Tilbake til Nyheter
EnterpriseAI Understanding orientering

Tencent sikrer 7 milliarder dollar leiekontrakt for 100 000 AI-brikker fra Oracle

Tencent har signert en femårig leieavtale for omtrent 100 000 avanserte AI-brikker fra Oracle, en avtale på 7 milliarder dollar som gir den kinesiske internettgiganten tilgang til maskinvare utestengt fra hjemmemarkedet.

4 min readRead the linked source
Source-provided image accompanying Tencent secures $7 billion lease for 100,000 AI chips from Oracle
KildereferanseKilde registrert
Utgiver
finance.biggo.com
Kilde lenke
finance.biggo.comhttps://finance.biggo.com/news/d760f401-8bde-41c5-aca6-97de15e96dd5
Kildetype
Koblet kilde – status for primærkilde er ikke etablert.
Også sitert

Historien sist revidert

KontekstForstå dette på 60 sekunder

Start her

Nøkkelord

Fundamentmodell
En stor ferdigtrent modell som kan tilpasses mange nedstrømsoppgaver.
Beregn
Behandlingsressursene som kreves for å trene og kjøre modeller, ofte målt i FLOPS- eller GPU-timer.
Test deg selvQuiz for forklaring av AI-modeller

Hva har endret seg siden publisering

  1. Først publisert
  2. The new report adds details about Oracle’s Project Jupiter force‑majeure claim, the immediate share price reaction, and Tencent’s recent preview of an image‑generation model, providing fresh context to the previously reported $7 billion lease for 100,000 AI chips.

Hva skjedde

Tencent Holdings signed a five‑year lease with Oracle to use data‑center capacity in Southeast Asia, granting access to about 100,000 advanced AI chips. The agreement is valued at roughly $7 billion, with an upfront payment of about 30 percent of the total.

According to a Financial Times report citing people familiar with the matter, Tencent will lease Oracle’s Southeast Asian data‑center capacity for a five‑year term, providing access to roughly 100,000 AI chips that are otherwise barred from sale in China. The deal is valued at about $7 billion, with an upfront payment estimated at 30 percent of the total value.

Oracle did not immediately comment, and the arrangement could not be independently verified at the time of reporting. The lease represents Tencent’s largest overseas lease transaction to date and underscores its aggressive push for to train and serve its AI models, including the Hunyuan and a newly previewed image‑generation system for professional creators.

The agreement comes as other Chinese tech giants such as ByteDance and Alibaba also rely heavily on overseas data‑center capacity to sidestep U.S. restrictions on advanced processors.

Kildedetaljer: finance.biggo.com ↗

Hvorfor det betyr noe

The lease gives Tencent a route to the most advanced AI processors, which are currently unavailable for direct purchase in China due to U.S. export controls. For Oracle, the contract diversifies its AI customer base beyond OpenAI and supports revenue growth amid a costly data‑center build‑out. The deal also highlights the growing geopolitical tension over AI , as regulators consider closing the “cloud‑access loophole” that lets Chinese firms obtain restricted chips via overseas leases.

Access to Nvidia‑class AI chips is critical for training large foundation models. By leasing rather than purchasing, Tencent can bypass direct export bans while still obtaining the power needed to keep pace with domestic rivals like DeepSeek and Alibaba.

For Oracle, the contract helps reduce dependence on a single anchor tenant (OpenAI) and provides a new revenue stream as the company expands its AI‑focused cloud services. Oracle shares rose more than 2.5 percent after the report, though the stock remains down year‑to‑date.

The deal spotlights a broader geopolitical issue: U.S. policymakers are debating legislation to close the cloud‑access loophole that enables Chinese firms to obtain restricted AI hardware abroad. The lease’s structure—leasing rather than owning—places it squarely in the regulatory gray area.

Interactive Mechanism

Interaktiv mekanisme: Hvordan det faktisk fungerer

Utforsk den underliggende teknologien bak denne utviklingen interaktivt.

Model Parameter Size:8B Parameters
VRAM Required5.5 GBGPU memory footprint
Target HardwareMacBook / Single GPUDeployment tier
Privacy100% Air-GappedLocal device capability
Core takeaway: Small, quantized models (3B–8B) now run directly inside smartphones and laptops with complete data privacy, while mammoth 400B+ models remain the domain of datacenter clusters.
Interaktiv konseptsjekk+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

Hva du skal se neste

Watch U.S. legislative efforts to tighten export‑control rules on cloud‑based AI , Oracle’s progress on its $165 billion Project Jupiter data‑center in New Mexico, and Tencent’s rollout of its new Hunyuan image‑generation model that could benefit from the leased compute.

U.S. legislative proposals aimed at tightening export‑control rules for cloud‑based AI could force Chinese firms to seek alternative strategies or limit the effectiveness of such leases.

Oracle’s Project Jupiter, a $165 billion AI data‑center campus in New Mexico, has recently invoked a force‑majeure clause due to delays. The outcome of that project will affect Oracle’s capacity to meet growing demand from clients like Tencent.

Tencent’s upcoming release of its Hunyuan image‑generation model may test the practical benefits of the leased , offering a tangible measure of how the additional chips improve model performance and service offerings.

Relaterte guider og quizer

AI-modeller forklartKIs fremtidKI-etikkTest det du vet – prøv en gratis AI-quizSlå opp et AI-begrep i ordlisten vårFølg AI-finansieringssporeren

Oppdateringer og rettelser

Denne kanoniske historien oppdateres på plass når utviklingshendelsen endres vesentlig. Nettadressen og den opprinnelige publiseringsdatoen endres aldri.

  • The new report adds details about Oracle’s Project Jupiter force‑majeure claim, the immediate share price reaction, and Tencent’s recent preview of an image‑generation model, providing fresh context to the previously reported $7 billion lease for 100,000 AI chips.
Se den offentlige korreksjonsloggen
Fant du dette nyttig?