Hong Kong’s equity market delivered a clear signal this week about how traders are valuing artificial intelligence companies: user growth alone is no longer enough. Shares of MiniMax, a major AI model developer with a heavy consumer-app business, fell more than 20% during intraday trading on July 9, cutting its market value to about HK$90 billion. That was far below its March peak of HK$410 billion. Zhipu, by contrast, rose 13% a day earlier and added another 11% in the following session, keeping its market capitalization near HK$900 billion.
The sharp divergence has turned into one of the clearest market judgments yet on the AI business model. Traders are increasingly rewarding companies that can convert model usage into enterprise revenue, productivity gains, and recurring software contracts. They are punishing platforms that rely on mass consumer traffic but struggle to cover the rising cost of computing power behind every prompt, video, image, or chatbot conversation.
The combined market-capitalization gap between Zhipu and MiniMax now stands at roughly HK$800 billion. That gap reflects more than short-term trading. It shows a deeper shift in how the AI economy is being priced. The earlier belief that attention, downloads, and user engagement would automatically translate into value is being tested by a harsher reality: every AI interaction carries a real compute cost.
For consumer AI products, that cost can rise faster than advertising revenue, subscriptions, or in-app payments. For enterprise AI platforms, the same computing burden may be easier to absorb if customers use the technology to improve coding, automate workflows, reduce labor costs, or manage industry-specific operations.
a reversal in market expectations
Zhipu and MiniMax entered the public market with very different narratives.
Zhipu listed in January and posted a modest first-day gain of 13.17%, closing with a value of about HK$55.5 billion. At the time, many traders viewed its business as slower and more capital intensive. The company generated 73.7% of its revenue from customized enterprise deployments across sectors including energy, finance, and government. These contracts could be large, but they required deep integration, sales cycles, and technical support.
MiniMax went public the following day and received a much warmer market reception. Its shares more than doubled on debut, closing up 109% and lifting its value above HK$105 billion. The company drew enthusiasm because about 67% of its income came from consumer-facing products, including international applications such as Talkie and the video-generation tool Hailuo AI. By March, MiniMax’s market value had surged to HK$410 billion, briefly placing it in the same valuation discussion as some major internet companies.
That early gap has now reversed. Zhipu has become the preferred name among traders seeking enterprise AI exposure, while MiniMax has become a test case for the limits of consumer AI monetization.
The turning point came in the first half of 2026, when both companies changed pricing for their AI services. Zhipu raised its API rates three times between February and April, increasing prices by a cumulative 83%. The increase did not weaken demand. Instead, API usage rose 400%, while annual recurring revenue reached 1.7 billion yuan by March, about 60 times the level recorded a year earlier.
MiniMax followed a different path. On June 1, it launched its M3 model with prices nearly double those of the previous version. Within a week, it reversed course and cut charges by 50% after demand weakened. Days later, a major U.S. bank reduced its target price for MiniMax to HK$400 from HK$1,100, identifying pricing power as an important measure of competitiveness in the AI sector.
pricing power becomes the key test
The market reaction suggests that traders are focusing less on model rankings and more on pricing durability. In the current AI cycle, a company’s ability to charge more without losing usage has become a critical sign of whether customers see the product as essential.
Zhipu’s price increases sent a stronger message than its early listing performance. Higher API fees, combined with rising usage, suggested that developers and enterprise clients continued to rely on its tools even at higher cost. That gives the company room to build revenue from software, coding tools, model access, and customized integration.
MiniMax’s rapid price reversal sent the opposite signal. If customers reduce usage when prices rise, the platform may have limited ability to pass rising compute costs on to users. That is a problem in consumer AI, where each interaction can require significant graphics-processing capacity and cloud spending.
MiniMax’s overall consumer gross margin stood at 4.7%. Its 3 million monthly active users generated about US$5 each on average. In 2025, the company recorded revenue of roughly US$79 million while reporting a net loss of about US$250 million. Those figures illustrate the pressure facing consumer AI companies: large audiences can produce large costs before they produce sustainable earnings.
The issue is not limited to MiniMax. Doubao, a major domestic AI app with 345 million monthly users, faced annual compute costs running into the tens of billions of yuan before introducing paid subscriptions in June 2026. Its subscription tiers ranged from 68 yuan to 500 yuan per month. That move reinforced a broader conclusion spreading across the sector: free consumer AI services are becoming commercially difficult to sustain at scale.
enterprise AI gains an advantage
Zhipu’s current advantage comes from its focus on enterprise software, developer infrastructure, and industry-specific deployments. These areas can support higher prices because customers are often measuring AI spending against productivity gains, labor savings, coding speed, or operational efficiency.
In 2025, Zhipu recorded revenue of 724 million yuan and a net loss of 4.718 billion yuan. Its current valuation near HK$900 billion implies a high price-to-sales multiple, and it still faces significant financial pressure. Gross margin fell from 56.3% to 41% as compute expenses increased. That means Zhipu is not immune to the same cost pressures affecting the wider industry.
Still, traders appear more willing to pay for its revenue structure because enterprise customers may be stickier than consumer users. Zhipu’s contracts reportedly include large-scale projects with power-grid and energy clients, along with work involving domestic processor integration. Its strategy has been described as a combination of base models, domestic compute capacity, and a developer platform.
That positioning has helped it attract attention from traders looking for a local enterprise AI company with characteristics similar to global business-focused model providers. Forecasts cited in the market put Zhipu’s 2027 forward price-to-sales ratio at 57 times, compared with 29 times for MiniMax. The premium reflects expectations that Zhipu may have stronger long-term monetization potential, even though its losses remain large.
MiniMax is not without options. Its enterprise platform revenue grew 198% year on year and carried gross margin near 70%. The challenge is whether that business can grow fast enough to offset losses from consumer products. The company’s future valuation may depend on how quickly it shifts from traffic-driven applications toward paid tools and enterprise services.
compute costs reshape the AI economy
Behind the stock moves is a broader structural problem: AI growth is colliding with hardware scarcity and rising compute demand. Training and running advanced models require large quantities of graphics processors, high-performance memory, data-center power, and networking infrastructure. As models become more capable, they often require more processing power, not less.
Nvidia chief executive Jensen Huang has repeatedly warned that computing demand is expanding at an extraordinary pace as AI models move into more industries and more complex tasks. The core concern is that centralized cloud providers and large data-center operators may not be able to build physical capacity fast enough to satisfy industrial demand.
That bottleneck matters for equity traders and digital-asset traders alike. If compute becomes the limiting resource for AI, then the market may increasingly value companies and networks that can secure, reduce, distribute, or monetize computing power.
The decline in prices for some consumer AI software companies has strengthened that argument. A platform may have millions of users, but if the cost of serving those users rises with every interaction, scale can become a liability rather than an advantage. In contrast, infrastructure that provides paid compute to enterprise clients may have a more direct path to revenue.
decentralized compute draws attention
The pressure on centralized computing has also brought renewed attention to decentralized physical infrastructure networks, often known as DePIN. These networks use blockchain-based incentives to coordinate physical resources such as graphics cards, storage, wireless equipment, sensors, or other hardware owned by independent operators.
In the AI compute segment, the basic idea is to connect idle or underused graphics processors from many providers and make that capacity available to developers or companies that need processing power. Supporters argue that this can create distributed supercomputing markets outside traditional cloud platforms.
The market value of decentralized computing projects was estimated at about US$12.2 billion in 2024, according to figures circulated in the digital-asset sector. Some industry projections have suggested that the broader distributed physical infrastructure market could reach multitrillion-dollar scale by 2028, including estimates as high as US$3.5 trillion. Such forecasts remain uncertain and depend on adoption, regulation, hardware availability, network reliability, and enterprise demand.
The appeal is cost. Corporate cloud giants typically charge developers several dollars per hour for access to premium graphics processors, with pricing often cited in the range of US$3 to US$8 per hour depending on chip type, region, and contract terms. Decentralized compute protocols claim they can offer comparable workloads at discounts of 50% to 80%, although performance, uptime, support, compliance, and data-security standards can vary widely.
More than 13 million machines are said to contribute computing resources to distributed networks globally. If those machines can be reliably organized for enterprise workloads, the model could create a new layer of AI infrastructure. If they cannot meet corporate requirements for speed, stability, privacy, and accountability, the opportunity may remain limited to smaller developers and experimental use cases.
digital-asset traders reassess utility tokens
The sell-off in consumer AI names has also influenced sentiment in digital-asset markets. Traders who follow AI-linked tokens are increasingly separating projects tied to free consumer applications from those connected to paid infrastructure services.
The distinction is important. Tokens linked mainly to attention, social engagement, or basic chatbot traffic may face the same monetization problem as consumer AI apps. If users do not pay enough to cover compute costs, the underlying economics remain weak. By contrast, tokens or networks that provide measurable computing resources to paying customers may have clearer revenue logic.
That does not mean infrastructure-linked tokens are automatically safer or more valuable. Digital assets remain volatile, and network claims about usage, revenue, and hardware capacity require careful verification. Traders may need to examine actual utilization rates, payment flows, customer concentration, hardware quality, and the share of protocol activity tied to real enterprise workloads rather than speculative incentives.
The key question is whether decentralized compute networks can attract sustained demand from companies that need AI processing at scale. If they can, their economics may be supported by real usage. If they rely mostly on token rewards to attract hardware supply, they could face pressure when market conditions weaken.
the new measure of value
The split between Zhipu and MiniMax shows how quickly market assumptions can change. During the mobile-internet era, user attention was often treated as the foundation of value. The larger the audience, the stronger the platform was assumed to be.
AI changes that formula. Every query, generated image, video edit, chatbot response, or coding request consumes compute. That means usage creates both revenue opportunity and cost exposure. A large user base is valuable only if the company can monetize that usage above the cost of delivering it.
For Zhipu, the current market reward reflects confidence in enterprise demand and pricing power. For MiniMax, the decline reflects concern that consumer traffic may not be enough to support its valuation unless monetization improves or enterprise revenue grows much faster. Both companies still face significant challenges, including high losses, margin pressure, and rising hardware costs.
The broader lesson is already shaping capital allocation across technology and digital assets. Traders are looking beyond raw user numbers and asking which platforms can turn AI activity into paid output, measurable efficiency, and recurring revenue. In that environment, productivity tools, workflow integration, enterprise APIs, and compute infrastructure are becoming more important than downloads and daily app engagement.
The HK$800 billion valuation gap between Zhipu and MiniMax captures that transition. The market is moving away from the belief that mass usage alone creates value. It is moving toward a stricter test: who pays, how much they pay, and whether the platform can deliver the service profitably after compute costs are counted.
To see how AI reshapes crypto trading economics, explore Toobit’s detailed insights in today AI moves shake markets.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
