An overseas development in the AI computing market is attracting significant attention from capital markets.

According to reports from Indonesian media outlet Indonesia Newswire, Nasdaq-listed Polibeli Group is deepening its collaboration with Amazon Web Services (AWS) to enhance computing infrastructure in Southeast Asia. The initiative involves setting up computing centers focused on large-scale model inference services, with potential clients including leading global AI firms like Anthropic. Earlier, Polibeli also announced plans to assess the feasibility of building a large-scale AI computing center in Thailand, with an expected power capacity of around 100MW. As AWS ramps up its investments in cloud computing and AI infrastructure across Southeast Asia, Polibeli is transitioning from digital supply chain operations into the global AI computing infrastructure arena.

This overseas business expansion has led the market to take a fresh look at the recent strategic moves by A-share listed company Xingyun Technology.

From Billion-Yuan Domestic Orders to Global Computing Expansion

In recent months, Xingyun Technology has successively announced several significant computing service and related business contracts. These include a 3.6 billion yuan server leasing agreement, a 322 million yuan SSD hardware sales contract, a 1.014 billion yuan computing service contract, and a 5.508 billion yuan computing service contract, bringing the total contract value to tens of billions of yuan. For a company that has recently undergone a business transformation, such a surge in large-scale orders naturally raises questions among observers regarding the origins of its clients, its computing resources, and its contract fulfillment capabilities.

However, Polibeli's progress in computing collaboration with Amazon Web Services (AWS) in Southeast Asia offers additional insights into Xingyun Technology's industrial resources.

Polibeli is viewed by capital markets as Xingyun Technology's "sister company" in international markets. From a capital and industrial standpoint, it carries forward Xingyun Technology's industrial investment strategy and the global resource deployment of the Xingyun Group. Led by its actual controller, Wang Wei, the Xingyun Group has long been deeply entrenched in the global digital supply chain, amassing a vast network of resources in industrial capital, server procurement, cross-border channels, and international partnerships.

Polibeli's partnership with Amazon Web Services (AWS) on computing infrastructure in Southeast Asia further underscores the Xingyun ecosystem's capacity to engage with international tech giants, integrate global computing resources, and facilitate overseas project execution.

In essence, the recent large-scale computing orders secured by Xingyun Technology may not be isolated incidents. They likely stem from an industrial network encompassing upstream hardware resources, midstream computing centers, and downstream clients for large-scale models.

Large-Scale Model Competition Shifts to "Inference Efficiency"

The AI industry's demand for computing power is gradually evolving from a focus on model training to a balance between training and inference.

As the user base for large-scale models grows, context lengths expand, and Agent applications become more prevalent, the demand for inference computing power is surging. Especially in scenarios involving long texts with millions of tokens, complex task planning, and multi-Agent collaboration, model operations impose higher requirements on memory capacity, data throughput, KV Cache management, and system latency.

The competition among computing service providers is also transitioning from merely acquiring GPUs to optimizing hardware and software collaboration.

Xingyun Technology's recent appointment of Tang Bo as Chief Scientist is a significant development. Tang Bo founded AlayaDB.AI, which has long focused on infrastructure for large-scale model inference. Its integrated offloading architecture for Attention and KV Cache primarily tackles issues related to memory usage, data transmission, and response latency during long-text inference.

Under current hardware conditions, synergistic optimization of storage, memory, networking, and computing resources can enhance GPU utilization and per-card token output efficiency, thereby reducing unit inference costs for large-scale model clients.

This aligns closely with the practical needs of leading domestic large-scale model companies. Market speculations about the identity of Xingyun Technology's computing order clients abound, with rumors often pointing to domestic tech giant ByteDance. However, relevant information should be based on official company announcements. Regardless of the final client, the sustainability of large-scale order fulfillment depends on two core capabilities: securing a stable supply of servers, chips, and data center resources, and reducing clients' computing costs through technical optimization.

From its current business deployment, Xingyun Technology is simultaneously strengthening both capabilities.

Xingyun Technology's Value Merits Reevaluation

Xingyun Technology's stock performance over the past six months has garnered considerable market attention. Controversies mainly revolve around its relatively short tenure in the cross-border computing sector, the large scale of its orders, and the incomplete disclosure of client information.

These questions require further validation through subsequent order delivery, revenue recognition, cash flow, and client payment situations.

However, Polibeli's computing collaboration with Amazon Web Services (AWS) in Southeast Asia suggests that the resource organization capabilities of Xingyun Technology's related industrial network may hold greater potential than previously assessed by the market.

Domestic computing contracts worth billions of yuan, overseas computing deployment targeting global industry chain players such as AWS and Anthropic, combined with the technical expertise of Tang Bo's team in long-text inference and KV Cache optimization, are gradually forming a closed loop across these three domains.

Xingyun Technology's transformation logic extends beyond merely purchasing servers and leasing computing power. Its greater potential lies in establishing a synergistic computing service system that integrates "global hardware and capital resource organization capabilities, infrastructure construction capabilities for large-scale model inference, and inference efficiency optimization capabilities."

The computing industry ultimately competes not just on the quantity of cards owned but on the ability to secure resources, find clients, improve efficiency, and complete deliveries.

From this perspective, Xingyun Technology's recent surge in large-scale orders may not be as abrupt as it seems. The industrial resources and technical capabilities behind the company are increasingly being unveiled to the market through expanding business leads.