Key Takeaways

  • UK-based AI infrastructure company Nscale is facing increased regulatory scrutiny over its agreement to provide computing capacity to ByteDance, the Chinese parent company of TikTok.
  • The arrangement gives ByteDance’s Singapore-based subsidiary, Spring, remote access to 2,304 Nvidia B200 AI processors installed at Nscale’s data centre in Glomfjord, Norway.
  • The US Commerce Department has introduced new export-control rules targeting overseas data centres that provide significant computing power to companies headquartered in China and other restricted destinations.
  • Under the new rules, qualifying data centres already providing computing access to restricted companies have 90 days from November 19, 2026, to obtain the required US government authorisation if they want to continue those services.
  • Bloomberg reported that Nscale may need to apply for a US licence to continue supplying ByteDance with the computing capacity.
  • Nscale’s relationship with ByteDance has attracted attention because the Singapore subsidiary contributed approximately $24 million of the company’s $33 million revenue in 2025.
  • The deal highlights how Washington is attempting to close a loophole that allowed Chinese companies to rent advanced American AI chips through overseas cloud providers.
  • Nscale’s compliance exposure comes as the company expands its business with major Western technology customers, including Microsoft and Anthropic.

Nscale’s agreement to provide artificial intelligence computing capacity to ByteDance is facing increased scrutiny after the US government introduced new restrictions targeting overseas access to advanced American semiconductor technology. The British AI infrastructure company operates data centres equipped with Nvidia processors, allowing customers to rent computing capacity remotely instead of purchasing the physical chips. One of its important customers is Spring, a Singapore-based subsidiary of ByteDance, which has used 2,304 Nvidia B200 processors installed at Nscale’s Glomfjord data centre in Norway.

The arrangement has attracted attention because US export controls restrict China’s access to certain advanced AI chips, but Chinese technology companies have continued to obtain computing capacity through overseas cloud providers. Washington’s new rules, effective November 19, 2026, extend export-control requirements to certain remote computing arrangements involving companies headquartered in restricted countries. Bloomberg reported that Nscale may have 90 days to seek a US licence to continue providing the relevant computing services to ByteDance. The outcome could affect the customer relationship, Nscale’s infrastructure strategy and the broader market for international AI computing services.

What Is the Nscale–ByteDance Deal?

Nscale is a UK-based AI infrastructure company that provides high-performance computing capacity to organisations developing and operating artificial intelligence systems.

Its business model is commonly described as a neocloud model.

Rather than selling individual processors to customers, companies such as Nscale build or operate specialised data centres containing large numbers of high-performance graphics processing units, or GPUs.

Customers pay to access the computing resources remotely.

This allows AI developers to train models, run inference workloads and perform other computationally intensive tasks without building their own data centres or purchasing thousands of processors.

Nscale established an important relationship with ByteDance through the latter’s Singapore-based subsidiary, Spring.

The arrangement provides Spring with access to 2,304 Nvidia B200 processors at Nscale’s Glomfjord facility in Norway.

The physical chips remain installed in the Norwegian data centre, while ByteDance’s subsidiary accesses their computing capacity remotely.

This distinction is central to the regulatory dispute.

US export controls traditionally focused heavily on the physical export and transfer of advanced semiconductors. Remote access arrangements created a different route through which companies in restricted countries could use powerful American-designed processors without importing them directly.

The new US rules are designed to address that route.

Why the US Government Is Examining the Arrangement

The US government has been tightening restrictions on China’s access to advanced semiconductor technology because of concerns about national security and the potential military applications of high-performance computing.

Advanced GPUs are essential for training sophisticated AI models.

They can also support applications involving scientific research, cybersecurity, intelligence analysis and other computationally demanding activities.

Washington has argued that access to large amounts of advanced computing power could help restricted countries develop capabilities that pose national security risks.

However, overseas data centres created a regulatory challenge.

A Chinese company that cannot directly purchase certain advanced Nvidia processors may still be able to rent computing capacity from a provider operating in Norway, the United Kingdom or another country.

The chips do not physically enter China, but the customer can remotely use their computing resources.

The Nscale–ByteDance arrangement illustrates this model.

The US Commerce Department’s latest rules are intended to make data-centre operators responsible for obtaining authorisation when they provide qualifying levels of computing power to restricted foreign customers.

This means that the location of the physical data centre is no longer the only relevant consideration. The identity of the remote customer, the amount of computing power provided and the intended use of that capacity can also determine whether a licence is required.

What the New US Chip Restrictions Say

The US Department of Commerce’s Bureau of Industry and Security (BIS) published an interim final rule titled Addressing Remote Access to Advanced Semiconductors in the Federal Register on November 19, 2026.

The regulation introduces a new end-use control covering certain advanced computing chips and related technology used in overseas data centres.

Under the rule, a licence is required when covered advanced computing items are destined for data-centre use that provides more than 12 million aggregate total processing performance (TPP) to or for companies headquartered in Macau or countries included in the US Country Group D:5.

China is included in the restricted group.

The rules also cover situations where a non-restricted intermediary provides computing resources that are ultimately used to train or supply AI models to a restricted company.

This is important because the regulation is designed to prevent companies from avoiding restrictions through subsidiaries, intermediaries or linked data centres.

Regulatory provisionWhat it means
Remote computing restrictionsCertain overseas data centres must obtain authorisation before providing qualifying computing capacity to restricted customers
12 million aggregate TPP thresholdThe rule uses a defined computing-performance threshold to identify covered arrangements
Linked data centresConnected facilities can be assessed together rather than treated as entirely separate locations
90-day transition periodExisting qualifying data-centre arrangements must obtain a licence within 90 days of the rule’s effective date
Customer due diligenceOperators and suppliers must examine customer ownership, control and intended use
Case-by-case reviewLicence applications are assessed by US authorities individually

The regulation became effective on November 19, 2026.

For qualifying existing arrangements, the 90-day transition period is intended to provide time to seek authorisation rather than requiring every existing service to stop immediately on the effective date.

The rule also includes an exception for computing capacity used exclusively for commercial recommender systems, such as systems that rank or suggest products and content. It does not extend that exception to generative AI systems that create new text, images, audio, video or software.

The full application of the regulation to Nscale’s arrangement will depend on the covered chips, computing capacity, end use and other facts assessed by the relevant parties and US authorities.

Nscale May Need a US Licence Within 90 Days

The regulatory change creates a new compliance requirement for Nscale.

Bloomberg reported that the company may need to apply to the US government for a licence to continue providing ByteDance with access to the advanced computing capacity in its overseas data centres.

The 90-day transition period is particularly important for companies that were already supplying computing power to restricted customers before the new regulation took effect.

Under the rule, qualifying existing data-centre operators must obtain the required authorisation within 90 days of November 19 if they want to continue providing covered services.

The licensing process will be reviewed on a case-by-case basis.

That means approval is not automatic.

US authorities can examine the customer, the intended use of the computing resources, the data-centre configuration and other national-security considerations before deciding whether to grant permission.

For Nscale, the immediate challenge is determining how the new requirements apply to its existing customer arrangements and whether the necessary authorisation can be secured.

The company has not publicly confirmed that a licence application has been approved.

It is also important not to treat the new rule as an immediate blanket ban on every overseas cloud service used by Chinese companies. Its restrictions depend on the specific technology, computing threshold, customer and end-use conditions set out in the regulation.

ByteDance’s Singapore Subsidiary Was a Major Nscale Customer

The commercial significance of the relationship is another reason the regulatory scrutiny matters.

Nscale’s IPO-related disclosures and supporting financing documents brought attention to Spring, the Singapore-based ByteDance subsidiary that used the company’s Norwegian infrastructure.

According to reporting based on those documents, Spring accounted for approximately $24 million of Nscale’s $33 million revenue in 2025.

<Chart content={{“chartType”:”pie”,”meta”:{“description”:”Approximate breakdown of Nscale’s 2025 revenue based on reported figures.”},”nameKey”:”customer”,”valueKey”:”revenue”,”series”:[{“dataKey”:”revenue”,”label”:”Revenue”,”valuePrefix”:”$”,”valueSuffix”:”m”}],”data”:[{“customer”:”Spring / ByteDance subsidiary”,”revenue”:24},{“customer”:”Other customers”,”revenue”:9}]}}/>

This means the ByteDance-linked customer represented approximately 73% of the company’s revenue during that year.

The arrangement also helped Nscale finance its infrastructure expansion.

A $105 million loan from Macquarie, alongside $35 million in equity, helped fund the purchase of advanced Nvidia processors.

The financing agreement included requirements for monitoring unusual computing activity and conducting due diligence related to the customer and its use of the infrastructure.

The relationship therefore had significance beyond ordinary cloud revenue. It was connected to the financing and development of Nscale’s early AI infrastructure business.

However, the company’s customer mix has since evolved.

Nscale has secured much larger commercial agreements with Microsoft and Anthropic. The company has indicated that the Spring contract’s share of overall revenue would fall below 20% as its business with other customers expanded.

This could reduce its dependence on ByteDance over time, although it does not automatically remove the regulatory obligations associated with an existing computing arrangement.

Why the Nvidia B200 Chip Is Important

The Nvidia B200 is part of the company’s Blackwell generation of AI processors, designed for demanding artificial intelligence workloads.

<AsyncImage query=”Nvidia Blackwell B200 GPU server rack data center” aspectRatio=”16:9″ width=”100%”/>

These processors are designed to support large-scale AI model training and inference.

Training involves using large datasets to develop the parameters of an AI model, while inference refers to using a trained model to generate responses or perform tasks.

Both processes can require substantial computing resources, particularly for large language models and multimodal AI systems.

The B200’s capabilities make it attractive to companies developing advanced AI products.

For ByteDance, access to overseas computing infrastructure can help support its AI development activities without requiring the company to import the physical processors directly into China.

For Nscale, offering access to advanced Nvidia hardware helps it compete with larger cloud providers seeking to serve the growing demand for AI computing.

The challenge is that the same technology has become a central focus of US export-control policy.

Washington wants to ensure that advanced American semiconductor technology does not reach restricted customers through indirect arrangements that undermine its national-security objectives.

Nscale’s Supplier Relationships Create Another Regulatory Challenge

The ByteDance arrangement is not the only regulatory issue facing Nscale.

The Financial Times reported on November 19 that the company had warned investors about potential exposure to suppliers with ownership links to a Chinese group already placed on the US Entity List.

The suppliers identified in related reporting include Kaytus Systems and Aivres Systems.

According to the Financial Times, both companies are ultimately majority-owned by IEIT Systems, a Chinese technology group on the US Entity List.

The Entity List imposes restrictions on exports, reexports and transfers of certain US-controlled technology to designated organisations.

The suppliers themselves were not identified as being on the Entity List in the reporting. The concern relates to their ownership connections and the possible impact of a planned expansion of US restrictions to certain affiliates of listed entities.

Nscale’s Glomfjord data centre in Norway has been associated with equipment supplied by Kaytus and Aivres.

The Financial Times reported that the use of Kaytus services at the facility had been requested by ByteDance, citing people familiar with the matter.

Nscale warned investors that changes to US export-control rules could require it to modify or terminate contractual arrangements, delay or cancel projects, or decline new business to remain compliant.

These supplier-related risks are separate from the new remote-computing restrictions, although both demonstrate how US export controls can affect an international AI infrastructure company.

Nscale’s Microsoft and Anthropic Deals Could Reduce Dependence on ByteDance

Nscale has been expanding its commercial relationships with major Western technology companies.

The company has secured large agreements involving Microsoft and Anthropic, two important players in the global AI industry.

<AsyncImageGroup query={[“Nscale AI data center exterior Norway”,”Microsoft hyperscale AI data center server racks”,”Anthropic Claude AI data center computing infrastructure”]} aspectRatio=”5:4″ layout=”carousel”/>

These agreements are important because large AI developers require access to substantial computing capacity to train and operate their models.

For infrastructure providers, long-term contracts can help justify investment in expensive GPUs, electricity supply, cooling systems and data-centre construction.

Nscale’s expanding customer base could reduce the proportion of revenue generated by Spring.

However, replacing one customer with another does not automatically resolve the regulatory challenge.

The company must still ensure that its infrastructure, suppliers and customer arrangements comply with the relevant export-control requirements.

Its ability to maintain access to Nvidia technology is also important. Advanced GPUs are central to its business model, and restrictions affecting procurement or deployment could influence project timelines and customer commitments.

Why the US Wants to Close the Overseas Cloud Computing Loophole

The Nscale case reflects a wider change in US semiconductor policy.

For several years, Washington’s export-control strategy has focused on restricting China’s ability to obtain advanced American chips directly.

However, cloud computing has complicated that approach.

A company does not necessarily need to own a GPU to use it. It can rent access to a data centre that already has the hardware installed.

This creates a distinction between owning a semiconductor and accessing the computing power it provides.

From the perspective of AI development, the practical result can be similar: a customer can use powerful processors to train models without taking physical possession of the chips.

The new BIS rule is intended to address this issue by regulating certain remote computing arrangements.

It also introduces stronger customer due-diligence expectations.

Data-centre operators, chip exporters and other relevant parties may need to examine direct and indirect ownership, customer identity and intended use.

The regulation also considers linked data centres, preventing companies from simply distributing computing capacity across connected facilities to avoid the threshold.

This approach could affect cloud providers operating in Europe, Asia and other regions that serve customers based in restricted countries.

What the New Rules Mean for the Global AI Industry

The regulatory change could reshape the international market for AI computing.

Cloud providers have increasingly built infrastructure outside the United States to serve customers seeking access to advanced Nvidia processors.

European data centres, in particular, have become important because they offer access to sophisticated infrastructure while operating outside China’s domestic technology restrictions.

The new rules could make these arrangements more complicated.

Potential consequences include:

  • Higher compliance costs: Data-centre operators may need additional legal, technical and customer-verification systems.
  • More licensing requirements: Providers serving restricted customers may need US government approval for certain services.
  • Changes in customer contracts: Existing agreements could require amendments or termination.
  • Infrastructure delays: Uncertainty over chip access may affect the timing of data-centre projects.
  • Customer diversification: AI infrastructure companies may place greater emphasis on customers that do not create export-control concerns.
  • Greater scrutiny of ownership: Suppliers and customers may face closer examination of their corporate structures and indirect relationships.

The rules could also influence how companies select locations for future AI infrastructure.

A data centre’s access to electricity, cooling, network connectivity and land will remain important, but its ability to serve customers under US export-control requirements could become equally significant.

The Bigger Picture

The scrutiny surrounding Nscale’s ByteDance agreement illustrates how the AI infrastructure business is becoming closely connected to international trade policy.

Advanced GPUs are no longer simply high-value technology products. They are strategic resources at the centre of competition between the United States and China.

The ability to access these processors can influence how quickly companies train AI models, develop new products and compete in the global technology market.

Nscale’s business model demonstrates how cloud infrastructure can provide access to powerful chips without transferring physical ownership to the customer. However, Washington’s new rules show that regulators are increasingly focused on the computing services those chips enable, not just where the hardware is shipped.

For Nscale, the challenge is to manage the regulatory requirements while expanding its customer base and infrastructure operations. For ByteDance, the issue is whether it can continue accessing the advanced computing resources needed for its AI development. For the wider industry, the new rules signal that overseas cloud access will no longer necessarily provide a route around restrictions on advanced semiconductor technology.

Looking Ahead

The immediate focus will be on how Nscale and other overseas data-centre operators respond to the 90-day licensing requirement. US authorities will need to assess qualifying applications, while providers will have to establish the identity, ownership and intended use of their customers. The outcome for the Nscale–ByteDance arrangement remains uncertain, and the new regulation does not by itself establish that the existing contract has been terminated or that Nscale has violated export-control law. Its ability to continue supplying the relevant computing capacity will depend on the rule’s application and any required authorisation.

Over the longer term, the restrictions could reshape the economics of the global AI cloud market. Providers may diversify their customer bases, strengthen compliance systems and reconsider where they deploy advanced Nvidia processors. Nscale’s growing relationships with Microsoft and Anthropic could reduce its commercial dependence on ByteDance, but the company will still need reliable access to advanced chips and compliant suppliers. The broader contest between Washington and Beijing is therefore likely to influence not only semiconductor sales but also the international market for AI computing services.

Get the day’s top stories in your inbox

One concise email. No spam, unsubscribe anytime.