Ali Ikhwan
In the high-stakes arena of artificial intelligence, where computing power is the new global currency, a London-based infrastructure firm has found itself at the epicenter of a geopolitical tug-of-war. Nscale, a burgeoning "neocloud" provider, has disclosed that a significant majority of its 2025 revenue was derived from a single, high-profile client: Spring, a subsidiary of the Chinese technology giant ByteDance.
This partnership, which facilitated ByteDance’s access to 2,304 of Nvidia’s ultra-powerful B200 artificial intelligence accelerators located in a data center in Glomfjord, Norway, highlights the creative, and often controversial, lengths to which firms will go to navigate the tightening web of U.S. export controls. While the arrangement appears to operate within the current letter of international law, it has ignited a fierce debate regarding the efficacy of Washington’s efforts to curb China’s technological ascent.
The Architecture of the Deal: Main Facts
At the heart of the controversy is a $24 million contract—a figure that represents roughly 73% of Nscale’s total $33 million annual revenue for 2025. This deal was not merely a standard cloud-hosting agreement; it was a strategic mobilization of high-end silicon. By leveraging a specialized facility in the remote Norwegian town of Glomfjord, Nscale provided the computational muscle required by Spring to fuel its demanding AI research and development pipelines.
The hardware in question, Nvidia’s B200 "Blackwell" chips, represents the current pinnacle of AI processing capability. These units are designed to handle the massive, complex calculations required to train large language models (LLMs) and generative AI architectures. Because the U.S. Department of Commerce has imposed sweeping restrictions on the export of such hardware to China, ByteDance—the parent company of TikTok—has been forced to seek indirect access to these capabilities. Nscale, by hosting these chips on European soil and providing "infrastructure-as-a-service" (IaaS) to a Chinese subsidiary, effectively bypassed the physical constraints that would otherwise have prevented ByteDance from utilizing this specific generation of hardware.
A Chronology of Strategic Expansion
The path to this deal was paved by a series of calculated moves by Nscale to position itself as a specialized provider of GPU-intensive cloud services.
- Early 2024: Nscale secures significant venture capital funding, earmarked for the acquisition of high-end Nvidia hardware. The company pivots its business model toward the "neocloud" sector, focusing on providing raw compute power to companies training frontier AI models.
- Mid-2024: Nscale announces the expansion of its data center footprint, specifically targeting regions with abundant, low-cost, and green energy, such as Norway. The Glomfjord facility is identified as a primary hub for this expansion.
- Late 2024: Negotiations between Nscale and Spring reach a climax. The deal is structured to ensure that the hardware remains within a NATO-member country, satisfying the technical requirements of the U.S. Bureau of Industry and Security (BIS) export regulations, which govern the export of "controlled" technology to specific countries, including China.
- Q1 2025: The deployment of 2,304 Nvidia B200 chips is completed in Glomfjord. Spring begins remote processing operations, effectively utilizing the chips for its internal AI workflows.
- Q3 2025: Financial disclosures emerge, revealing that the Spring contract accounts for nearly three-quarters of Nscale’s revenue, sparking inquiries into the nature of the partnership and its compliance with the spirit, if not the letter, of international trade policy.
Supporting Data and Technical Context
To understand why this deal is so significant, one must look at the technical disparity between the hardware available to Chinese firms and the hardware available to their Western counterparts.
The Nvidia B200 is an iteration of the Blackwell architecture, offering performance gains that are orders of magnitude greater than the "China-compliant" chips Nvidia is authorized to sell directly to the Chinese market. For a company like ByteDance, which is locked in a global race to develop competitive generative AI, the ability to train models on unrestricted hardware is a massive competitive advantage.
| Metric | Nvidia B200 (Restricted) | China-Compliant Alternatives |
|---|---|---|
| Compute Performance | Ultra-High (Frontier AI) | Mid-Range (Limited LLM Training) |
| Memory Bandwidth | Superior | Constrained |
| Availability to China | Prohibited (Direct Export) | Authorized |
Nscale’s revenue model is heavily reliant on this high-end compute density. By concentrating 2,304 B200 chips in a single cluster, they have created a "compute island" capable of training complex models in a fraction of the time required by standard cloud providers. The $24 million contract reflects the premium price that entities like ByteDance are willing to pay for access to these restricted assets in a jurisdiction that is not subject to U.S. direct export bans.
The Regulatory Loophole: Official Responses and Legal Scrutiny
The Financial Times and other investigative outlets have noted that this arrangement exploits a "jurisdictional gap" in U.S. policy. Current export controls are largely predicated on the physical destination of the hardware. If the hardware is located in a country that is not on the list of prohibited destinations, and if the firm operating the hardware is not on the U.S. Entity List in a way that restricts access to remote compute, the transaction may be classified as "legal."
Nscale has maintained that it complies with all applicable laws. In statements provided to investors, the company emphasized that it performs rigorous "Know Your Customer" (KYC) checks and ensures that no hardware is physically shipped to China. However, the company has declined to comment on the long-term sustainability of the relationship given the evolving political climate.
From the perspective of Washington, the situation is increasingly viewed as an "end-run" around national security measures. U.S. policymakers have argued that allowing Chinese companies to access top-tier compute via remote cloud providers in third-party countries undermines the entire premise of the export controls. There is growing sentiment on Capitol Hill that the "Export Administration Regulations" (EAR) must be updated to cover "Compute-as-a-Service" (CaaS) to Chinese entities, regardless of where the data center is located.
Implications: A Looming Regulatory Reckoning
The Nscale-ByteDance deal serves as a case study for the risks inherent in the modern, globalized tech supply chain. For Nscale, the reputational risk is twofold. First, there is the risk of being labeled an "enabler" of Chinese AI ambitions, which could alienate Western venture capital firms and potential U.S. enterprise customers. Second, there is the existential risk that the U.S. government will issue an executive order or update export rules to specifically target this type of remote-access cloud computing.
1. The Fragmentation of Cloud Computing
If the U.S. succeeds in closing these loopholes, we are likely to see a "balkanization" of cloud infrastructure. Companies like Nscale may be forced to choose between serving the U.S. market or the Chinese market, as the two spheres of influence become mutually exclusive for high-end hardware providers.
2. The Rise of "Sovereign Clouds"
The push for stricter controls may accelerate the development of "sovereign clouds" in Europe and Asia, where countries aim to build their own AI infrastructure independent of U.S. hardware dominance. However, as long as Nvidia remains the undisputed leader in GPU technology, this independence remains a distant goal.
3. Investor Uncertainty
For investors, the Nscale case is a warning sign. High-growth revenue streams that rely on regulatory arbitrage are inherently fragile. A single policy change in Washington could render 73% of Nscale’s revenue illegal overnight, creating a significant risk of insolvency.
4. The Future of Export Controls
The Nscale-ByteDance saga suggests that the next phase of the "Chip War" will move away from physical border controls and toward the regulation of access. As AI development becomes increasingly dependent on massive, centralized compute clusters, the focus will shift to who is allowed to "rent" time on these machines, rather than just who is allowed to purchase the silicon.
Conclusion
The story of Nscale and its $24 million deal with ByteDance is not just about a single company in Norway; it is a symptom of a larger, systemic shift in the global order. As AI capabilities become synonymous with national power, the U.S. is struggling to define the boundaries of its influence in a digital world where data and compute have no physical borders.
For now, the Glomfjord data center continues to hum with the activity of thousands of Nvidia chips, silently crunching data for a Chinese tech giant while operating under the umbrella of a UK-based firm. Whether this arrangement remains a viable business model or becomes a casualty of the escalating tech cold war remains to be seen. One thing is certain: the era of frictionless global cloud computing is coming to an end, replaced by a complex, regulated landscape where every cycle of compute is subject to the watchful eyes of global superpowers.
Lina Hope
Suro Senen
