Tencent Leases 100,000 AI Chips From Oracle as US Export Controls Shift Global Supply

The Financial Times reported on 1 October that Tencent had agreed to lease 100,000 AI accelerators from Oracle over five years for $7 billion. Startup Fortune carried the same figure the same day. Neither company has confirmed the arrangement publicly, and the dossier does not say which chips are involved or where they will sit.

The timing is what makes the deal notable. It lands in the same week that Washington and Beijing are still arguing over the scope of US semiconductor controls. It suggests Chinese buyers are willing to pay a premium for compute they cannot buy directly.

Controls Keep Moving

US export controls on advanced chips are administered by the Commerce Department's Bureau of Industry and Security under the Export Administration Regulations, according to a Kovara research explainer. The first rule came on 7 October 2022 as an interim final rule covering advanced chips, supercomputer items and chipmaking equipment. A chip falls under ECCN 3A090.a if its total processing performance reaches 4,800, or 1,600 with a performance density of 5.92 or more, the explainer says. Chips below those thresholds and not designed for data centres are carved out of parts of the controls.

The rules have been rewritten repeatedly since. October 2023 removed the interconnect parameter, added performance density, and extended licence requirements to more than 40 additional countries. December 2024 brought controls on high-bandwidth memory above 2 gigabytes per second per square millimetre. The 2025 AI Diffusion Rule was rescinded before its obligations took effect, and in January 2026 BIS moved to case-by-case review for China exports of chips up to roughly H200 level, Kovara writes.

That last shift matters because it followed a series of reversals. The Register reported on 30 September that the Trump administration first imposed licensing requirements on Nvidia's and AMD's China-spec accelerators in April 2025, then reversed course that summer and later cut the US government in on a percentage of certain China-bound chip sales. By late last year, Nvidia was allowed to sell more potent H200-series parts to approved Chinese customers for the first time.

Huawei Claims the Lead at Home

Huawei says the damage to Nvidia's position in China is already done. In a Q&A-style release this week, rotating chairman Eric Xu claimed its Ascend line of neural processing units had surpassed Nvidia in Chinese market share. "It's pretty hard to collect data about the market share of Nvidia in China, but based on the data we have collected, Ascend has surpassed Nvidia," Xu said, according to The Register.

Nvidia's own numbers give the claim some context. During its Q2 earnings call, executives said H200 shipments to the region amounted to less than 1 percent of datacenter revenues. Xu also downplayed Huawei's international ambitions, saying the company lacks capacity to meet Chinese demand and will supply only a limited number of overseas customers.

"Even though our chips may be less advanced, at least their supply is assured, so that you don't have to worry about chip supply day in and day out," Xu said.

The software gap that once protected Nvidia is narrowing too. Deepseek released open-source programming tools for Ascend chips, with TileLang at the centre, according to The Decoder. The Chinese developer says Huawei "fully supported" the work, and the two companies optimised a supernode cluster of 128 Ascend 950 chips. Separately, research firm SemiAnalysis tested OpenAI's Jalapeño inference chip and called the CUDA moat "potentially dead," though it cautioned that its tests covered relatively easy scenarios.

Enforcement Questions Linger

Not every story this month is about policy. On 24 September, Tom's Hardware reported that SemiAnalysis head Dylan Patel had called for an investigation into AMD after China's Puzhi Electronic Technology crowdfunded a system based on AMD's Zynq UltraScale+ XCZU47DR RFSoC, an export-controlled dual-use part. Patel wrote on X that the chip is quoted at $1,000 in China against a US list price of $36,000. AMD told Tom's Hardware it does not sell or ship the chips directly into restricted regions and investigates all suspected diversion cases, without confirming a specific investigation.

Meanwhile the diplomatic track stays stuck. The South China Morning Post reported on 27 September that export controls were not a major sticking point during Xi Jinping's state visit to Washington, with no direct mention in either side's readouts. Export Compliance Daily reported on 28 September that former officials and analysts see the renewed US-China trade truce as more fragile than either side admits, as Beijing keeps pushing for controls to be lifted and Chinese firms removed from restricted lists.

China's Ministry of Commerce has kept up the pressure in public. A Beijing Post summary notes that spokesperson He Yongqian said communication between the two sides has continued since the Geneva talks and called for an end to what China terms discriminatory restrictions. The same report cites a US directive telling EDA vendors including Cadence, Synopsys and Siemens EDA to halt software sales to Chinese entities, and notes share moves: Synopsys down 9.6 percent, Cadence down 10.7 percent, while Chinese EDA firms Primarius and Semitronix rose.

The Tencent-Oracle lease is not proof that controls are failing. It is evidence that the market is routing around them, and that the price of doing so is now measured in billions.