China’s AI Chip ‘Little Dragons’ Are Growing Fast, but Durable Profits Remain Elusive

China’s best-known domestic AI-chip challengers—MetaX, Moore Threads, Biren and Enflame—are no longer simply promising design houses. In 2026, all four are demonstrating meaningful revenue momentum, and public-market financing is becoming part of the industry’s growth model. Yet their commercial maturity should be judged less by a single revenue surge or statutory profit than by whether they can turn deployments into diversified, repeatable orders and sustainable operating earnings.

The short version: sales are scaling faster than profits

The group’s first-half figures point to a sharp increase in demand for domestic AI compute:

• MetaX reportedly posted first-half revenue of about RMB 1.32 billion and a statutory profit. Its adjusted result nevertheless remained negative, an important warning that non-recurring items and core operating performance can tell different stories.
• Moore Threads reported RMB 1.736 billion in first-half revenue, up 147.4% year over year. Its attributable loss narrowed substantially, but it remained loss-making on a recurring basis.
• Biren forecast first-half revenue of RMB 1.15 billion to RMB 1.30 billion, compared with RMB 58.9 million a year earlier. That exceptional percentage growth reflects both stronger sales and a very low prior-year base; it does not, by itself, establish a durable profit model.
• Enflame remains in investment mode. Its own prospectus history shows revenue rising from RMB 301 million in 2023 to RMB 990 million in 2025, while adjusted losses persisted. 7

These are meaningful commercial signals. But semiconductor companies must fund successive chip generations, systems engineering, software tools and customer support long before a revenue ramp can reliably cover the cost base.

Why MetaX’s profit needs context

MetaX’s reported first-half profit is a milestone, but the continuing adjusted loss matters more for assessing commercial maturity. Statutory net income can include items that do not recur in ordinary chip sales and services. Adjusted earnings are therefore a more useful test of whether gross profit, R&D and operating costs have reached a self-sustaining balance.

The next proof point is not merely another profitable reporting period. It is evidence that MetaX can maintain product shipments, software support and repeat customer orders while continuing to invest in its GPU roadmap.

Moore Threads and Biren: a revenue ramp, not a finished business model

Moore Threads’ first-half revenue growth and narrower loss suggest that larger sales volumes may be improving operating leverage. Its challenge is to preserve that trajectory as it spends on new products and expands deployment support.

Biren’s projected leap in revenue is even more dramatic in percentage terms. However, a rapid ramp from a small base can be uneven. For both companies, the important questions are the same: how broadly demand is distributed across customers, whether orders repeat after initial trials, and whether software and systems costs decline as deployments grow.

Enflame’s IPO shows both progress and capital intensity

Enflame is the final member of the group to move toward the public market. It set an IPO price of RMB 142.18 per share and expected to raise about RMB 6.1 billion, or roughly $908 million, through an offering of approximately 43 million shares on Shanghai’s STAR Market. 1

The proposed use of funds is revealing. Enflame planned to allocate roughly RMB 3.3 billion to AI hardware-software collaborative innovation, RMB 1.5 billion to fifth-generation chips and RMB 1.2 billion to sixth-generation chips. 10 This is a commercial milestone, but also a reminder of the capital required to compete in AI infrastructure.

Enflame’s financial record reinforces that point. Revenue reached RMB 990 million in 2025, while the company still reported a substantial adjusted loss; it has said it expects profitability in 2026 or 2027. 713

Two technical paths: GPU compatibility versus a specialized accelerator stack

The four companies are not pursuing identical product strategies.

MetaX, Moore Threads and Biren are generally associated with general-purpose GPU platforms and software environments intended to reduce migration friction for developers familiar with Nvidia’s CUDA-centric workflows. Their commercial task is to make porting, optimization and cluster operations sufficiently straightforward that customers can use domestic hardware without accepting a large productivity penalty.

Enflame takes a more specialized accelerator approach, built around its own hardware and TopsRider software stack. A purpose-built design can be attractive for targeted AI training and inference workloads, but its success depends heavily on tooling, framework support, performance tuning and customer confidence in the surrounding ecosystem.

In both approaches, software is not an accessory. It is part of the product.

Domestic demand is creating an opening—but not removing the hard parts

China’s push for technology self-sufficiency and restrictions affecting access to Nvidia’s most capable AI products have strengthened incentives for cloud providers, enterprises and AI developers to evaluate domestic accelerators. That opening is helping turn pilot projects into larger commercial opportunities.

But the companies still face the difficult elements of platform competition: chip performance, supply-chain execution, high-bandwidth memory and interconnect availability, compiler and framework compatibility, reliability at cluster scale, and the depth of developer support. Fast top-line growth can coexist with high R&D requirements and low or negative adjusted earnings.

What would demonstrate real commercial maturity?

The four companies are closest to commercial maturity when several conditions appear together:

1. Repeat, diversified customer orders rather than a small number of large deployments.
2. Positive adjusted earnings, not only statutory profits or narrower losses.
3. Reliable cluster-scale delivery, including systems integration and after-sales support.
4. A durable software ecosystem that lowers migration and optimization costs for developers.
5. A funded roadmap for successive generations without relying indefinitely on external capital.

The 2026 evidence supports a cautious conclusion: China’s AI-chip “little dragons” are becoming credible commercial suppliers, with revenue growth and capital-market access to show for it. The decisive phase is still ahead—converting demand created by a strategic market opening into profitable, repeatable and ecosystem-backed businesses.