MetaX Turns Profitable as GPU Shipments Rise, but the Headline Hides a Harder Test

MetaX reported a 612.4 million yuan first-half profit after losing money a year earlier, while revenue climbed 44.7% to 1.32 billion yuan. The results, first highlighted by 36Kr and detailed in the company filing, make rising GPU shipments the central story.

The reversal looks decisive until one accounting distinction enters the frame. MetaX still recorded a 48.9 million yuan loss after excluding nonrecurring items. Its hardware business is approaching operating profitability, but the reported profit does not yet prove that GPU sales alone can sustain earnings.

That gap puts MetaX against a harder opponent than another Chinese chipmaker. It must turn government-backed demand, domestic substitution, and growing shipments into repeatable commercial profit. Nvidia’s software advantage remains the reference point, while Chinese competitors are chasing many of the same buyers.

MetaX’s GPU Shipments Turned Revenue Growth Into a Headline Profit

The important change is not simply that MetaX sold more chips, but that shipment growth brought its core business close to profitability.

MetaX reported first-half revenue of 1.324 billion yuan, up 44.67% from the same period in 2025. The company attributed the increase to broader customer acceptance, continued purchasing, and significantly higher GPU shipments.

The company’s half-year filing provides the clearest account of that shift. It says customers continued adopting MetaX products and services, increasing the volume shipped during the reporting period.

Gross margin reached 57.2%, rising 1.1 percentage points from a year earlier. That matters because shipment growth can destroy value when a hardware company discounts products heavily or carries excessive manufacturing costs.

A rising margin suggests MetaX did not rely entirely on lower prices to expand sales. It retained more gross profit from each unit of revenue, although the filing does not disclose enough detail to separate product mix from pricing.

Second-quarter results offer another useful signal. Revenue reached 762 million yuan during the quarter, exceeding the first quarter’s implied contribution.

Quarterly adjusted net profit was 54 million yuan, according to the figures reported with the results. That represented a sequential reversal from an adjusted loss in the first quarter.

This quarterly progression is more informative than the six-month headline. It indicates that MetaX’s operations crossed into adjusted profitability during the latest quarter, rather than relying only on an annual comparison.

Still, the distinction between reported and adjusted earnings is crucial. First-half net profit attributable to shareholders reached 612.4 million yuan, compared with a 186 million yuan loss one year earlier.

After nonrecurring gains were removed, MetaX reported a 48.9 million yuan loss. That adjusted deficit narrowed by 153.3 million yuan, or 75.83%, from the previous year.

The difference between those figures means most first-half net profit came from items outside recurring operating performance. Investors should therefore resist treating 612.4 million yuan as the current earnings capacity of MetaX’s GPU business.

The more defensible conclusion is narrower. Higher shipments lifted revenue, protected gross margin, and brought recurring operations close to break-even.

That conclusion still represents real progress. MetaX entered public markets after years of high research spending and losses, alongside peers pursuing expensive alternatives to imported computing hardware.

Its listing documents described a market where domestic GPU suppliers faced long verification cycles, software migration barriers, and persistent research costs. Those constraints have not disappeared.

Customers running AI infrastructure rarely replace accelerators as if they were interchangeable components. They must test models, software libraries, networking, servers, drivers, and operational tools before approving large deployments.

A shipment increase therefore carries more meaning than a simple unit sale. It suggests that additional MetaX products passed enough technical and procurement checks to reach customer environments.

The remaining question is whether those deployments become durable, repeat purchases. That determines whether the latest quarter marks a lasting operating turn or a temporary concentration of orders.

The Profit Puts Pressure on China’s Other GPU Contenders

MetaX has raised the commercial benchmark for domestic GPU vendors, but it has not escaped their shared dependence on large infrastructure buyers.

China’s domestic accelerator market includes several technical routes. MetaX, Moore Threads, Biren Technology, and Iluvatar CoreX develop GPUs or GPU-like computing platforms for AI workloads.

Cambricon, Huawei, and other suppliers compete through architectures designed around AI acceleration. These products can serve overlapping training and inference requirements, even when their designs differ from general-purpose GPUs.

The competitive pressure is no longer limited to performance claims. Suppliers must demonstrate shipments, recurring orders, workable margins, and software compatibility across real deployments.

MetaX’s revenue growth places it among the domestic vendors showing measurable commercial scale. A market comparison described diverging first-half results among Chinese chip designers, with MetaX producing one of the clearest profit reversals.

That performance pressures rivals to show that customer trials are becoming production purchases. Announced partnerships and benchmark results matter less when another vendor can point to expanding revenue and a profitable quarter.

Moore Threads is one of the most direct reference points. Both companies entered the Shanghai Stock Exchange’s Star Market in December 2025 after pursuing public listings during the same domestic GPU investment cycle.

Their strategies are not identical. Moore Threads has emphasized a broad GPU portfolio spanning graphics and intelligent computing, while MetaX concentrates heavily on data-center and AI workloads.

Biren focuses on high-performance accelerators and large computing systems. Iluvatar CoreX also targets training and inference through domestically developed general-purpose GPU products.

These vendors face Nvidia indirectly and one another directly. Nvidia defines customer expectations for performance, developer tooling, and production reliability, even where export controls limit product availability.

Domestic companies compete for buyers seeking alternative supply, local support, or compliance with procurement requirements. They also compete for software engineers, manufacturing capacity, and access to large computing-center projects.

MetaX’s software stack, called MXMACA, is central to that contest. A software stack combines drivers, libraries, compilers, and development tools that allow applications to use the underlying processor.

The company says MXMACA is compatible with software built around the CUDA ecosystem. CUDA is Nvidia’s programming platform and the default environment for much AI development.

Compatibility can reduce migration work, but company descriptions do not establish complete equivalence. Production teams must verify individual models, operators, frameworks, and distributed training configurations.

That verification burden creates a commercial barrier. A chip can perform well in a controlled benchmark while requiring substantial engineering work inside a customer’s actual system.

MetaX reports that its products have reached thousand-card cluster deployments. The company also says its platforms support training and inference across several model architectures.

Those claims show the type of deployment MetaX is pursuing. They do not disclose utilization rates, total cost of ownership, failure frequency, or performance across independently selected workloads.

Competitors can respond by improving software support, lowering migration costs, or targeting specific inference tasks. They can also bundle accelerators with servers, networking, and technical services.

This makes the market less like a simple chip race. It is a contest over complete computing systems and the engineering effort required to keep them productive.

The pressure will intensify if buyers begin comparing deployments through operating outcomes. Useful measures include completed training runs, inference throughput, energy consumption, uptime, and engineering hours spent on migration.

Revenue shows that MetaX is winning orders. It does not yet reveal whether customers receive enough value to standardize future infrastructure around the platform.

Why Higher GPU Shipments Matter More Than the 612 Million Yuan Profit

The operating mechanism is a feedback loop between deployments, software validation, repeat orders, and better absorption of fixed development costs.

GPU development requires substantial spending before meaningful revenue appears. Architecture design, verification, software development, packaging, and production preparation all happen before a customer deploys the finished product.

These costs make scale unusually important. Once a product reaches volume production, higher sales can spread fixed engineering expenses across a larger revenue base.

MetaX’s results show that mechanism beginning to work. Revenue increased faster than many operating costs, while gross margin improved and the adjusted loss narrowed.

The shipment expansion also creates more software feedback. Each deployment exposes missing operators, compatibility issues, performance bottlenecks, and failures that internal testing might not uncover.

Engineers can use those findings to improve drivers, libraries, and deployment tools. Better software then reduces the difficulty of subsequent customer projects.

This cycle matters because Nvidia’s largest advantage is not limited to processor specifications. Developers have spent years building applications, documentation, workflows, and expertise around CUDA.

A domestic GPU vendor must therefore solve two problems. It needs sufficient hardware performance, and it must lower the cost of leaving an established software environment.

MetaX says MXMACA supports migration from mainstream GPU software. Its filing describes compatibility as a core product strength and positions software usability alongside single-card and cluster performance.

Customers will judge that claim through practical work. A model that starts successfully is only the beginning of validation.

Teams must confirm numerical accuracy, performance stability, memory behavior, distributed communication, monitoring, and recovery from hardware failures. They must repeat those checks after framework and model updates.

This challenge grows inside large clusters. Distributed training divides computation across many accelerators, while high-speed interconnects coordinate data movement among them.

MetaX has developed MetaXLink for card-to-card communication. The company says it supports configurations ranging from small groups to larger interconnected systems.

Interconnect performance influences how much theoretical chip capacity becomes useful cluster capacity. Slow communication can leave processors idle while they wait for data from other devices.

This is why higher shipment volume matters beyond revenue. Larger deployments create opportunities to prove that hardware, networking, and software can function together under sustained workloads.

MetaX has reported deployments in public computing platforms, telecom infrastructure, commercial AI centers, and research projects. Its filing also mentions cooperation with Shanghai Unicom and a domestic GPU center in Wuxi.

The company describes applications across finance, healthcare, energy, education, transportation, and media. These examples indicate a broad target market, but they provide limited detail about revenue generated by each sector.

MetaX also reports support for research and emerging applications. One project used its computing resources for cancer research, while another placed a MetaX-based computing payload aboard a satellite.

These projects demonstrate technical range. They should not be confused with evidence that each emerging use case has become a substantial commercial market.

Near-term revenue will probably depend more on data centers, telecom operators, public computing platforms, and large enterprises. These buyers can purchase clusters at a scale that materially affects financial results.

Their procurement can also be uneven. A few large projects may produce rapid revenue growth during one period, followed by slower recognition during another.

That pattern makes shipment quality as important as shipment quantity. A diversified base of repeat buyers would provide stronger evidence than a small group of large, project-based customers.

The balance sheet gives additional context. MetaX held 1.43 billion yuan of inventory at June 30, down from 1.50 billion yuan at the end of 2025.

Within that total, finished goods increased, while raw materials declined. Semi-finished products also rose, reflecting continued manufacturing activity and the timing of production.

Inventory remains large relative to first-half revenue. That is not automatically a warning because chip supply planning requires long lead times and advance commitments.

However, it increases the importance of demand forecasting. Products can lose value quickly when architectures change, customers delay projects, or newer chips enter production.

MetaX recorded almost 197 million yuan of inventory impairment reserves at the end of June. The figure shows that management already recognizes some risk within the inventory position.

The commercial mechanism is therefore clear but demanding. More shipments can improve margins, strengthen software, and create reference deployments.

The same expansion can also raise working-capital exposure and increase dependence on continued procurement. Scale helps only when products remain competitive and customers keep ordering.

What the Profit Figure Does Not Show

MetaX has demonstrated improving operations, but its adjusted loss, inventory exposure, and software validation burden prevent a clean profitability claim.

The first limitation is accounting quality. Reported net income reached 612.4 million yuan, yet adjusted operations remained 48.9 million yuan in the red.

Nonrecurring gains are legitimate accounting items, but they do not provide the same evidence as profit generated through repeated product sales. They should not anchor expectations for future earnings.

The second quarter provides a stronger signal because adjusted profit reached 54 million yuan. Even so, one profitable quarter cannot establish a durable earnings pattern.

Large hardware orders often follow project schedules. Revenue recognition can cluster around delivery, acceptance, or commissioning milestones.

A stronger test would require several quarters of adjusted profit alongside continued revenue growth. Stable gross margin would add confidence that competition has not forced aggressive discounting.

The second limitation is customer concentration. MetaX’s pre-listing disclosures showed that a large share of sales came from a limited group of customers.

This pattern is common among young infrastructure suppliers. A single computing-center contract can represent a meaningful portion of annual revenue.

Concentration accelerates growth when those customers expand. It also exposes the supplier to budget changes, project delays, and bargaining pressure.

Investors should watch whether MetaX reports a broader customer base or repeat purchasing across multiple sectors. Named partnerships alone do not answer that question.

The third limitation is software maturity. MetaX says its platform supports migration from the dominant international GPU ecosystem, but independent evidence remains limited.

Software compatibility is not binary. A framework can run while still requiring custom engineering, unsupported operations, or performance tuning for important models.

Fast-moving AI workloads make this problem harder. New model structures, quantization methods, attention implementations, and distributed techniques appear frequently.

A domestic platform must keep pace without breaking existing deployments. That requires sustained research spending even after hardware begins generating revenue.

MetaX held 426 authorized Chinese patents at June 30, including 414 invention patents, according to its filing. Patents demonstrate engineering output but do not measure customer usability.

The more relevant commercial measures are migration time, supported workloads, deployment stability, and customer renewal. MetaX does not publish enough standardized data to compare those measures across vendors.

The fourth limitation is competition from several directions. Domestic GPU companies are not the only alternatives for Chinese buyers.

ASIC suppliers build processors optimized for narrower AI tasks. An ASIC, or application-specific integrated circuit, sacrifices some flexibility to improve efficiency for defined workloads.

MetaX itself expects GPUs and ASICs to coexist. GPUs can adapt to changing models, while specialized chips can offer better efficiency in stable applications.

This creates a two-sided challenge. MetaX must narrow the usability gap with Nvidia while defending suitable inference workloads against specialized domestic accelerators.

Export controls provide demand for local alternatives, but they do not eliminate technical competition. Buyers still compare performance, energy use, software support, availability, and lifecycle risk.

The fifth limitation involves procurement incentives. China’s push for technological self-sufficiency supports domestic chip adoption through policy, investment, and infrastructure construction.

That environment can accelerate validation and create early demand. It can also make it difficult to separate market-led purchasing from policy-led deployment.

The distinction matters for long-term economics. A sustainable supplier needs products that customers continue buying because they solve operational problems at an acceptable total cost.

Policy demand can help a platform reach the scale needed to improve. It cannot permanently substitute for reliability, developer adoption, and competitive system performance.

MetaX’s IPO review coverage documented the heavy cumulative losses that preceded its listing process. Those losses reflected the long development cycle shared by domestic GPU companies.

Public capital gives MetaX more room to fund new products and software. It also increases pressure to convert technical progress into recurring returns.

The company’s financial position provides time. At midyear, it reported substantial cash and financial assets, supported partly by proceeds raised during its public listing.

Yet the core test remains unchanged. MetaX must show that recurring GPU economics, rather than isolated gains or favorable financing conditions, can support the business.

None of these risks erases the first-half improvement. Revenue growth, a higher gross margin, and second-quarter adjusted profit are meaningful operational evidence.

They simply support a more precise judgment. MetaX has approached recurring profitability, but it has not completed the transition.

Three Signals Will Show Whether MetaX’s Turnaround Lasts

The next phase will be decided by adjusted profit, repeat GPU orders, and evidence that MetaX software reduces real migration costs.

The first signal is another quarter of adjusted profitability. This measure removes many items that can make statutory earnings appear stronger than normal operations.

If MetaX remains profitable after those exclusions, the second-quarter result will look like the start of a durable turn. A return to adjusted losses would weaken that interpretation.

Gross margin should be read alongside adjusted earnings. Stable or rising margin would suggest that MetaX can grow without relying on discounts that undermine operating leverage.

The second signal is the composition of future orders. Revenue growth becomes more persuasive when existing customers expand deployments and new customers reduce concentration.

Repeat purchasing would indicate that buyers found sufficient value after completing technical validation. It would also suggest that MetaX hardware is supporting sustained production workloads.

New customer additions matter for a different reason. They would reduce exposure to individual infrastructure projects and strengthen MetaX’s negotiating position.

Watch inventory and receivables as supporting indicators. Inventory that rises faster than demand can signal optimistic production planning, while expanding receivables can slow cash conversion.

MetaX ended June with 969.2 million yuan in accounts receivable, up from 725.7 million yuan at the end of 2025. The increase deserves attention as sales expand.

Higher receivables can follow ordinary revenue growth. They become concerning when collection slows or when revenue depends on a few customers with long payment cycles.

The third signal is independent software and cluster validation. MetaX needs evidence that MXMACA supports current AI workloads without excessive migration work.

Useful disclosures would include broader benchmark methodology, supported model lists, customer deployment reports, and measured cluster availability. Results selected and published only by the vendor carry less weight.

Large-scale repeat deployments would provide particularly strong evidence. Customers rarely expand a cluster when software instability or engineering costs outweigh the hardware’s benefits.

The competitive response will also matter. Moore Threads, Biren, Iluvatar CoreX, Cambricon, and Huawei will continue improving products and software.

Their progress can weaken MetaX even if overall demand for domestic processors rises. Buyers can divide orders across suppliers or select different chips for training and inference.

MetaX therefore needs more than an expanding market. It needs a defensible position within that market, supported by software, system performance, and customer retention.

For developers, this contest affects which toolchains become practical outside Nvidia’s environment. More mature alternatives can increase hardware choice, but fragmented platforms can multiply testing work.

Enterprise buyers face a related calculation. A domestic accelerator can improve supply options, yet migration expenses and long-term software support may outweigh a lower acquisition barrier.

Technical teams evaluating MetaX should document compatibility at the workload level. They should test the exact models, frameworks, precision formats, and distributed configurations intended for production.

They should also preserve benchmark results, driver versions, unresolved issues, and vendor commitments in a searchable technical knowledge base. This record makes later purchasing decisions easier to audit.

The first-half report changes MetaX’s position in China’s GPU race. The company is no longer defined only by investment, product road maps, and accumulated losses.

It now has rising shipments, more than 1.3 billion yuan in half-year revenue, and one quarter of adjusted profit. Those results establish commercial momentum.

The 612.4 million yuan headline should still be handled carefully. Adjusted results show that recurring operations remained slightly unprofitable across the full six months.

The next reports will reveal whether MetaX can repeat the second quarter without sacrificing margin or accumulating financial strain. That is the threshold between a promising shipment surge and a sustainable GPU business.

Readers following MetaX should ask three direct questions. Does adjusted profit remain positive, do customers place larger repeat orders, and does independent deployment evidence validate the software claims?

If all three answers improve together, MetaX’s turnaround will look operational rather than accounting-driven. If they diverge, the first-half profit will remain an encouraging but incomplete milestone.