Oracle drops $7 billion Tencent AI chip lease deal, FT reports
For the sanctions officer on a corporate-banking credit desk
Oracle has abandoned a proposed $7 billion, five-year deal to lease Tencent up to 100,000 advanced AI chips, the Financial Times reported, as cited by Reuters on Nov. 4.1
The FT attributed the collapse to pressure from the Trump administration, and the deal ended before it was signed; US officials raised concerns about the arrangement, according to the report.1 The chips would have been hosted in Southeast Asian data centres, because US export controls bar direct shipment of such chips to China; Chinese officials had also urged Tencent to support domestic chipmakers.1
This account comes from a secondhand source, a Chip Dispatch article relaying the FT report via Reuters; neither Oracle nor Tencent is quoted, and the article does not say whether either company has confirmed the cancellation.1
Nvidia has reportedly told cloud service providers that at least 51 percent of their services must go to non-Chinese customers; the article does not name the source of that report.1
ByteDance and Alibaba rank among the largest Chinese clients for computing infrastructure in Southeast Asia, according to the article.1
The article says Tencent has increased its infrastructure spending as it develops its Hunyuan AI models, which the Oracle deal was meant to support.1
For banks and fintechs that lend to, invest in or buy from data-centre and cloud operators, offshore location no longer settles export-control risk, and the customer base now matters. Tenant lists, including any Chinese customers, belong in credit, compliance and vendor due diligence.
Our take. The data centres were in Southeast Asia; the customer was in China, and Washington judged by the customer. A loan repaid from lease income paid by Chinese tenants can lose that income before the lease is even signed.
Action. Credit risk and sanctions: Flag data-centre loans whose lease income depends on Chinese tenants; test them against a 51% non-Chinese customer threshold.
What our records say
NVIDIA designs GPUs and AI systems and sells them with a licensed software stack, and banks run the stack in their own data centres or rent it in the cloud. BNY said in March 2024 that it was the first major bank to deploy an NVIDIA DGX SuperPOD with DGX H100 systems, used for deposit forecasting, payment automation, predictive trade analytics and end-of-day cash balances. American Express trains LSTM fraud models on NVIDIA DGX, optimises them with TensorRT and serves them on Triton Inference Server within a two-millisecond latency budget. Capital One's agent servicing tool for thousands of customer-service agents uses the NVIDIA enterprise AI ecosystem, and its multi-agent workflows run on an NVIDIA inference stack with Triton and TensorRT-LLM. PNC is buying Nvidia chips for its own data centres as part of what CEO Bill Demchak calls an 'AI factory' (June 2026), to rely less on usage-priced external models. Santander said on 19 October 2026 that its AI Lab specialises models with NVIDIA NeMo Customizer, uses NVIDIA Kumo models for structured data and serves models with NVIDIA Dynamo-Triton; the two set a three-year collaboration framework. Deutsche Bank announced a partnership and said it plans to use NVIDIA AI Enterprise for risk-model development and high-performance computing, on-premises and on Google Cloud (Dec 7, 2022). Starling is building a proprietary transaction foundation model with Nvidia's financial-services team on data from over 15 billion anonymised transactions, trained and hosted on a UK sovereign cloud run by Civo, which is due in production in early 2027 for fraud detection.12
From our records
- NVIDIAvendor · banks using it (6) BNY, Amex, Capital One · alternatives SambaNova
Sources
- ReportOracle Cancels $7B Tencent AI Chip Deal — Chip Dispatch Chipdispatch