A growing number of AI giants are burning through cash as they ramp up investments in artificial intelligence.

As AI strategies accelerate, leading companies are making aggressive investments in hardware and computing capabilities to stay ahead.

On April 20, sources revealed that ByteDance’s net profit for 2025 plummeted by over 70% year-on-year, with profit margins taking a significant hit due to a sharp increase in AI investments during the third and fourth quarters of the previous year. Meanwhile, ByteDance’s overseas revenue surged nearly 50% in 2025, far outpacing domestic growth of around 20%. Overseas revenue now accounts for over 30% of the total—a new record. TikTok Shop’s GMV (Gross Merchandise Volume) soared nearly 70% year-on-year in 2025, driving the increase in ByteDance’s overseas revenue share.

The sources stated that to support the training and inference demands of its Doubao large language model and multimodal AI products, ByteDance sharply increased AI-related investments in the second half of 2025. This covered high-end AI chip procurement, foundational model R&D, and other core areas. These massive expenditures directly caused the full-year net profit to drop over 70% year-on-year. ByteDance plans to further escalate AI investments. During shareholder communications, the company disclosed that technical resource investments will expand further in 2026, along with compliance-related spending in overseas markets, keeping short-term profit margins under pressure. [Additional infrastructure investments will also be made to meet the computing demands of continuous large model iterations.]

How much has ByteDance invested in AI? Media reports in late 2025 stated that ByteDance planned to invest $23 billion (~RMB 160 billion) in AI infrastructure in 2026, up from RMB 150 billion in 2025. Of this RMB 160 billion, RMB 85 billion is earmarked for AI chip procurement, with the remaining RMB 75 billion allocated to data centers and supporting facilities. RMB 50 billion of the budget is designated for overseas computing investments.

ByteDance is not alone in betting big on AI—domestic tech firms are going all-in on AI development. From Lunar New Year marketing battles to year-round strategic priorities, giants like Alibaba and Tencent are prioritizing AI, even at the cost of short-term profit declines, to secure a foothold in the AI era.

In February 2025, Alibaba Group CEO Wu Yongming announced that Alibaba will invest over RMB 380 billion over three years to build cloud and AI hardware infrastructure, surpassing the total investment of the past decade. Alibaba is redirecting its core technical expertise, top talent, and e-commerce focus toward AI. On April 8, Alibaba established a Group-level Technical Committee led by Wu Yongming, comprising Alibaba’s top tech leaders: Zhou Jingren, Wu Zeming, and Li Feifei. Zhou Jingren was appointed Chief AI Architect, Li Feifei oversees Alibaba Cloud’s technology and AI cloud infrastructure, and Wu Zeming leads the Group’s business technology platform and AI inference platform. This centralizes technical resources previously dispersed across business lines, breaking down silos and aligning decision-making and resource allocation directly with AI goals. Additionally, Tongyi Lab was substantially upgraded to become the Tongyi Large Model Business Unit, fully overseen by Zhou Jingren. The rebranding reflects a strategic shift: from pure R&D to integrating R&D with direct commercialization, pushing large models to the frontlines of business and converting technology into revenue.

Tencent invested RMB 18 billion in AI product development last year, with over RMB 7 billion spent on new AI products in Q4 alone. This year’s investment will at least double. Tencent CSO James Mitchell previously noted that surging AI computing demand extends beyond GPUs. When users engage with AI agents, they effectively create and execute software, with most computations handled by CPUs and significant memory consumption. Thus, demand growth spans not just GPUs, DRAM, or HBM, but also CPUs, standard RAM, SSDs, hard drives, and other hardware, showing an overall upward trend. Since the 2023 release of Baidu’s ERNIE Bot, Baidu’s cumulative AI investments have exceeded RMB 100 billion.

Shortages: First GPUs, Then Memory, Now CPUs

Tech firms’ massive AI capital expenditures are directly boosting the semiconductor supply chain. Initially, GPUs were in short supply, followed by memory. Now, the focus has shifted to CPUs. TrendForce estimates that due to strong AI demand and NVIDIA’s push for high-chip-count integrated GB/VR cabinet solutions, NVIDIA’s high-end GPU shipments will grow significantly in 2026, with an estimated ~26% year-on-year increase.

Memory has entered a “super cycle,” with all product categories entering price hike phases. TrendForce’s latest survey indicates that in Q2 2026, DRAM manufacturers will aggressively shift capacity to HBM and server applications, adopting “catch-up pricing” strategies to narrow price gaps across products. Despite potential downside risks in end markets, overall conventional DRAM contract prices are expected to rise 58-63% quarter-on-quarter. The NAND Flash market remains driven by AI and data center demand, with across-the-board price hikes continuing. Overall contract prices are projected to increase 70-75% quarter-on-quarter in Q2.

When will memory price hikes end? SK Group Chairman Choi Tae-won publicly stated that the primary cause of this shortage is insufficient wafer supply, which will take at least 4-5 years to resolve. He expects memory supply shortages to persist until 2030, with a supply gap exceeding 20%.

With the explosive growth of Agentic AI, CPUs now face severe shortages. Semianalysis’ Dylan Patel notes that GPUs are no longer the bottleneck for cloud providers—that role has shifted to CPUs. Previously, AI GPUs handled only simple inference tasks, but new models have fundamentally changed task dynamics. Agentic AI is now widely used for database queries and CPU-intensive tasks like physical simulations and computational modeling. These frequent database accesses and CPU-heavy computations have caused CPU utilization rates in cloud data centers to spike sharply.

Capacity Expansion: Upstream Firms Increase CapEx, But Ramp-Up Takes Time

With such high demand, upstream manufacturers are actively expanding capacity. TSMC is advancing global capacity expansion plans, particularly for 3nm processes. At its latest earnings call, TSMC stated that Taiwan remains its core R&D and advanced process hub, prioritizing R&D retention in Taiwan to ensure tight integration with new process development. It will add a 3nm line in the Tainan Science Park, with mass production expected in H1 2027. Its second Arizona fab is complete and will introduce 3nm in H2 2027, while its second Japan fab is slated for 3nm mass production in 2028.

Samsung Electronics plans to invest over KRW 110 trillion in capital expenditures and R&D in 2026, aiming to triple AI chip capacity and quintuple HBM capacity. Its technology roadmap is clear: it is advancing 10nm fifth-gen 1b DRAM process conversions and new line expansions at its Hwaseong and Pyeongtaek sites in Korea, while accelerating HBM4 mass production to close the gap with SK Hynix in the HBM market.

SK Hynix is investing over KRW 1 trillion in its Jiangsu Wuxi DRAM fab and Dalian NAND flash fab, focusing on DRAM process node upgrades from 1Z to 1A and 321-layer ninth-gen NAND line conversions. Notably, the Wuxi fab contributes over 30% of its global DRAM output, while the Dalian fab accounts for 40-45% of NAND output.

Micron expects capital expenditures to exceed $25 billion in FY2026 and rise further in FY2027, focusing on HBM and advanced DRAM capacity expansion. Its strategic layout is clear: it is accelerating EUV equipment adoption across all lines to pave the way for 1β node DRAM and HBM4/5 production, while advancing fab construction in New York, Idaho, and Hiroshima, Japan, with total investments reaching $100 billion. Its new Singapore fab is also expected to start production in H2 2028, strengthening high-end NAND flash supply.

Gap: Compute Demand Growth ~3x NVIDIA's Projected CAGR for Compute Supply

Based on OpenRouter platform tracking, global weekly token usage surged from 6.4 trillion to 22.7 trillion from early January to March 2026, a ~250% increase in three months. The rapid adoption of agentic AI tools (e.g., OpenClaw) has dramatically accelerated demand-side growth. Multiple LLM providers have begun imposing token usage caps on users to manage runaway demand. Morgan Stanley forecasts that future compute demand growth will be ~3x NVIDIA's projected CAGR for compute supply, indicating that compute shortages will persist and intensify long-term.

Against this backdrop, Morgan Stanley believes any company that can break through compute scalability bottlenecks will see significant upside. This includes not just chip manufacturing supply chains but also memory, optical networking equipment, and data center core components. Morgan Stanley is highly bullish on these “Merchants of Compute,” expecting them to directly benefit from systemic supply-demand imbalances.

Beyond immense compute demand, the time required for capacity expansion also contributes to supply-demand gaps. Take TSMC: its new 3nm fabs in Taiwan, the U.S., and Japan will not start production until H2 2027 at the earliest.

Against this industry backdrop, global tech giants are embroiled in a high-stakes AI arms race, with hardware price hikes, compute shortages, and profit pressures becoming the norm. This also urges us to revisit AI's fundamentals. Looking back at AI's evolution, in 2016 Demis Hassabis and his company DeepMind defeated Lee Sedol with AlphaGo, then bested Ke Jie the following year, officially heralding the AI era. In 2022, OpenAI released ChatGPT, marking AI's inflection point. “Hassabis: The Brain Behind Google AI” not only chronicles the journey of this top global AI scientist but also reveals how DeepMind, through scientific rigor and steady execution, regained its position as a global AI leader beyond the speed-and-traffic-chasing model. For the global AI industry at this critical investment and expansion phase, this long-term R&D philosophy and science-driven development path offer vital reference points.