ByteDance Negotiates $500M Azure Deal for OpenAI Models
ByteDance, the owner of TikTok, is in advanced talks to spend about $500M more on Microsoft's Azure cloud, mainly to use OpenAI models, according to The Information, which cites two people familiar with the talks. OpenAI does not sell its models to most Chinese companies, but Microsoft does not apply that restriction when it resells OpenAI technology through Azure, the outlet reports. The deal reportedly irritates OpenAI. The practical takeaway: access rules for a model depend on which provider you buy it from.
AI-processed from TNW; edited by Hamidun News
The same model can be closed to a customer when bought directly but available to them through a partner's cloud. That is the story The Information describes: ByteDance, the owner of TikTok, is in advanced talks to spend roughly $500M more on Microsoft's Azure cloud. The money would go mainly toward using OpenAI models. Two people familiar with the negotiations told the outlet about this. TNW has not independently verified the information, so this is still a leak, not a confirmed deal.
The loophole around OpenAI's ban
OpenAI does not let most Chinese companies use its models. But Microsoft holds the rights to sell OpenAI technology through Azure and, according to The Information, does not apply the same ban. The scheme is simple: there is no direct access, but there is access through Microsoft's cloud.
The new contract would be added to ByteDance's existing large agreement with Azure. According to the outlet, it has upset OpenAI. In recent months the company has complained to Microsoft that Chinese firms are "distilling" its AI. Distillation means feeding queries to someone else's model and using its answers as training material for your own, competing model. A report last month mentioned DeepSeek and Moonshot in this context.
An important caveat: the publication does not say that ByteDance itself distilled OpenAI's models. The company has its own developments, including the Seedream family of image generators. Buying access to other companies' models is not the same as copying them, and the two should not be conflated.
Another route to chips is also closing
At the same time, another path ByteDance had to American AI hardware is closing. Banks have suspended work on a $480M loan to buy Nvidia chips that the company intended to use, Bloomberg reports. The loan was requested by Singapore-based Aolani Cloud: the funds were needed for GPUs for a data center in Malaysia.
On Monday, the US introduced a rule requiring data center operators in a number of countries, including Singapore and Malaysia, to obtain a license for advanced AI chips. The measure targets Chinese companies that rent such chips remotely. Bankers at five lenders discussed exiting the deal, and some called the loan dead. Aolani said it is working with regulators and lenders to complete the project under the new rule. DBS, which is organizing the loan, declined to comment.
What this means for practitioners
First: access to models is not only a question of technology but also of the supply channel. The same OpenAI model follows different rules on different platforms, and those rules are set by contracts between vendors, not only by the model developer's policy. If you build products on cloud APIs, it is worth finding out in advance what happens when a vendor and a reseller disagree.
Second: restrictions on chip exports are shifting from direct sales to renting capacity. Buying GPUs in Malaysia and renting them out to Chinese customers is becoming harder, so the compute shortage for Chinese companies is unlikely to ease. This explains why ByteDance is simultaneously seeking ready-made models through Azure and hardware through loans.
Third: the risk for Microsoft itself. If OpenAI insists on strict restrictions, partners will have to negotiate how exactly to apply them in the cloud. For now, these are reports citing anonymous sources: according to the material, neither ByteDance, Microsoft, nor OpenAI has publicly confirmed the deal. The final terms may change or may not happen at all. But the arrangement itself shows where the real boundaries of access to frontier AI lie: not in the statements of labs, but in cloud provider contracts and chip export permits.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What are the key details on The loophole around OpenAI's ban?
OpenAI does not let most Chinese companies use its models. But Microsoft holds the rights to sell OpenAI technology through Azure and, according to The Information, does not apply the same ban. The scheme is simple: there is no direct access, but there is access through Microsoft's cloud. The new contract would be added to ByteDance's existing large agreement with Azure. According to the outlet, it has upset OpenAI.
What are the key details on Another route to chips is also closing?
At the same time, another path ByteDance had to American AI hardware is closing. Banks have suspended work on a $480M loan to buy Nvidia chips that the company intended to use, Bloomberg reports. The loan was requested by Singapore-based Aolani Cloud: the funds were needed for GPUs for a data center in Malaysia. On Monday, the US introduced a rule requiring data center operators in a number of countries, including Singapore and Malaysia, to obtain a license for advanced AI chips.
What this means for practitioners?
First: access to models is not only a question of technology but also of the supply channel. The same OpenAI model follows different rules on different platforms, and those rules are set by contracts between vendors, not only by the model developer's policy. If you build products on cloud APIs, it is worth finding out in advance what happens when a vendor and a reseller disagree. Second: restrictions on chip exports are shifting from direct sales to renting capacity.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.