exponentially.

Since entering the GPU market, Iluvatar CoreX has pursued dual-track development in training and inference, producing the Iluvatar InferX series. In 2025, this series generated 339 million yuan in revenue—a year-on-year increase of 238 percent, double the growth rate of training chips. The Iluvatar InferX Gen1 and Gen1X, released in 2022, gained broad adoption. Gen2 shipped in the fourth quarter of 2025 and has entered small-batch delivery. Gen3 will begin mass production in the second half of 2026.

According to IDC statistics, Huawei Ascend shipped 812,000 units in 2025, ranking first in the domestic GPU market, followed by Alibaba T-Head and Cambricon. Iluvatar CoreX shipped 45,000 units. This ByteDance order alone exceeds all of Iluvatar CoreX's GPU shipments from the prior year. Its achievement represents not merely a scale jump but validates that the company has obtained "certification from large enterprises."

Guosen Securities calculates that Iluvatar CoreX's inference chips carried an average unit price of 11,700 yuan in 2025. Applied to this order, the company would receive nearly 600 million yuan from ByteDance—equivalent to 60 percent of its total annual revenue from last year.

As of March 2026, China's daily Token consumption exceeded 140 trillion, a more than 1,000-fold increase from early 2024. Following large-scale AGI deployment, inference demand is growing faster than training. Morgan Stanley's year-end report noted that China's AI chip industry is undergoing significant structural adjustment, with inference computing demand gradually replacing simple compute stacking as the core growth driver. Every call and conversation depends on underlying inference computing power.

Amid both surging market demand and industrial scaling, Iluvatar CoreX has entered another critical phase following its IPO.

The Chinese AI chip industry was still in its infancy in 2015. At that moment, NVIDIA had recently entered artificial intelligence. Li Yunpeng, then R&D director of Oracle's database department, recognized the potential of China's technology sector and resigned to return and launch a startup. He founded Iluvatar CoreX in Shanghai's Minhang district in December 2015—five years before Moore Threads and four years before Biren Technology.

The future of AGI remained uncertain. Li Yunpeng focused on edge-side AI inference chips and released the company's first AI chip, "Iluvatar CoreX I," in 2019, targeting visual intelligent algorithm applications. Less than a year later, investors urged the company toward the general GPU direction, while Li Yunpeng held firm to his vision and ultimately departed. Diao Shijing, former co-president of Tsinghua Unigroup and chairman of Unigroup Guowei, assumed leadership.

In 2021, Iluvatar CoreX launched "Tian Gai Gen 1," its first mass-produced general GPU, filling a gap in high-performance domestic AI training chips. After Diao Shijing's disappearance in July 2022, Gai Lujiang became the new leader, serving as chairman and CEO. That year, the company released the "Iluvatar InferX Gen 1 inference accelerator card," completing its dual product-line layout and becoming the first in China to construct a unified "training + inference" architecture.

Centurium Capital holds the largest institutional shareholder position, with approximately 22.92 percent ownership prior to the IPO. The firm's notable industry portfolio includes an investment in Luckin Coffee. Gai Lujiang, also a core investor in Shanghai Xishi, serves as another major shareholder. Unlike his predecessors, Gai Lujiang lacks chip technology background; his career has centered on finance and investment.

Technically, CTO Lü Jianping brings experience from world-class chip firms including Samsung, NVIDIA, MediaTek, and Intel, with deep expertise in general GPUs. Vice-presidents Sun Yile and Shi Jiasheng both come from AMD, leading chip R&D and mass production and software development respectively. Vice-president Song Yu previously worked at IBM.

Similar to its early investment in Luckin, Centurium Capital completed an early-stage financing of Iluvatar CoreX through affiliated enterprises in 2017. The firm has accompanied the company throughout, leading key funding rounds and supporting joint-stock system reform and board reorganization.

In a 2022 interview, Lü Jianping stated: "Iluvatar CoreX has always approached graphics from the perspective of general computing, rather than struggling to catch up from the traditional graphics pipeline. We will be the leader in this area."

Through strategic adjustments by successive leaders and a capable technical team, the company placed its bet on the challenging domestic general GPU market and achieved "three firsts": the first domestic mass-produced training GPU, the first mass-produced inference GPU, and the first to achieve 7nm mass production.

[The text appears to end mid-sentence and may be incomplete.]