腾讯发行最长 30 年期债券，为 AI 资本开支补充资金

NextFin News — Tencent rarely looks short of cash. WeChat, games, advertising and payments still generate enormous operating inflows. At the end of 2025, liquid resources approached 500 billion yuan; net cash after borrowings and notes remained above 100 billion yuan and rose further early in 2026. Yet the company is again raising long-term debt.

On October 14 it priced four tranches totaling 1.8 billion dollars and 26 billion yuan. The dollar notes stretch to ten and twenty years; the renminbi notes include a thirty-year piece. Coupon rates sit higher in dollars than in yuan, reflecting different rate environments. Net proceeds are earmarked for general corporate purposes. The deal follows a sizable multi-currency issue only four months earlier and an inaugural renminbi note program in 2025.

The funding rhythm tracks a sharp rise in capital intensity. In 2023, capital expenditure under the company’s reporting stayed near 24 billion yuan. In 2024 it jumped more than threefold as generative AI raised the cost of competing. Spending remained elevated in 2025, at roughly a tenth of annual revenue. The first half of 2026 then exceeded the prior full year: quarterly outlays climbed into the tens of billions of yuan, and cash payments for IT infrastructure, servers and data centers nearly doubled year on year.

Booked capital expenditure and cash payments do not move in lockstep because of timing differences, but both series tell the same story. A business once defined by software, relationships and content is adding another core asset class: compute capacity that must be bought, powered and depreciated over years.

Operating cash generation has not disappeared. In 2025 free cash flow remained strongly positive even after heavy investment. The warning sign appeared in the second quarter of 2026. Operating cash inflows were outpaced by capital payments for hardware, content commitments and leases, leaving free cash flow negative for the period. Net cash on the balance sheet fell sharply in the same quarter, reflecting both infrastructure prepayments and the annual dividend.

Management has described the company’s portfolio in two layers. Mature franchises in messaging, games and advertising continue to produce cash. Newer AI efforts—foundation models, consumer and office applications, and the infrastructure that supports internal products and cloud customers—require large outlays before revenue scales. Training and inference both consume GPUs. Management has characterized the densest phase of capacity build as concentrated over the next couple of years and has said funding should draw on cash, the investment portfolio and prudent debt capacity rather than operating cash flow alone.

Long bonds fit that logic. Spreading a multi-year infrastructure wave across ten- to thirty-year liabilities reduces the chance that a single quarter’s hardware payments will force abrupt cuts elsewhere. Investor demand for the latest issue was strong, according to market reports, suggesting credit markets still view Tencent as a high-quality borrower. The open question is not access to capital. It is conversion: how much of today’s server and data-center spend will show up as durable AI-related revenue or as measurable lift in existing advertising and games businesses.

Some benefits are already visible in management commentary—improved ad targeting, faster content tools, stronger recommendations. The harder test sits with AI-native products that must still build distribution and prove unit economics while inference costs run in the background. Second-quarter profit growth slowed sharply under international reporting standards even as capital expenditure approached the scale of quarterly net income and cash infrastructure payments exceeded it.

Leadership has stressed that capacity will not expand without regard to returns. Foundational training spend is framed as a front-loaded necessity; incremental inference capacity is supposed to track demonstrated demand. That distinction matters. Unlike a single failed acquisition, underused AI hardware leaves a trail of depreciation for years. Tencent’s bond calendar now stretches deep into the 2050s. In practical terms, the company is financing early-stage AI infrastructure on a horizon far longer than any current product cycle—an expensive ticket to remain in the next round of platform competition.

更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App