The landscape of intelligent computing, known as "Intelligent Computing" in Chinese, is undergoing dramatic changes in 2025, spurred by the advent of DeepSeek, a powerful computational modelAs industries grapple with the ramifications of this breakthrough, a flurry of interest from both consumers and businesses has emerged, signaling a resurgence of demand for computing power.

Initially, when DeepSeek was introduced, there were skeptics who feared that its optimization capabilities might reduce the need for computational resourcesHowever, after several weeks of market observation, a notable uptick in the demand for computational power has become evidentThis surge was highlighted by Liu Jun, Senior Vice President of Inspur Information, during the launch of the "2025 China Artificial Intelligence Computing Power Development Assessment Report." He noted that inquiries regarding AI servers capable of running the comprehensive 671B model of DeepSeek have skyrocketed in just the past two weeks, demonstrating the rapid market shift.

The collaborative report by IDC and Inspur Information dives deep into how DeepSeek is reshaping the intelligent computing market and outlines the latest trendsThe report paints a picture of evolution, where DeepSeek acts as a catalyst, revitalizing a market that is now embracing the new possibilities presented by this innovative technology.

What is remarkable is the level of excitement permeating all user groupsFrom grandparents to children, individuals are becoming increasingly aware of DeepSeek's capabilitiesEnterprises and governmental organizations are intensifying their exploration of applications, with fresh announcements of incorporating DeepSeek into their operations appearing dailyThe enthusiasm translates into a rapid escalation of computational demands.

In the immediate aftermath of the Lunar New Year, chip manufacturers worldwide have accelerated their adaptation processesIndustry insiders predict that optimizing inference capabilities will be prioritized, while adjustments for training processes may take longer to implement

Advertisements

The increased inquiries and purchase orders being received by server manufacturers underline this growing momentum.

Looking towards a more extended horizon, numerous experts predict that the DeepSeek wave will instigate significant transformations across the pre-training, fine-tuning, and inference segments of the intelligent computing marketFor example, there were whispers of the Scaling Law approaching obsolescence last year, prompting some major players to shy away from pre-trainingHowever, fueled by the success narrative of DeepSeek, confidence among these players is rebounding, and many are considering returning to the battlefield.

As IDC’s China Vice President, Zhou Zhenggang, articulated, if DeepSeek can deliver a model on par with what other firms need ten times the resources to achieve, it creates an itch for these companies to explore the potential of the DeepSeek framework on their grander projectsThis sentiment inspires every participant in the landscape of large models.

In recent announcements, OpenAI’s CEO, Sam Altman, revealed plans to unveil GPT-5, a model consolidating a wealth of OpenAI's technological advancementsShortly thereafter, tech mogul Elon Musk introduced Grok 3, marking another leap in the ongoing arms race for cutting-edge AI models.

On the fine-tuning front, the efficiencies introduced by DeepSeek have also fortified this market sectorReports indicate the influence of Scaling Law is now extending into fine-tuning and inference phases, allowing advanced algorithms such as reinforcement learning and chain-of-thought reasoning to incorporate substantial computational investmentsSuch developments can dramatically enhance the cognitive capabilities of large models.

Platforms like Hugging Face are already experiencing a flood of new versions that are being fine-tuned and distilled based on DeepSeek architecturesExperts believe this will significantly propel the entire intelligent computing landscape forward.

On the inference side, the industry views this sector as brimming with opportunity

Advertisements

One analyst likened DeepSeek to the "Watt moment" in history when James Watt’s improvements to the steam engine enabled it to become a stable source of power that could penetrate various industriesIn this context, large models emerge as the modern steam engine, ready to be adapted across all sectors.

Moreover, DeepSeek has ignited enthusiasm among enterprise clients for deploying large models within their operationsFollowing their trials, clients are contemplating how to scale deployments more broadly, indicating that the upcoming rounds of procurement will likely outpace current demand levels.

The report encapsulates this underlining trend, suggesting that, akin to Jevons’ Paradox, the efficiency gains brought about by DeepSeek haven’t stunted resource demands; rather, they've accelerated the adoption of large models and relevant applicationsAs more users and applications surface, the landscape of innovation is being redefined, leading to an upsurge in requirements for data centers and edge computing infrastructure.

Market forecasts are optimistic: by 2024, China's AI computing power market is expected to reach $19 billion, skyrocketing to $25.9 billion by 2025, which equates to a staggering annual growth rate of 36.2%. Looking further ahead, by 2028, the market could exceed $55.2 billion.

The intelligent services sector is also projected to experience rapid expansion, with an anticipated market size of $5 billion by 2024, potentially ballooning to $26.7 billion by 2028 at a compound annual growth rate of 57.3%. Within this growth, the integrated intelligent services market and GenAI IaaS (Infrastructure as a Service) will emerge as key drivers, expected to achieve respective growth rates of 73% and 79.8% over the next five years.

Transitioning from mere scalability to efficiency optimization is imperativeSolving the issues of high-performance computing scarcity and low power utilization necessitates more than just increasing capacity; it requires a paradigm shift toward enhancing operational efficiency.

The industry has widely understood that boosting capacity—via expanded computing resources—is crucial

Advertisements

In 2023, a wave of interest in intelligent computing centers led to the initiation of significant projects, with over 460 related tenders documented for 2024 alone, where contract values for major projects exceeded $1 million.

Overall projections indicate a robust compound growth rate of 46.2% and 18.8% for intelligent and general computing sectors from 2023 to 2028, significantly bettering previous expectations.

While enhancing capacity is a priority, efficiency gains are vital as wellLowering operational costs reduces energy consumption, becoming increasingly critical for the implementation and sustainability of large modelsHence, applying an application-oriented approach to AI infrastructure planning can help mitigate waste, encouraging efficient deployment strategies tailored to various sectors.

District-specific resource planning is now becoming standard practice for many intelligent computing centers, adapting to local industry needs—whether that pertains to manufacturing, robotics, or drone technologyThe ongoing rollout of DeepSeek across various centers, from Henan Airport Center to Wuxi Taihu, reflects a promising uptick in opportunities spurred by applications unlocking new capabilities.

However, as Liu Jun cautioned, merely deploying DeepSeek's API won’t sufficeCompanies must synergize these AI technologies with their operational datasetsSuccessful implementation hinges upon targeted, personalized optimizations that enhance user interaction and satisfaction levels—a necessity, especially where response times can make or break operational success.

Additionally, improving the efficiency of existing models, as demonstrated by DeepSeek's integration of FP8 and the MoE architecture, exemplifies how innovative approaches can elevate performance while reducing computing overheadsDifferent organizations renowned for their endeavors in advanced architecture are pivoting towards methods for efficient resource deployment.

With the current trend revealing a shift away from sheer volume, organizations are beginning to prioritize efficiency within their systems

Liu Jun's reflections highlight an industry-wide evolution aimed at creating more efficient systems to leverage large models more effectively.

Furthermore, optimization of computational infrastructure is paramountEnhanced architectures improve node-level performance, decrease data transit delays, and facilitate intelligent task distribution and management across clusters to ensure optimal resource utilization.

To stimulate substantial model production integration, establishing high-quality datasets is vitalA cohesive data architecture promotes streamlined access and sharing practices, bolstering AI model effectiveness during training engagements.

The report reveals that within the upcoming 18 months, alongside the infrastructure needed for hardware upgrades, increased investments in software and services are expectedAs user demand surges for tailored solutions, companies will need to adapt accordingly in response to DeepSeek's integration.

Looking ahead, as Liu Jun emphasizes, both central computing and private cloud sectors are witnessing escalating competition as major firms rally around supporting DeepSeek's expansive capabilitiesCorporations are increasingly interested in deploying small-scale intelligent computing centers to house between 1 and 20 servers, laying a foundation for future growth that aligns with their business model needs.

Ultimately, while the rise of inference demands is palpable—with projections suggesting a workload share of 67% in 2025—the focus on user experiences and cost-effectiveness will dictate the survivability of organizations within this growing inferential landscape.

As both inference and training markets navigate an environment steeped in potential, the intelligent computing sphere stands on the precipice of unprecedented growthEmbracing enthusiasm without recklessness is the key behavioral tenet guiding the industry toward the upcoming wave of technology integration.