Infinigence AI Raises $97M to Power Token Economy

Infinigence AI (无问芯穹), one of China's leading AI-native infrastructure companies, has closed a new funding round exceeding 700 million yuan (approximately $97 million), reinforcing its position as the most heavily funded startup in China's AI infrastructure sector. The round, which was quietly completed several months ago, signals growing investor confidence in the middleware layer that connects heterogeneous chips to real-world AI productivity.

The company simultaneously introduced what it calls an 'AI Productivity Formula' — a framework for measuring and optimizing the full value chain from raw electricity to AI tokens to tangible business output — positioning itself as a critical hub in the emerging Token economy.

Key Takeaways

• Infinigence AI raised over $97 million (700M+ yuan) in its latest round, maintaining the largest cumulative funding among China's AI-native infrastructure companies
• The round was co-led by Hangzhou High-tech Jinji Group and Huiyuan Capital, with participation from over a dozen strategic and financial investors
• Funds will target 3 strategic priorities: heterogeneous computing leadership, power-to-token efficiency, and token-to-productivity conversion
• The company debuted an 'AI Productivity Formula' connecting compute infrastructure to enterprise value creation
• Existing shareholders including Legend Capital, Shanghai Guotou Futeng, and Yuanzhi Future all increased their stakes
• Infinigence is building an enterprise-grade AI agent service platform for cross-industry deployment

Who Backed the Round — And Why It Matters

The investor roster reads like a who's who of Chinese strategic capital. Hangzhou High-tech Jinji Group and Huiyuan Capital co-led the round, bringing both municipal government backing and private sector conviction. Follow-on investors include Guoxing Capital, data center operator Chindata Group (秦淮数据), GF Qianhe, Lihe Qingtong, China Insurance Investment, AEF NextGen, Tengrui Capital, Colorlight (卡莱特), CITIC Securities Capital, and Will (Kuande Intelligent Learning Lab).

Notably, 3 existing shareholders — Legend Capital, Shanghai Guotou Futeng, and Yuanzhi Future — doubled down with additional investments. In a market where many AI startups struggle to retain existing backers, repeat investment from prior-round shareholders is a strong endorsement of Infinigence's execution and technology moat.

The diversity of the investor base is itself a signal. The participation of Chindata Group, one of Asia's largest hyperscale data center operators, suggests deep infrastructure-layer synergies. Meanwhile, insurance and government-backed capital indicate that Infinigence's technology is seen as strategically important to China's broader AI compute ambitions.

The AI Productivity Formula: From Electricity to Enterprise Value

At the heart of Infinigence's pitch is a deceptively simple idea: the entire AI value chain can be expressed as a productivity equation. The company breaks this into 2 critical conversion steps:

1. Electricity → Tokens: How efficiently raw power is converted into usable AI inference and training capacity (measured in tokens processed per kilowatt-hour)
2. Tokens → Productivity: How effectively those tokens translate into real business outcomes — whether that is customer service automation, code generation, or industrial optimization

This framework matters because it shifts the conversation from raw chip performance to end-to-end system efficiency. In a world where NVIDIA H100 GPUs remain scarce and expensive — and where Chinese companies face ongoing U.S. export restrictions on advanced chips — squeezing maximum value from available hardware is not just an optimization exercise. It is a strategic imperative.

Infinigence's approach stands in contrast to companies that focus narrowly on either the chip layer (like AMD or domestic Chinese GPU makers) or the application layer (like enterprise AI platforms). By optimizing the full stack from silicon to software to business logic, Infinigence occupies a unique middleware position in the AI infrastructure landscape.

Heterogeneous Computing: Making Every Chip Count

The first of Infinigence's 3 investment priorities is strengthening its multi-chip heterogeneous computing technology. Unlike companies that optimize for a single GPU architecture, Infinigence has built a software layer that abstracts across different chip types — enabling workloads to run efficiently on whatever hardware is available.

This capability is particularly valuable in the Chinese market, where companies must work with a fragmented ecosystem of domestic and international processors:

• NVIDIA GPUs (where export-compliant versions are available)
• Huawei Ascend processors
• Cambricon and other domestic AI accelerators
• Intel and AMD CPUs and GPUs for hybrid workloads
• Custom ASICs from various Chinese chipmakers

By building a unified orchestration layer, Infinigence allows enterprises and cloud providers to maximize usable compute regardless of the underlying hardware mix. This is analogous to what companies like Run:ai (recently acquired by NVIDIA for $700 million) and Anyscale do in Western markets — but with the added complexity of managing truly heterogeneous chip architectures rather than variations within a single vendor's lineup.

The company claims its technology delivers what it calls 'extreme optimization' of available compute at scale, turning fragmented chip resources into a cohesive, high-performance AI compute fabric.

Enterprise AI Agents: The Token-to-Value Bridge

The third prong of Infinigence's strategy addresses the demand side of the equation. The company is building an enterprise-grade intelligent agent service platform designed to deploy AI solutions across diverse industry verticals.

This platform serves as the bridge between raw token generation capacity and actual business productivity. Key target sectors likely include:

• Financial services (risk modeling, automated trading analysis)
• Manufacturing (predictive maintenance, quality control)
• Healthcare (medical image analysis, clinical decision support)
• Government and public sector (smart city applications, administrative automation)
• Telecommunications (network optimization, customer service)

The agent-based approach reflects a broader industry trend. Rather than delivering AI as a monolithic model API, companies like Infinigence are packaging AI capabilities as autonomous agents that can reason, plan, and execute multi-step tasks within specific business contexts. This mirrors moves by Western companies like Microsoft (with Copilot Studio), Salesforce (with Agentforce), and Google (with Vertex AI agents).

What differentiates Infinigence is the vertical integration. Because the company controls the compute optimization layer, it can potentially deliver AI agent services at lower cost and higher efficiency than competitors who rely on third-party infrastructure.

Industry Context: China's AI Infrastructure Race Intensifies

Infinigence's fundraise comes at a pivotal moment for China's AI ecosystem. Following the shock debut of DeepSeek's cost-efficient models in early 2025, the Chinese market has seen a dramatic acceleration in AI infrastructure investment. The message from DeepSeek was clear: with the right software optimization, world-class AI performance is achievable even without access to the most advanced chips.

This has created a gold rush for companies that can maximize the performance of available hardware. Infinigence sits squarely in this sweet spot. While U.S. export controls continue to restrict China's access to cutting-edge NVIDIA GPUs, companies like Infinigence provide the software intelligence needed to close the compute gap through efficiency rather than raw hardware power.

The broader global AI infrastructure market is projected to exceed $300 billion by 2027, according to multiple industry estimates. Within China alone, government-backed 'computing power' initiatives across provinces have earmarked tens of billions of dollars for AI infrastructure buildout. Infinigence's positioning as a hardware-agnostic optimization layer makes it a natural beneficiary of these investments regardless of which chip vendors ultimately dominate.

What This Means for the Global AI Landscape

For Western observers, Infinigence's rise highlights several important trends. First, China's AI infrastructure stack is maturing rapidly, with specialized middleware companies emerging to solve real engineering challenges rather than simply copying Western architectures. Second, the concept of a 'Token economy' — where AI compute is measured, traded, and optimized like a commodity — is becoming a shared framework across both U.S. and Chinese markets.

Third, the heterogeneous computing challenge that Infinigence addresses is not unique to China. As Western companies increasingly deploy workloads across NVIDIA, AMD, Intel, and custom silicon (including Google TPUs and Amazon Trainium), the need for chip-agnostic orchestration will only grow. Infinigence's approach may offer a preview of infrastructure patterns that become standard globally.

Looking Ahead: The Road to Token Infrastructure Dominance

With over $97 million in fresh capital, Infinigence is well-positioned to accelerate across all 3 strategic pillars. The immediate priorities likely include expanding its heterogeneous computing support to cover the latest domestic Chinese chips, scaling its enterprise agent platform to production deployments, and deepening partnerships with data center operators like Chindata.

The bigger question is whether Infinigence can establish itself as an indispensable layer in China's AI stack — the equivalent of becoming the operating system for AI compute. If its 'AI Productivity Formula' gains traction as an industry standard for measuring infrastructure efficiency, the company could evolve from a technology provider into a market-defining platform.

For now, the fundraise confirms that investors are betting heavily on the infrastructure layer between chips and applications. In the Token economy, the companies that control how efficiently tokens are produced and consumed may ultimately wield more influence than those that make the chips or build the models themselves.