Yang Guang Bets 980 Million Yuan on an AI Data Center Despite Mounting Losses

Yang Guang plans to invest up to 980 million yuan in an AI computing center, despite losses and limited experience operating technology infrastructure.

The Shenzhen-listed property company disclosed the plan through its controlled subsidiary, Shenzhen Yang Guang Digital Technology. The facility would be built in Xingning, a county-level city within Meizhou, Guangdong Province.

According to the initial project report, commercial operations are scheduled to begin in October 2027. Funding would come from internal resources, external financing, or a combination of both.

The proposal puts a striking mismatch at the center of the story. Yang Guang still describes commercial property operations and property leasing as its main businesses. It is now considering a capital-intensive computing project larger than several years of recent revenue.

That contrast matters more than the announced investment ceiling. An intelligent computing center is a data center designed for accelerated AI workloads, usually involving specialized processors, high-speed networks, storage, cooling, and substantial electricity capacity.

Building the physical site is only the first test. The operator must also secure equipment, customers, reliable energy, technical staff, and enough utilization to cover continuing costs.

Yang Guang has already warned investors about approvals, financing, unfamiliar operations, and continuing losses. Those qualifications turn the announcement into a test of execution, rather than a simple expansion story.

The central contest is therefore clear: Yang Guang’s promise of a new computing business versus the financial and operational reality of entering one from outside the industry.

What Yang Guang Has Actually Approved

Yang Guang has authorized a large investment plan, but it has not yet demonstrated a functioning AI infrastructure business.

The proposed facility would sit in Xingning, within the northeastern part of Guangdong. Yang Guang’s controlled subsidiary would serve as the investment vehicle.

The announced budget is a ceiling of 980 million yuan. “Up to” is important because it does not mean the entire amount has been committed, financed, or spent.

The target date is also forward-looking. The company expects the center to enter formal operation in October 2027, leaving more than a year for approvals, financing, procurement, construction, testing, and customer preparation.

The disclosure does not establish that every required approval has been received. Yang Guang explicitly included project approval among its stated uncertainties.

That distinction separates a board-approved proposal from an operating asset. Data center projects commonly pass through planning, energy, construction, environmental, network, and equipment procurement stages before producing revenue.

The announcement also leaves several commercial questions unanswered. Publicly available reporting does not identify committed customers, signed capacity contracts, expected utilization, or the project’s planned computing configuration.

It does not provide a detailed split between buildings, power systems, cooling, servers, accelerators, networking, storage, and software. Without that allocation, readers cannot determine how much of the budget would create saleable computing capacity.

The financing structure remains open as well. Yang Guang says the project would use its own funds or external financing, but that description does not reveal borrowing terms or lender commitments.

For a profitable infrastructure operator, flexible financing language might attract limited attention. For a company reporting continuing losses, it becomes one of the central variables.

Yang Guang’s corporate position makes the scale more notable. A May 2026 filing said commercial operations and property leasing remained the company’s primary businesses.

That market disclosure reported first-quarter revenue of about 62.2 million yuan. It also reported a net loss attributable to shareholders of roughly 23.5 million yuan.

The same filing placed total assets near 4.31 billion yuan and shareholder equity around 2.05 billion yuan at March 31, 2026. The proposed project ceiling equals almost half that reported equity.

The comparison does not prove that the project is unaffordable. Financing can spread capital requirements, and spending could occur across several construction stages.

It does show that the proposal is material relative to Yang Guang’s financial base. Investors therefore need more than a headline investment number to evaluate the commitment.

The company’s recent technology acquisitions provide another piece of context. In May, Yang Guang paid 100,000 yuan for controlling stakes in two recently established Shenzhen companies.

Those targets had very small operations at the acquisition date. Yang Guang said their existing scale would not change its main business or materially affect revenue and profit.

The proposed Xingning center represents a much larger step. It moves the technology strategy from small corporate vehicles toward physical infrastructure with substantial capital requirements.

That is what changed. Yang Guang is no longer only exploring a digital business through small subsidiaries. It is proposing a facility large enough to test its balance sheet and management capacity.

Why the AI Data Center Bet Is Happening Now

National demand for AI infrastructure creates a favorable narrative, but policy support does not guarantee profitable demand for an individual center.

China continues to prioritize intelligent computing capacity as part of its broader AI strategy. National policy calls for larger computing clusters and better coordination among computing, electricity, data, and network resources.

The State Council’s AI policy directs authorities to improve the national computing network. It also supports standardized, scalable computing services that can match supply with demand.

That policy direction helps explain why companies outside traditional cloud infrastructure see an opening. AI training and inference require specialized capacity, while regional governments want technology investment and new digital industries.

Local and national plans also encourage cities to build computing infrastructure within coordinated networks. The objective is not simply to place more servers in more buildings.

Policymakers increasingly emphasize resource scheduling, efficiency, renewable electricity, and practical applications. Those requirements favor projects that can connect power, networks, customers, and software services.

Guangdong offers one obvious advantage: proximity to technology demand in Shenzhen and the wider Greater Bay Area. Meizhou may offer different land and operating conditions than the largest coastal cities.

However, proximity within one province does not automatically produce low-latency access, adequate transmission capacity, or customer contracts. Those outcomes depend on network architecture and the workloads being served.

Some AI tasks tolerate distance. Model training, batch processing, rendering, and other asynchronous workloads can run outside major business centers when network and scheduling systems work well.

Other tasks need tighter latency and service guarantees. Interactive inference, financial applications, industrial control, and real-time consumer services can impose different location requirements.

Yang Guang has not publicly described which workloads will anchor the Xingning project. That missing detail makes it difficult to judge whether the location matches the intended demand.

National infrastructure development also increases competition. China’s major telecommunications operators, cloud providers, specialist data center companies, and local state-backed platforms already operate computing networks.

China Mobile, for example, said its national computing infrastructure included more than 10,000 accelerator cards and over 60 EFLOPS of intelligent computing capacity. An EFLOPS represents one quintillion floating-point operations each second.

That national network also combines computing resources with backbone connectivity and centralized scheduling. Such integration creates a higher competitive bar than constructing a standalone machine room.

Regional programs are expanding too. Guizhou reported more than 150 EFLOPS of computing capacity in 2025, with intelligent computing representing over 90 percent.

The province also reported 50 data centers built or under construction. Its fixed-asset investment in computing facilities exceeded 22 billion yuan during 2025.

Those figures show strong infrastructure demand. They also illustrate how quickly capacity can accumulate when governments, carriers, and private operators pursue the same opportunity.

Yang Guang is therefore entering a market with policy momentum and abundant capital. That combination can support demand, but it can also produce excess capacity in poorly positioned locations.

The company must identify a specific customer problem that existing operators do not already solve. Lower land costs alone would not establish a durable advantage.

A credible strategy might involve regional enterprises, public-sector workloads, specialized model deployment, or partnerships with established infrastructure providers. No such anchor has been publicly confirmed.

The timing also reflects pressure within Yang Guang’s existing business. Property leasing and commercial operations have not generated enough momentum to restore consistent profitability.

A technology project can offer a new growth narrative. Yet the need for a new narrative does not make the economics of that project easier.

The market should judge the center by contracted demand, deliverable capacity, energy access, and operating returns. Broad AI adoption cannot substitute for those project-level measures.

The Promise Versus Reality Behind Yang Guang’s Computing Pivot

Yang Guang is presenting a path into AI infrastructure while its current financial results make every execution error more consequential.

The company’s 2026 half-year forecast sharpens this tension. Yang Guang expects a net loss attributable to shareholders between 43.5 million and 58 million yuan.

Its earnings forecast also projects an adjusted loss between 48.5 million and 65 million yuan. The comparable shareholder loss was about 39.2 million yuan one year earlier.

Yang Guang attributed the deterioration to lower income from property leasing and commercial operations. It also cited increased losses from an associate.

The estimate had not been pre-audited when published. Final figures were due in the company’s half-year report, so the forecast should not be treated as completed financial statements.

Even with that qualification, the direction is clear. Yang Guang is considering its largest visible technology commitment while its established operations remain under pressure.

This does not automatically make diversification irrational. Companies often invest during weak periods because their existing business offers limited growth.

The concern is sequencing. A new operator must spend before construction produces capacity, then attract customers before that capacity produces stable cash flow.

External financing can reduce the immediate use of internal cash. It also introduces interest costs, repayment obligations, covenants, and refinancing exposure.

Internal funding avoids some lender constraints. However, it can reduce liquidity available for existing properties, operating losses, maintenance, and other commitments.

A joint venture or phased procurement might distribute risk. The disclosed plan does not yet provide enough detail to assess whether Yang Guang will use either structure.

The company must also bridge a substantial operating gap. Property management and data center operations share some knowledge about physical assets, contracts, and facilities.

They differ in critical areas. AI infrastructure requires specialized procurement, high-density electrical design, cooling engineering, cybersecurity, hardware maintenance, and workload scheduling.

Accelerator supply introduces another challenge. The value of an intelligent computing center depends heavily on which processors it can deploy and which software stacks customers can use.

Hardware choices also affect network design, power density, cooling, maintenance, and depreciation. A procurement delay can therefore reshape the entire project schedule.

Technology changes faster than the physical shell. Equipment ordered too early can age before the facility reaches stable utilization.

Equipment ordered too late can delay customer acceptance and revenue. Operators must balance availability against obsolescence throughout construction.

Customer concentration adds another risk. One large contract can support early utilization, but losing that customer can leave expensive equipment idle.

A diversified customer base reduces dependence. It also requires stronger sales, support, security, billing, and capacity management capabilities.

Yang Guang has not disclosed committed tenants or computing customers. It also has not published expected occupancy, revenue, operating margin, or payback assumptions.

That absence does not show that customers are unavailable. It means outsiders cannot yet test the commercial case.

The company’s own warnings recognize this uncertainty. It has cited cross-industry operation as a risk, effectively acknowledging that corporate control does not equal technical competence.

Execution will depend heavily on partners and hires. Investors should therefore watch for named engineering contractors, equipment providers, network operators, and commercial customers.

Experienced partners could lower construction and operating risk. They would not eliminate financing risk or guarantee profitable utilization.

The planned October 2027 opening also creates a long exposure window. Funding conditions, equipment supply, policy rules, customer demand, and computing technology can change before launch.

The project’s size amplifies those uncertainties. A small pilot could validate demand with limited capital, while a 980 million yuan ceiling signals a more substantial commitment.

The strongest interpretation is that Yang Guang sees an opportunity large enough to justify a major pivot. The skeptical interpretation is that AI infrastructure offers an attractive story during a difficult earnings period.

Both interpretations remain plausible because the announcement lacks operational proof. Contracts and financing milestones will decide which one becomes more convincing.

Financing, Power, and Utilization Are the Real Constraints

The project succeeds only if Yang Guang converts construction spending into consistently used computing capacity at sustainable operating costs.

Data centers are sometimes described like ordinary property developments. That comparison understates how much their economics depend on equipment cycles, electricity, networks, and customer utilization.

A completed building does not constitute a productive AI center. The asset needs energized racks, usable processors, functioning software, network access, security controls, and paying workloads.

Utilization is especially important because many costs continue when demand is weak. Operators still carry depreciation, staffing, maintenance, financing, and baseline energy expenses.

Headline computing capacity can also mislead. Different chips, numerical formats, workloads, and software configurations can produce performance figures that are difficult to compare.

Yang Guang has not disclosed a capacity figure or benchmarking method. Readers should resist calculating implied value from the budget alone.

Power availability will shape both schedule and economics. High-density computing equipment requires reliable electricity delivery and cooling systems designed for concentrated heat.

China’s data center policies increasingly connect computing development with energy efficiency and renewable power. The National Data Administration has emphasized coordinated planning between computing and electricity.

Its green power guidance targeted an 80 percent green-electricity share for new data centers within national hub nodes during 2025.

The Xingning project is not necessarily governed by that specific hub target. The policy still shows the direction of regulatory and operating expectations.

Yang Guang will need to clarify available power capacity, grid connection timing, redundancy, renewable sourcing, and expected energy efficiency. None of those details appears in the initial report.

Cooling choices matter as well. Conventional air cooling remains useful, but denser AI systems increasingly require more advanced thermal management.

Liquid cooling can improve heat removal for high-density hardware. It can also introduce different design, maintenance, supply-chain, and operational requirements.

The company has not said which cooling architecture it plans to use. That prevents meaningful comparison with newer facilities designed around dense accelerator clusters.

Financing is the second constraint. The project’s maximum investment exceeds Yang Guang’s latest annual revenue by a wide margin.

The company reported 2025 revenue of roughly 341.5 million yuan in its May filing. It also reported a 2025 net loss attributable to shareholders near 218.2 million yuan.

Those historical results do not establish the project’s funding plan. They do explain why lender terms and capital contributions deserve close attention.

A financing announcement should reveal more than the total facility size. Investors need to know interest rates, maturities, collateral, guarantees, repayment schedules, and conditions for drawing funds.

They also need to understand whether spending is conditional on approvals or customer commitments. Conditional phases can protect liquidity when demand or permits arrive later than expected.

Unconditional procurement concentrates risk earlier. That approach can secure scarce equipment, but it can leave the company exposed if construction or customer onboarding slips.

The third constraint is commercial utilization. Yang Guang needs customers whose workloads fit the location, hardware, network, and compliance environment.

Possible users include model developers, regional enterprises, research organizations, manufacturers, and public institutions. These are potential categories, not confirmed customers for this project.

Each group has different requirements. Model training can demand large clusters and fast internal networks, while enterprise inference may prioritize reliability, security, and flexible consumption.

Public-sector workloads can involve procurement procedures and data governance. Industrial customers may need integration with factories, sensors, or private networks.

Selling capacity therefore requires more than announcing available compute. Yang Guang must package infrastructure into services that customers can buy and operate safely.

Large incumbents already provide cloud platforms, managed clusters, network connectivity, and developer ecosystems. A new entrant must compete on location, specialization, service, partnerships, or cost.

Yang Guang has not identified that differentiator. Until it does, the project remains an infrastructure proposal searching for a clearly documented market position.

The company could reduce this risk by disclosing binding capacity reservations before major equipment purchases. It could also establish partnerships with experienced operators.

Independent technical validation would help. So would reporting standardized capacity, energy efficiency, available network routes, and the percentage of equipment backed by customer agreements.

The central tradeoff is not simply growth versus caution. It is the potential value of entering AI infrastructure versus the cost of learning through a large live project.

Yang Guang’s property experience may support site development and asset management. The decisive capabilities lie in areas that its historical business has not yet demonstrated.

Three Signals Will Decide Whether the Plan Becomes a Business

Approvals, committed financing, and contracted utilization will provide the clearest evidence that Yang Guang’s proposal is advancing beyond an AI investment narrative.

The first signal is regulatory and construction progress. Yang Guang should identify major approvals, grid access, land arrangements, design milestones, and the start of physical work.

These events would strengthen confidence in the October 2027 schedule. Repeated delays or vague progress reports would weaken it.

Approval risk deserves attention because the company listed it explicitly. A project timeline should therefore be measured from verified milestones, not only from the announced target date.

Investors should also examine whether the final approved scope matches the 980 million yuan ceiling. A smaller first phase might represent disciplined execution rather than failure.

Conversely, an expanding budget would raise questions about procurement, design changes, and financing capacity. The reasons for any revision would matter more than the revision alone.

The second signal is a detailed funding package. Yang Guang needs to show which entity supplies the capital and what obligations accompany it.

A credible package would identify committed amounts, funding dates, borrowing conditions, collateral, and the subsidiary’s required contribution. It should also explain how existing operations retain adequate liquidity.

Financing backed by construction progress or signed customers would reduce some risk. Expensive debt secured against core property assets could transfer project uncertainty into the established business.

The market should distinguish lender willingness from project validation. A loan can fund construction without proving that the center will achieve sufficient utilization.

Third, Yang Guang needs commercial evidence. Signed capacity contracts, named anchor customers, or credible reservations would connect the facility to actual demand.

The quality of those contracts matters. Investors should look for duration, cancellation rights, minimum usage, payment protections, and any capital obligations placed on Yang Guang.

Short memoranda or broad cooperation agreements carry less weight than enforceable service commitments. Customer names alone also reveal little without commercial terms.

Utilization should become the primary operating metric after launch. Installed capacity, energized capacity, contracted capacity, and used capacity are separate measures.

Reporting only the largest number could obscure weak demand. Yang Guang should disclose each stage consistently and explain the measurement basis.

Revenue quality will matter alongside utilization. A facility can show high activity while producing poor margins if promotional contracts or energy costs absorb most income.

Cash collection is another test. Long receivable periods would be particularly concerning for a company already managing losses in its established operations.

The October 2027 target remains far enough away for conditions to change. Hardware generations will advance, competitors will add capacity, and customer preferences will evolve.

Yang Guang will need to update its design without allowing technology changes to destabilize procurement. That balance is difficult even for experienced operators.

The company should also disclose its operating model before launch. It could run the center directly, hire a specialist, lease capacity wholesale, or combine these approaches.

Each model distributes risk differently. Direct operation offers more control but demands deeper technical and commercial capabilities.

A specialist operator can supply expertise, though fees and shared economics reduce Yang Guang’s returns. Wholesale leasing simplifies sales but can create customer concentration.

These choices should become visible well before the facility begins formal operation. If they remain unresolved near the planned launch, the schedule will look increasingly fragile.

For developers and enterprise buyers, the project matters because regional capacity can broaden infrastructure options. More options can improve availability, workload placement, and negotiating leverage.

Yet buyers should evaluate delivered service, not construction budgets. Hardware compatibility, network performance, security controls, support, and contractual reliability determine whether capacity is useful.

Knowledge workers tracking this development can preserve filings, financing updates, and technical claims in a searchable AI knowledge base. The timeline matters because individual announcements reveal only part of the risk.

Yang Guang has made a consequential opening move, not completed a transformation. The company has placed a 980 million yuan ceiling and an October 2027 target around that ambition.

The next evidence must be harder: permits, committed capital, credible partners, customer contracts, and measurable utilization.

Watch those three signals in order. Confirmed approvals show the facility can be built, financing shows it can be funded, and binding demand shows it deserves to operate.

If Yang Guang supplies all three, the project will look like a disciplined entry into AI infrastructure. If it supplies only announcements, the gap between promise and reality will keep widening.