DataCanvas Debuts Global AI Computing Paradigm at GITEX Global 2025

GITEX Global 2025, one of the world’s top three IT expos, recently kicked off in Dubai, UAE. The event gathered over 6,500 enterprises and 200,000 professionals from 180+ countries. DataCanvas, an invited participant, shared regional market insights and showcased its forward-looking global AI infrastructure strategy and achievements, standing out as a highlight at the Middle East’s tech showcase with its “new paradigm” for global intelligent computing.

Thousands of years ago, Arab manuscripts noted the “Seven Seas” journey to China—an imprint of ancient East-West cultural exchanges. Today, amid the global AI boom, exporting China’s advanced AI infrastructure expertise to empower global partners has become key to addressing worldwide computing demand and fostering industrial collaboration.

Dr. Xu Jiang, DataCanvas’s Head of International Business, addressed the GITEX forum on the AI computing market’s split between “super giants” and “agile innovators.” He pinpointed the pain point for SMEs and developers: unmet elastic computing needs for 10-100 GPUs. The dialogue on resolving supply-demand mismatches and achieving true computing inclusivity offered fresh solutions to global AI innovators’ bottlenecks.

Middle Eastern nations see AI as core to post-oil economic transformation. The UAE, for example, targets AI contributing 20% of non-oil GDP by 2031—yet its computing supply remains stuck in the pre-cloud era. “This imbalance limits resource flexibility and raises barriers,” Xu explained. “Most players—SMEs and researchers needing 10-100 GPUs—are shut out by costly rigid bare-metal services, while standard cloud instances fail large-scale training. We must break this bottleneck to avoid stifling AI ecosystem vitality.”

To tackle the clash between surging AI demand and structural supply gaps, DataCanvas pioneered a software-defined data center solution. It converts GPU power into divisible digital assets via resource pooling, paired with its innovative “One Data Computing Unit” metering to enable atomic-level scheduling.

This cuts total costs by over 60% and builds a Computing-as-a-Service infrastructure: SMEs access computing like utilities, research teams deploy training environments in minutes, and manufacturers rapidly roll out AI quality-inspection algorithms. Computing power transcends organizational size to become an accessible innovation driver.

Beyond regional needs, DataCanvas eyes global expansion. Leveraging its AI infrastructure engineering expertise, it offers tailored support: end-to-end smart computing center services for emerging markets, lightweight solutions for mature ones, and foundational support for global AI players. This cross-border, cross-economy inclusive computing practice is redefining global development.

As AI shifts toward “active decision-making,” Reinforcement Learning (RL) is moving from lab to industry—2025 is hailed as its “scaling year” for real-economy penetration. RL now delivers value in high-complexity scenarios: optimizing oil refining, aiding nuclear safety decisions, and accelerating drug development via protein design—far beyond gaming.

These breakthroughs rely on advancing world modeling, multimodal collaboration, and training paradigms. Building intelligent computing platforms for these needs has become critical for countries to lead AI tech and industry.

At GITEX’s ZTalk, DataCanvas’s Chief AI Scientist Miao Xu unveiled AgentiCTRL—the world’s first industrial-grade RL cloud platform—and its agent solutions. “In AI’s post-training era of multi-model collaboration,” Miao noted, “RL-agent integration demands more than traditional cloud platforms: high-frequency data interaction, real-time feedback for ‘perception-decision-action’ loops, and stable elastic scheduling for multi-agent training. Mainstream clouds, designed for static inference, can’t handle RL’s dynamic, resource-heavy nature—blocking large-scale adoption.”

AgentiCTRL breaks this bottleneck as the first 10,000-card heterogeneous computing RL platform. Its MoE architecture enables full agent training-inference with minimal code, seamless model deployment, boosted large-model inference, and fast agent iteration. Versus traditional RL, it cuts costs by 60% and boosts end-to-end efficiency by 500%, solving "high costs, long cycles" in agent development. 

Miao stressed: “Advanced RL drives agent AI adoption. Our ‘platform + applications’ model makes this globally inclusive—AgentiCTRL as the computing base, and ‘XinQiye’ translating RL into ready-to-use agent tools for businesses worldwide, ensuring equal tech access for all.” 

From ancient “Seven Seas” exchanges to today’s Middle East foray via GITEX, DataCanvas uses inclusive computing to meet global AI needs—unlocking greater value from China’s innovative computing paradigm.