Tencent

Microsoft sells OpenAI models to China’s biggest tech firms while rivals stay away

AI News, Institutional Ant Group, Azure, ByteDance, China, GPT Models, Microsoft, OpenAI, Tencent

AI NewsJune 18, 2026·6 min read

Microsoft is the sole authorized distributor of OpenAI’s GPT models to Chinese tech giants, generating Azure AI revenue that tripled year-over-year while competitors OpenAI and Anthropic refuse direct market entry due to IP theft concerns. This arrangement positions Microsoft as a strategic intermediary in U.S.-China AI competition, extracting substantial licensing fees while shielding American model creators from direct exposure to Chinese distillation practices.

ByteDance spending over $1 billion annually on Microsoft Azure AI services, making it the largest customer in China region
Azure AI revenue in China tripled in fiscal year ending June 2025, following a 400% increase the prior year
Microsoft’s China business represents 1.5% of overall revenue despite regional growth, indicating vast total scale of operations

$1B+ ByteDance’s projected annual spending on Microsoft AI and cloud services in China
3x Azure AI revenue growth rate in China year-over-year through June 2025
1.5% Microsoft’s China revenue as share of total 2024 company revenue

Microsoft has secured an exclusive commercial foothold in China’s artificial intelligence market by leveraging its contractual authority with OpenAI to license GPT models directly to Chinese technology firms.

ByteDance, Ant Group, Tencent, and Meituan now access advanced OpenAI capabilities through Microsoft’s Azure cloud platform, creating a revenue stream that has grown faster in China than anywhere else globally.

The arrangement sidesteps the direct market entry that OpenAI and Anthropic have rejected, allowing both American companies to maintain distance from Chinese operations while Microsoft captures the licensing premium and relationship economics of serving the world’s second-largest AI ecosystem.

This configuration emerged from Microsoft’s unique contractual position with OpenAI. Unlike other partners bound by OpenAI’s own geographic and ethical restrictions, Microsoft negotiated independent authority to set terms for selling GPT models internationally.

That autonomy became commercially decisive when OpenAI and Anthropic declined to operate in China, leaving Microsoft as the exclusive authorized channel for Chinese enterprises seeking access to frontier U.S. AI capabilities.

The arrangement creates a stable, defensible competitive moat: Chinese customers cannot bypass Microsoft to license directly from OpenAI, and American AI companies avoid the regulatory and IP risks of Chinese operations while maintaining indirect market presence through Microsoft’s infrastructure.

ByteDance spending $1 billion annually on Microsoft as China’s largest AI customer

ByteDance’s consumption of Microsoft’s Azure AI services has emerged as the primary driver of regional growth, with the Beijing-based company on track to spend over $1 billion per year on licensing and cloud infrastructure.

That spending level makes ByteDance Microsoft’s single largest AI customer in the China region, a position that reflects both the parent company’s dominance in Chinese consumer technology and its aggressive investment in competing large language models through its Doubao chatbot platform.

ByteDance’s appetite for GPT model access suggests the company views licensing advanced capabilities as faster and more cost-effective than building comparable systems internally, at least for certain enterprise and consumer applications.

The magnitude of ByteDance’s commitment underscores the pricing power Microsoft wields as the exclusive distributor. At over $1 billion annually, the customer represents a material revenue stream even for Microsoft’s massive scale.

The spend reflects not just API consumption but also underlying cloud infrastructure costs, professional services, and integration work required to operationalize advanced AI models across ByteDance’s portfolio of platforms. That bundled arrangement locks ByteDance into Microsoft’s ecosystem and creates switching costs that would be substantial if alternative sources became available.

ByteDance’s parallel investment in its own Doubao chatbot indicates the company is simultaneously building independent capabilities while licensing OpenAI’s models, a diversification strategy that insulates ByteDance from overreliance on Microsoft but also demonstrates confidence in its own technical trajectory.

Azure AI revenue in China tripled year-over-year, outpacing all other Microsoft regions

Microsoft’s Azure AI business in China expanded at an exceptional rate, tripling in the fiscal year ending June 2025 after posting a 400% increase in the prior year. That growth trajectory far exceeds the company’s cloud and AI expansion in Europe, North America, and other regions, signaling that Chinese demand for frontier AI models has eclipsed adoption rates in mature markets.

The acceleration reflects both the concentration of Chinese tech spending among a handful of megacap companies and the absence of competing suppliers willing to serve the market at scale.

The growth metrics reveal two distinct patterns: the base-building phase of the prior fiscal year, when ByteDance and other early adopters deployed GPT models broadly across operations, followed by deeper penetration and higher consumption per customer in the most recent year.

That progression typically precedes margin expansion and increased switching costs, suggesting Microsoft’s China AI business is transitioning from land-and-expand mode into a mature revenue stream with predictable retention dynamics.

The 3x expansion also outpaced general cloud adoption in China, indicating that GPT licensing specifically, not baseline infrastructure, drove the outsized performance.

Despite the regional growth rates, Microsoft’s president Brad Smith characterized the China business as representing approximately 1.5% of the company’s overall 2024 revenue.

That statement provides scale context: even with tripling growth, China remains a fractional portion of Microsoft’s nearly $250 billion annual revenue, though the fractional nature of the percentage does not diminish the strategic importance of maintaining exclusive access to the world’s second-largest technology market or the leverage Microsoft derives from being the sole U.S. AI company operating at meaningful scale in China.

OpenAI and Anthropic cite IP theft risks, leaving Microsoft as sole authorized distributor

OpenAI and Anthropic have publicly and privately refused direct entry into China, citing concerns about intellectual property theft and the risk of Chinese companies using distillation techniques to reverse-engineer advanced models.

Distillation represents a specific technical threat: Chinese firms could train their own smaller models by observing the outputs of GPT systems, effectively learning the underlying weights and reasoning patterns without paying for OpenAI’s research and computational investment.

That technique has already surfaced as a concern in OpenAI’s private communications with Microsoft, indicating the company views Chinese distillation as a material risk rather than a theoretical one.

OpenAI has reportedly pressed Microsoft to strengthen protections against this specific threat, pushing for contractual restrictions on how Chinese customers can use GPT outputs and deploy models downstream.

Microsoft has responded by implementing a technical architecture that keeps OpenAI models off servers physically located in China; instead, Chinese customers access the models remotely over the internet from data centers in Singapore and other nearby countries.

That geographic isolation reduces, though does not eliminate, the risk of local server compromise or wholesale model extraction by Chinese government actors or advanced technical teams.

Microsoft has also deployed automated monitoring tools to track how Chinese customers use the models and has restricted Azure AI access to established companies rather than offering service to startups or individuals.

Those restrictions create a managed customer base where Microsoft can implement oversight and contractual enforcement, reducing the surface area for distillation attacks or unauthorized model replication.

The monitoring infrastructure and customer vetting represent a form of risk mitigation that OpenAI was unwilling to undertake itself, but that Microsoft deemed necessary to unlock the China market opportunity while maintaining acceptable risk parameters for its partner.

Anthropic’s refusal to enter China alongside OpenAI suggests the IP theft concerns are industry-wide and not unique to OpenAI’s circumstances, indicating a broader consensus among U.S. AI labs that direct Chinese operations carry unacceptable compliance and security risks.

Microsoft’s exclusive distributor role creates strategic leverage in U.S.-China competition

Microsoft’s position as the sole authorized channel for U.S. frontier AI models into China creates strategic value that extends beyond the licensing revenue. The company has described itself as the primary business connecting the AI innovation centers of the U.S. West Coast and China’s East Coast, a characterization that understates the strategic gatekeeping function Microsoft exercises.

By controlling access and monitoring usage, Microsoft can observe how Chinese companies deploy advanced AI, gather competitive intelligence on Chinese AI development priorities, and maintain a relationship with major Chinese technology firms that would otherwise be unavailable to U.S. companies under current export control regimes.

That intermediary position also allows Microsoft to serve both sides of the U.S.-China technology competition simultaneously without direct conflict. The company profits from licensing American technology to Chinese customers while maintaining operations and market share in the United States. Microsoft leadership has

Get this in your inboxThe Crypto Coin Show newsletter covers the policy and market moves institutional crypto investors are pricing in.

DeepSeek is seeking funding at a valuation above $20 billion

Blockchain AI Funding, Alibaba, Chinese AI Startups, DeepSeek, Huawei Chips, Nvidia, Tencent

AI NewsApril 22, 2026·5 min read

DeepSeek’s valuation has more than doubled to above $20 billion in less than a week as Tencent and Alibaba enter funding discussions, signaling intensifying competition between U.S. and Chinese AI stacks at a moment when chip performance gaps remain wide but closing. For institutional crypto and blockchain investors, this mirrors broader geopolitical technology decoupling that could reshape compute infrastructure markets and alter which blockchain protocols and AI-native chains gain adoption in different regions.

DeepSeek’s valuation jumped from $10 billion to above $20 billion within days as Tencent and Alibaba initiated investment talks.
Nvidia CEO Jensen Huang warned that if DeepSeek optimizes AI models for Huawei chips instead of American hardware, it would be a “horrible outcome” for the U.S.
Huawei’s Ascend 910C currently delivers only 60% of the inference performance of Nvidia’s H100, with American chips estimated to be 5x more powerful today but potentially 17x by 2027.

$20B+ DeepSeek’s target valuation versus $10 billion floor announced Friday
60% Huawei Ascend 910C performance relative to Nvidia H100 benchmark
17x Projected U.S. chip performance advantage over Chinese rivals by 2027

DeepSeek’s valuation has surged past $20 billion in funding discussions with China’s largest tech firms, less than a week after the AI startup disclosed it was seeking at least $300 million at a $10 billion floor. Tencent Holdings and Alibaba Group are now in active talks to invest in the Chinese startup, according to reports on Wednesday citing four people familiar with the negotiations.

The rapid repricing reflects intensifying competitive pressure in AI infrastructure and a fundamental shift in how geopolitical technology competition plays out at the infrastructure layer, a concern with direct implications for how compute resources, and by extension blockchain networks, become distributed globally.

The speed of this valuation increase mirrors the broader urgency surrounding AI chip sovereignty. Within days of opening its first outside investor discussions, DeepSeek has attracted attention from two of China’s most valuable technology firms, each with incentives to reduce dependence on American semiconductor supply chains.

The talks remain ongoing, meaning the final valuation and funding amount could still shift, but the direction signals how quickly capital mobilizes around AI infrastructure bets perceived as strategically important to national technology ecosystems.

Nvidia CEO warns DeepSeek optimization for Huawei chips would harm U.S. competitiveness

Jensen Huang, Nvidia’s chief executive, articulated the core concern driving this funding competition during a Wednesday podcast appearance. He described it as a “horrible outcome” for the United States if DeepSeek’s AI models were optimized to run efficiently on Huawei chips rather than American processors.

Huang’s warning cuts to the heart of why institutional investors should track this funding round closely: the outcome determines whether future AI model development follows an American technology stack or a Chinese one, with cascading effects on which compute platforms, and which regional blockchain and protocol ecosystems, gain adoption in different markets.

If future AI models are optimised in a very different way than the American tech stack, and as AI diffuses out into the rest of the world with Chinese standards and technology, China will become superior to the United States.

Jensen Huang, Chief Executive, Nvidia

Huang framed the competition not as a near-term chip performance race, but as a standards-setting moment. Once large AI model developers optimize for a particular chip architecture, downstream applications, including blockchain infrastructure, AI-native protocols, and decentralized compute networks, tend to follow that same hardware assumption.

If DeepSeek V4 and future versions run optimally on Ascend silicon rather than H100s, that creates a self-reinforcing ecosystem where Chinese-standard hardware becomes the assumed baseline for global AI development, analogous to how x86 architecture became the assumed standard for decades of enterprise computing.

The strategic concern Huang raised extends beyond raw performance metrics. He noted that China possesses “abundant energy” for training and inference workloads and a “large pool of AI researchers,” meaning the country can overcome near-term chip deficits through scale and talent concentration.

DeepSeek’s own emergence as a credible alternative to OpenAI and Claude demonstrates this: the startup achieved competitive performance despite working under U.S. export controls on advanced semiconductors.

Huawei Ascend 910C delivers 60% of H100 performance but gap narrows as Chinese production scales

Current performance data suggests American chips still hold a decisive advantage. Huawei’s Ascend 910C, the chip preceding its newer 950PR model, delivers approximately 60% of the inference performance that Nvidia’s H100 achieves, a significant shortfall.

The H100 itself, however, is already two generations behind Nvidia’s current flagship processor, meaning the real-world gap between Huawei’s available silicon and Nvidia’s most advanced offerings is even wider.

The absolute performance spread remains substantial: American AI chips are estimated to be roughly five times more powerful than Chinese alternatives today.

However, that lead is projected to narrow considerably, with estimates suggesting American chips could hold only a 17-times performance advantage by 2027, a compression driven by both Chinese chip design improvements and anticipated slowdowns in American chipmaker gains as physical design limits tighten.

Production volume compounds the picture. Huawei is targeting 750,000 AI chip shipments in 2026, but that total production represents only 3% to 5% of the combined computing power Nvidia generates across all its customers. This means Chinese AI development remains fundamentally constrained by silicon availability, not just efficiency.

DeepSeek’s funding round therefore carries geopolitical weight precisely because capital flowing to the startup can accelerate model optimization techniques that extract maximum performance from scarce Huawei chips, reducing China’s dependence on incremental hardware improvements alone.

DeepSeek funding surge signals broader AI infrastructure competition reshaping compute markets

DeepSeek’s valuation jump coincides with a wider funding surge in AI infrastructure. Vast Data, a storage and compute company, announced a $1 billion funding round on the same day Huang made his remarks, indicating that capital markets are pricing in sustained competition for alternative AI stacks and infrastructure layers outside the dominant American ecosystem.

For institutional investors tracking blockchain and cryptocurrency infrastructure, this bifurcation matters directly.

If AI model training and inference optimizes around region-specific hardware stacks, blockchain protocols and decentralized compute networks will likely follow similar geographic and technical boundaries.

A Chinese-optimized AI stack built on Huawei chips creates natural incentives for blockchain protocols to optimize for deployment in that ecosystem, potentially fragmenting the global compute market into American-standard and Chinese-standard spheres.

Conversely, if Nvidia and American chipmakers maintain their advantage, the current concentration of AI infrastructure in U.S.-aligned jurisdictions will likely persist, benefiting blockchain platforms already embedded in that ecosystem.

The funding discussions involving Tencent and Alibaba also carry strategic weight because both firms operate blockchain or blockchain-adjacent platforms.

Tencent’s involvement in digital asset ecosystems and Alibaba’s blockchain infrastructure initiatives mean their investment in DeepSeek is not purely financial, but helps coordinate AI and blockchain infrastructure strategies at the firm level, accelerating the development of integrated Chinese technology stacks that reduce reliance on American components across multiple infrastructure layers.

The next decisive indicator is whether DeepSeek publicly discloses optimization metrics for Huawei Ascend chips versus Nvidia processors in model performance benchmarks. If inference latency and throughput on Ascend hardware approaches parity with H100 performance within the next two quarters, it signals that model optimization rather than raw chip performance is becoming the binding constraint on Chinese AI development, a turning point that would trigger accelerated adoption of Huawei chips by other Chinese AI firms and potentially force institutional investors to reprice both American chip export restrictions and regional blockchain infrastructure competition.

Get this in your inboxThe Crypto Coin Show newsletter covers the policy and market moves institutional crypto investors are pricing in.