Tencent has lost access to a large block of advanced AI computing capacity after Oracle pulled out of a planned chip leasing arrangement. The deal would have given the Chinese tech giant use of 100,000 high-end AI chips for five years.
Despite the setback, investors reacted calmly. Tencent's shares climbed in the days that followed, buoyed by the release of a new version of its in-house AI model, according to a weekly China roundup published by Growth Dragons.
A Workaround Closes Off
The chips were to be housed in Oracle data centers in Southeast Asia, with the lease valued at around US$7 billion. Renting computing power overseas has become one of the few remaining routes for Chinese companies to tap top-tier US chips, since export controls bar them from purchasing the hardware outright.
Citing the Financial Times, the roundup reported that US officials voiced concerns about the arrangement, after which Oracle backed out. The episode shows how cloud-based access to restricted hardware is increasingly drawing regulatory attention, not just direct chip sales.
Shares Rise on the Hy4 Launch
On the same day the lease news emerged, Tencent introduced Hy4, the latest iteration of its Hunyuan AI model. The stock gained 7% the following day and finished the week up 9%.
That rally came against a weak backdrop. Tencent's shares remain down roughly 30% for the year, suggesting investors are still cautious about how its AI spending will translate into returns.
What Hy4 Can and Cannot Do
Tencent says Hy4 can interpret charts and images, operate a computer autonomously and handle long, multi-step tasks. The company also claims it produces far fewer fabricated answers than the preview version released in August.
Tencent kept developer pricing for the model unchanged, a reflection of the intense price competition among Chinese AI providers. Its own published benchmarks, however, show Hy4 behind Anthropic's Claude on all 12 tests shown, and behind Moonshot's Kimi on six of them.
Leaning on Its Own Ecosystem
Tencent's main advantage lies in deploying AI across products it already controls. Employees have built thousands of small AI tools on its WorkBuddy platform to handle HR, finance and administrative tasks, cutting the time needed for chores such as reconciling payments against invoices.
The company can offer similar tools to businesses using WeCom, its workplace messaging app, and to international cloud clients. One recent example is OmniOps, a new partner in Saudi Arabia.
The Compute Question Remains
Losing the Oracle capacity leaves Tencent with less computing power for training future models, a constraint that matters as model development grows more hardware-intensive. Heavy AI investment is also expected to pressure profits before it pays off.
The newsletter's authors argued the lease cancellation may matter less than it appears, reasoning that an affordable, capable model embedded in WeChat and WeCom could be more valuable to Tencent than a market-leading one.
What It Signals for the AI Hardware Market
For the broader industry, the cancelled deal indicates that leasing overseas data center capacity is no longer a reliable path around US export rules for Chinese firms. Cloud providers may face closer scrutiny over who uses their most advanced chips.
For Chinese AI companies, the likely response is to squeeze more value from limited compute through efficient models and tight product integration. Tencent's approach offers an early test of whether that strategy can hold up as hardware access narrows.
Source: Growth Dragons
