Byte boosts again! Global AI investment continues to rise, computing power enters full-chain inflation.

According to reports, ByteDance is increasing its investment in artificial intelligence infrastructure, with the company planning to exceed 200 billion yuan in capital expenditures this year, an increase of 25% from the initial plan. The increase in capital expenditure is mainly due to two reasons: ByteDance's increasing investment in artificial intelligence, and the rise in memory chip costs. Several cloud giants are also increasing their capital expenditures along with ByteDance. TrendForce released a report on May 6 stating that due to recent upward revisions in capital expenditure guidance for 2026 by most major North American cloud service providers in response to strong AI demand, the estimated total capital expenditure for the nine major cloud vendors (Google, AWS, Meta, Microsoft, Oracle, ByteDance, Tencent, Alibaba, Baidu) is expected to be around 830 billion US dollars for the year, with a year-on-year growth rate increasing from the original 61% to 79%. Furthermore, many companies attribute the increase in capital expenditure to constraints and price pressures on hardware infrastructure: Microsoft stated that out of its 190 billion US dollars in capital expenditure, 25 billion US dollars are used to counter the rise in component prices; Google stated that its capital expenditure for the quarter was 35.7 billion US dollars, including real estate, servers, data centers, and other infrastructure; Meta attributed the increase in capital expenditure to the higher prices of hardware components this year and the additional data center costs needed to support future capacity. As AI commercialization enters the validation stage, some companies' AI businesses have begun to gradually monetize. Guolian Minsheng Securities pointed out that from the financial reports of overseas cloud giants, AI competition has shifted from simple technical and computational expansion to a heavy asset competition constrained by capital expenditure efficiency. This round of financial reports releases three key signals: First, AI commercialization is entering the validation stage, and the market is starting to demand quantifiable revenue indicators from companies, such as ARR, cloud business growth rate, backlog of orders, or advertising conversion efficiency, rather than just narrative-driven. Second, there is a clear differentiation in AI monetization paths, with different companies building commercial loops through cloud service expansion, computational and chip sales, software subscription embedding, and advertising system optimization based on their own endowments, but the industry has not yet formed a unified paradigm. Third, capital expenditure constraints are significantly strengthened, with a focus on investment return cycles, profit margin stability, and free cash flow performance under high Capex levels in the coming quarters. Domestic model ARR and Token volume have seen exponential growth, signaling the "golden age of domestic computational power." Just a few days ago, ByteDance's AI application Douyin launched a paid version service statement on the App Store, introducing three tiers of value-added services on top of the free version. The initiation of paid testing for Douyin has been praised by many securities firms as "a milestone," with Bohai Securities pointing out that this could become an industry benchmark, verifying the scarcity of computational resources and marking the initial commercialization of AI. As the largest AI chat application in terms of monthly active users in China, as of March 2026, Douyin has reached 345 million monthly active users. However, the expansion of the user base also means a rise in customer acquisition costs and the consumption of existing user tokens: according to data released by Volcano Engine in early April, Douyin's large model daily Token call volume exceeded 12 trillion in March this year, doubling in three months; compared to when it was first introduced to the public in May 2024, this has grown by 1000 times. At the current moment, domestic model factories have seen a significant increase in Token and ARR. Data shows that Minimax ARR has exceeded $150 million; KNOWLEDGE ATLASAPI ARR has surpassed $250 million, with demand still outstripping supply despite three price hikes. In the first quarter of 2026, Token call volume increased by 400%, signaling a rapid release of computational power to meet the expanding commercial demands. Sinolink pointed out that the golden age of domestic computational power has begun, where usable computational power itself is the most core asset in a supply-constrained environment. Under strong logic from both supply and demand sides, the computational power industry chain in 2026 has materially entered a phase of comprehensive inflation, with industry vitality shifting from core chips to AIDC, cloud, and computational service segments: CPUs/GPUs face capacity bottlenecks and continuous rental price increases, while major players like Alibaba and Baidu are also increasing pricing for their cloud computational resources and related services in multiple rounds. The continuous price hikes across multiple segments strongly confirm the explosive demand for real computational power and the extreme scarcity of supply. This article is reprinted from "Cai Lianshe," written by Zheng Yuanfang; GMTEight editor: Huang Xiaodong.