Token Wave Reshapes China’s Internet Landscape

Two developments last week crystallized a notable shift in China’s internet sector. On April 2, ByteDance announced that its average daily Token calls had doubled over the prior three months to reach 120 trillion. On the same day, Goldman Sachs published its quarterly review of China’s internet industry and quietly reordered its sub‑sector preferences: cloud and data centers retained the top position, e‑commerce and freight platforms moved up to second, while gaming and entertainment slipped down the ranking.

Viewed separately, the two items might appear unrelated; taken together, they point to the same structural change. Over the past two years, the underlying competitive logic of China’s internet ecosystem has been rewritten. Token consumption has emerged as the new proxy for user engagement, and for business‑to‑business AI competition Token usage is becoming the decisive metric. Goldman Sachs’ tracking shows China’s aggregate daily Token consumption rose from 0.12 trillion in May 2024 to 140 trillion in March 2026, an increase of more than 1,000‑fold in under two years. ByteDance alone accounts for roughly 100 trillion of that total, while the remainder is shared among other players. Alibaba’s MaaS platform Bailian recorded a six‑fold increase in Token calls over the same period, and Goldman Sachs projects Alibaba Cloud’s revenue growth to accelerate from 36% in the quarter ending December 2025 to 40% in the quarter ending March 2026.

A critical but easily overlooked implication is that Agent‑era Token consumption is driven primarily by enterprise automation workflows rather than individual conversational sessions. Continuous customer‑service agents, automated code review pipelines and other B‑side processes consume Tokens at magnitudes tens of times greater than single user queries. Commercial traction is already visible: MiniMax reported ARR of USD 150 million in February, and Zhipu AI’s ARR reached USD 250 million by the end of March, a six‑fold increase since the start of the year, indicating that this commercialization pathway is materializing.

For cloud providers, demand has become highly deterministic because many enterprise AI workloads are essential rather than discretionary. This dynamic has inverted previous cloud economics: where price competition once dominated, leading providers are now able to lead on pricing, particularly for small and medium‑sized enterprise customers. Goldman Sachs notes that recent price increases have been driven by leading cloud vendors and concentrated at the SME end. Chinese cloud capital expenditure as a share of operating cash flow is about 58%, compared with an 89% average among U.S. peers, suggesting Chinese vendors retain financial flexibility to continue investing. Goldman Sachs forecasts Alibaba’s fiscal‑2027 capital expenditure to rise about 34% year‑on‑year to roughly RMB 180 billion, and Tencent’s fiscal‑2026 capital expenditure to increase about 25% to roughly RMB 100 billion.

However, sustaining Alibaba Cloud’s above‑40% revenue growth over the next five years would, by Goldman Sachs’ estimates, require annual capital spending that ultimately exceeds one year’s profit from Taobao and Tmall (approximately RMB 190 billion). In effect, e‑commerce profits would be largely redirected to support cloud expansion, creating a material funding pressure if losses from instant retail and food‑delivery businesses do not narrow as expected. This funding tension is the principal near‑term constraint weighing on Alibaba’s valuation.

The food‑delivery price war illustrates how these strategic pressures manifest in market competition. Goldman Sachs estimates the industry posted losses of RMB 70 billion and RMB 48 billion in the third and fourth quarters of 2025 respectively, against a pre‑war annual profit pool of about RMB 30 billion. In early April, the State Administration for Market Regulation circulated commentary interpreted as signaling regulatory intervention to curb excessive competition, analogous to the 2021 measures that constrained destructive price wars in the courier sector. Meituan’s market share has compressed from roughly 75–80% to about 50–55%; despite this, Meituan retains superior unit economics relative to Ele.me and, with regulatory support, its path to profit recovery appears clearer, albeit with long‑run per‑order profit expectations at roughly half 2024 levels.

At the model layer, the competitive picture remains unsettled. OpenRouter’s rankings show nine Chinese models among the top fifteen, with Xiaomi’s MiMo‑V2‑Pro leading and MiniMax, DeepSeek, Kimi and Zhipu also present. Price differentials are stark: Chinese models commonly charge between USD 0.4 and USD 3.6 per million Tokens, while Anthropic and OpenAI list prices near USD 25 and USD 15 respectively. Yet developer‑oriented price comparisons do not directly translate into enterprise procurement decisions. Zhipu AI’s model revenues, for example, are dominated by To‑G and state‑owned enterprise deployments, where pricing logic diverges from API price competition. More fundamentally, the Agent era raises questions about the depth of model‑level moats: distillation techniques and open‑source proliferation are narrowing performance gaps, and investors are increasingly uncertain about how high the entry barriers truly are. ByteDance, Alibaba and Tencent have each reorganized and incentivized AI teams to retain talent, reflecting the urgency of securing core capabilities.

Pinduoduo presents a particularly complex valuation case. The stock trades at about nine times expected 2026 earnings, below the sector median of 14 times, while net cash on the balance sheet stands near USD 70 billion (approximately USD 60 billion excluding restricted cash), close to half the company’s market capitalization. Goldman Sachs projects Temu’s 2026 GMV could exceed USD 100 billion, which, if realized, would position Temu at the scale of a mid‑sized listed cross‑border e‑commerce platform; current market pricing, however, assigns essentially no value to that international business. Bulls point to Temu’s local‑fulfillment transition in the U.S. and Europe as evidence of upside, while bears cite tariff risk, management turnover and limited segment disclosure as reasons for caution.

Goldman Sachs frames 2026 as a strategic inflection point for China’s internet giants: accelerated consumer AI investment, competition for AI super‑entry points, and simultaneous defense of core market positions. These concurrent priorities will strain resources. ByteDance stands out as the most formidable competitor, advancing across model development, content traffic, local services, e‑commerce and cloud. Over the next three to six months, three developments merit close attention: the outcome and regulatory boundaries of food‑delivery antitrust reviews, whether Alibaba Cloud and Tencent Cloud can sustain accelerated AI revenue growth in Q1, and Temu’s GMV and profitability under evolving tariff policies. Token consumption is the observable outcome, compute investment is the driving cause, and the ability of business models to monetize these shifts will determine long‑term winners. The competitive landscape has not yet produced a definitive victor, but the field is narrowing rapidly.