The new journey of large models has already begun, moving from a tech carnival to value realization, and from野蛮生长 to rational prosperity. This is the inevitable choice of the industry, and even more so, the development trend of the times.
In May 2026, the global AI industry witnessed significant changes. ByteDance (字节跳动)'s Doubao launched a paid subscription service, and Elon Musk also announced the merger of xAI into SpaceX, completing a business restructuring. One represents domestic commercialization, the other an overseas structural adjustment; these two events echo each other, directly reflecting an industry trend: the phase of extensive expansion for large models has ended. The past development model relying on stacking parameters, burning capital, and competing for traffic is no longer sustainable, and the industry has officially entered a new stage of commercial implementation, computing power optimization, and compliance control.
1. Restructuring of overseas tracks: Stepping out of model competition and increasing investment in computing infrastructure
Recently, Musk confirmed that xAI has cancelled its independent operations, merged entirely into SpaceX, and been renamed SpaceXAI. This integration is a long-term strategic adjustment, not a temporary business change. At the time of its establishment, xAI focused on the R&D of general large models, following industry trends to participate in the parameter race. However, issues such as frequent personnel turnover, high R&D costs, and unclear paths to profitability have long existed. As general large model technology gradually hits a bottleneck, the returns from simply optimizing dialogue models continue to decrease, and the cost-effectiveness of continuing to delve deep into the general model track keeps declining. Therefore, Musk chose to adjust direction in a timely manner to avoid homogeneous competition.
After the restructuring, SpaceXAI is no longer focusing on the R&D of general-purpose chat models for ordinary users, but is instead pivoting to high-end computing infrastructure. Public information shows that the Colossus1 supercomputing cluster under SpaceX is equipped with 220,000 Nvidia GPUs. It has been exclusively leased to Anthropic, and the two parties have reached a long-term cooperation to jointly explore the integrated application of space computing power and artificial intelligence. At the same time, SpaceX continues to deploy space orbital data centers, building a proprietary computing system relying on the Starlink communication network.
In terms of strategic logic, current large model technology is trending towards homogenization; the experience improvement brought by parameter increases is limited, while computing power costs continue to rise. Compared to deeply cultivating the general model track, SpaceX (SpaceX) chose to transform into the computing power supply end, shifting from a model R&D enterprise to a computing power infrastructure service provider. This layout targets the most core computing power resources of the AI industry and also builds a solid underlying support for overseas high-end AI services.
2. Shift in domestic trends: Abandoning pure technical gimmicks, commercialization has become the main line of development.
In sync with the strategic adjustments of overseas enterprises, the development logic of the domestic AI industry is also undergoing a shift. In the past few years, a "parameter race" prevailed in the domestic large model industry, where companies blindly pursued ultra-large parameters and prioritized product demonstration effects while neglecting practical applications, leading to a continuous accumulation of industry bubbles. Now, the logic of industry competition has been completely rewritten, with financial strength, computing power reserves, and implementation capabilities becoming the core factors of competition. Among them, Doubao (豆包) attempting to launch paid services has become a landmark move in the commercialization process of domestic AI.
Previously, most domestic AI products long adopted a free model to exchange for user traffic, relying on capital infusion to maintain operations, resulting in insufficient stability in their business models. To break the industry's inherent development model, Doubao (豆包) adopts a tiered operation approach: basic functions such as simple chat and basic copywriting remain permanently free, while high-consumption functions like data analysis, HD generation, and professional office work are included in the paid system. This adjustment marks the official shift of domestic mass-market AI products from traffic acquisition to commercial monetization, providing a reference for the exploration of industry profit models.
High computing costs are a key factor forcing enterprises to undergo commercial transformation. Large-scale users continuously calling models generates huge computing expenses, which cannot be sustained for long by relying solely on capital subsidies. Only by closing the commercial loop and achieving positive cash flow can the continuous iteration and optimization of models be supported. The capital market is also sensing changes in the industry, with the investment focus gradually shifting from conceptual technology to implementation capabilities. Currently, Kimi has completed a new round of financing of 14 billion yuan, deeply cultivating commercial long-text services; DeepSeek saw its valuation break through 300 billion yuan in its first round of financing, focusing on high-performance open-source models. Capital continues to concentrate on high-quality, practical enterprises, and the speed of industry consolidation continues to accelerate.
3. Precise Regulatory Measures: Upholding the Safety Baseline, Empowering the Upgrading of the Real Economy
With the rapid iteration of the industry, the domestic regulatory system continues to improve, forming a pattern where compliance control runs parallel to industrial development. In terms of security, relevant departments are strengthening industry control to build a solid line of defense for technological security. Recently, the NDRC (National Development and Reform Commission) halted foreign capital's acquisition of the Manus project, and AI large models have been included in the key review category for foreign investment security. By controlling the outflow of core algorithms, training data, and high-end computing power, the aim is to avoid the risks of malicious capital mergers and technological monopolies, ensuring the self-control and autonomy of the local AI industry.
At the industry level, regulatory authorities have introduced multiple support policies to guide AI technology to take root in the real economy. Policies encourage the deep integration of AI with e-commerce, manufacturing, healthcare, government affairs, and other fields, discarding flashy but impractical technology demonstrations and promoting the intelligent transformation of industries. Domestic tech companies are adjusting to the trend by embedding AI technology into production chains, leveraging artificial intelligence to optimize processes, reduce costs, and improve production efficiency. Strict regulation combined with industry support not only rectifies industry chaos but also clarifies the direction for the high-quality development of the AI industry.
4. Industry Future Outlook: Two Major Trends Reshaping the Global AI Industry Landscape
(I) Transformation of Functional Attributes: From Interaction Tools to Productivity Carriers
In the future, large models will shed the label of being just entertainment chat tools, and the focus of application will shift from human-computer interaction to actual production. In the past, the industry prioritized optimizing dialogue fluency and focusing on user interaction experience; in the future, the core of industry competition will center on task execution capabilities, building digital employees suitable for enterprise use. AI agents can integrate into various links such as office work, production, and R&D, autonomously completing tasks like data processing, business analysis, and process execution. At the same time, embodied intelligence is accelerating its implementation, linking up with industrial equipment and laboratory instruments to complete practical work, driving AI to extend from the digital realm to physical production scenarios. Practicality will become the core standard for judging the value of models.
(II) Divergence of Business Models: Differentiated Development Paths Formed in Domestic and Overseas Markets
The global AI industry will gradually form differentiated business models. Overseas markets, relying on high-end computing resources and mature enterprise service systems, will take a closed-source, high-premium route, deeply cultivating fields such as finance and high-end technology, generating revenue through customized high-end services, and solidifying their own technological barriers.
The domestic AI industry relies on massive application scenarios, comprehensive computing infrastructure, and a vast manufacturing base to focus on cost-effective inclusive services. On one hand, it retains basic free features to cover ordinary consumers and micro, small, and medium-sized enterprises; on the other hand, it delves deep into vertical industries, customizes specialized industrial models, and empowers traditional industries through large-scale applications. Both development models are adapted to local market characteristics, with no distinction between superior and inferior, while deep cultivation of the real economy will become the core competitive advantage of the domestic AI industry.
Doubao (豆包) testing commercialization and xAI restructuring are a microcosm of the current global AI industry transformation. The era of barbaric development characterized by capital speculation and traffic supremacy has ended, and industry competition has escalated into a comprehensive contest of computing power, ecosystems, implementation capabilities, and commercial value. Computing power management, compliant operations, real economy empowerment, and stable profitability will become essential conditions for the survival and development of AI enterprises. The industry bubble is gradually fading away, artificial intelligence is returning to its technological essence, and the future will focus on serving the real economy, entering a long-term, steady, and rational development cycle.