MiniMax boosts registered capital to $4 billion and expands headcount to 276 as it prepares for an A-share listing, signaling a shift in the large-model race from technical competition to a battle of capital and organization.
A major shift in momentum often happens silently. As the public's novelty with generative AI fades, investment institutions are scrutinizing large model companies with unprecedented severity. Recently, a document quietly disclosed on the CSRC website tore away the thin veil shrouding the technological glory of domestic AI unicorns. MiniMax, a star company long ranked at the forefront of China's large model tier and known for its exceptionally high technical ambitions, has officially signed a counseling agreement with CITIC Securities, making no secret of its ultimate capital goal: the A-share market.
Capital markets have always been extremely sensitive, and they tolerate no pretense. During the past several years of rapid large model development, AI startups were often carefully packaged as an elitist, asset-light game. A few top algorithm talents from prestigious backgrounds, a few papers that caused a huge stir at leading international conferences, and a grand narrative about humanity's silicon-based future and general artificial intelligence were enough to easily pry hundreds of millions of dollars in funding from the primary market dominated by dollar-denominated venture capital.
However, in the current macroeconomic cycle, this kind of castle in the air—relying purely on storytelling and industry rankings to maintain extremely high valuations—has completely collapsed. The secondary market, especially the A-share listing review system, which is known for its extremely strict scrutiny of profits and cash flow, has long moved past a period of tolerance for purely technical expectations. It has fully shifted toward extreme pickiness about a company's real cash-generation ability, financial health, and closed-loop commercial implementation. MiniMax's decision to make a counter-trend push toward the A-share market at this highly sensitive moment of market sentiment is by no means blind confidence in its current book profits. Rather, it is a strategic rush and a life-or-death breakout forced by the unfathomable computing costs behind it and the brutal pace of industry reshuffling.
The Underlying Interest Chain of the Four Billion Capital Defense Line
This extreme hunger for heavy capital has long left an extremely clear and unignorable evolutionary track in the company's underlying corporate structure and capital path. If we try to strip away the complicated media PR releases and trace upward through the equity and registration tracks recorded in the Tianyancha system, we can see the heavy-asset transformation this AI unicorn has undergone in preparation for this hard battle.
System information reveals with extreme precision that Shanghai Xiyu Jizhi Technology Co., Ltd., the core operating entity affiliated with MiniMax, has seen its registered capital undergo a violent reverse expansion in the recent period, jumping directly from RMB 1 billion to RMB 4 billion, an overall increase of 300%.
The RMB 3 billion in new registered capital is absolutely not a simple stacking of numbers to cope with the listing counseling process. For a technology entity wholly owned by Hong Kong Xiyu Jizhi, permanently locking in massive offshore structure funds or real money from earlier funding rounds as onshore paid-in registered capital through extremely high legal binding force means management is building the highest financial breakwater for a protracted computing power war of attrition.
We must clearly recognize that by mid-2026, the global supply chain for cutting-edge computing chips had evolved into a zero-sum game among major tech companies. The procurement price of a single high-performance computing card has long broken through conventional commercial pricing logic, with the premium space stretched indefinitely. Training and inference for large models are essentially a bottomless furnace that devours money. Every iteration and reorganization of a hundred-billion or trillion-parameter model is backed by tens of thousands of high-end GPU clusters humming day and night, along with electricity and cloud server costs running into the tens of millions or even hundreds of millions of yuan.
This hard foundation of RMB 4 billion is MiniMax's hard-core deterrent to the entire large model track. At a time when high-end computing hardware is under extremely strict geographical blockades and domestic replacement chips are still in a fragile period of capacity ramping and ecosystem alignment, possessing extremely large paid-in capital means having the absolute confidence to bulk-purchase computing hardware through special channels. It can not only be used to lock in long-term computing power leases for the next three to five years in advance, but also, at critical moments, use cash flow advantages to buy out the scarcest high-quality industry corpus data suppliers.
This strategy of turning intangible technological competition into pure capital crushing is extremely cold-blooded. MiniMax knows full well that in the future large model war, the underlying logic is no longer about who writes more elegant code, but who can maintain full-load operation of a 10,000-card cluster under extreme supply chain pressure. Those waist-level startup teams that still try to start with tens of millions of yuan and rely on fine-tuning low-cost open-source models will completely lose their bargaining power at the computing power negotiation table under the gravitational crush of this RMB 4 billion.
Organizational Fission at a Breakneck 137-Fold Expansion
Beyond the geometric explosion in capital base, the company's extremely rapid expansion in organizational structure and headcount is also profoundly upending outsiders' stereotypes about AI startups. From just two insured employees when it was founded in 2021 to a surge to 276 people by 2024—a 137-fold personnel fission—this is not just a cold set of social security data; it reveals a profound qualitative change in the large model industry: the entire track is shifting from a small workshop-style research effort with a handful of people in a lab to labor-intensive, resource-intensive modern industrial mass production.
Outsiders and large numbers of retail investors chasing hot topics often naively assume that the nearly 300-person team is entirely composed of algorithm geeks and frontier scientists earning million-yuan salaries. This kind of talk, which lacks basic industry common sense, completely obscures the real difficulties and dirty, tiring work involved in bringing large models to commercial implementation.
The clearest reality is that once the underlying reasoning capability of a base large model reaches a certain threshold, the marginal utility of pure algorithm optimization is rapidly diminishing. What truly determines whether a model can earn real money in the market is the large engineering delivery team behind it, the meticulous data-cleaning team that endures monotonous work, and the government-enterprise sales specialists who can go deep into various industries to crack hard nuts.
These 276 employees with real social security contribution records are a direct reflection of MiniMax's transformation from a single model provider to a complex integrated AI service solutions provider. Beneath the tip of this organizational iceberg, they must staff a large number of senior product managers to repeatedly polish the extremely subtle interaction experiences on the application side; they must build professional compliance and risk-control teams to cope with increasingly strict generative AI filing and content security review requirements; and they need a formidable, highly executable commercial field sales force to convince conservative clients in traditional industries that integrating large models can genuinely deliver cost reductions and efficiency gains, rather than just adding an extremely expensive chat toy inside the enterprise.
The extremely rapid expansion in headcount not only brings heavy rigid payroll pressure, but also places extremely high demands on the young founding team for organizational culture integration and management control. Managing a few geniuses in a lab who share a common technical belief may be easy, but managing several hundred modern corporate employees bearing strict performance targets and fighting across various non-standard commercial fronts is a challenge of internal friction management and interest allocation that is no less difficult than training a giant network from scratch.
The Interest Game of an A-Share Listing and the Countdown to Survival
Choosing to charge toward the A-share market at this particular moment is an extremely dangerous but also the most practical capital gamble for MiniMax. For a long time, the inertial thinking of Chinese internet tech companies and the exit paths of venture capital institutions have almost always been to seek overseas listings without question. However, amid the profound restructuring of global capital flows today, overseas capital markets' valuation logic for China's core underlying hard technologies has undergone irreversible and deeply distorted changes. Overseas listings face extremely strict technology reviews and compliance red lines for core data leaving the country; meanwhile, other capital channels have to endure prolonged liquidity droughts and severely discounted valuations.
By contrast, although the A-share market imposes almost harsh hard requirements on net profit indicators and sustainable operating capability, it is backed by national-level institutional investors with a strong willingness to support independently controllable core hard technology. If MiniMax can successfully pass the strict review of the A-share issuance review committee and get listed, it will gain not only an extremely high P/E premium and tens of billions in direct fundraising, but also a fully opened trust link with domestic government industrial guidance funds and large state-owned central enterprises.
Being listed on the A-share market is itself a highly valuable safety and compliance pass backed by national credit. For MiniMax, which urgently needs to expand into domestic government-enterprise major clients and break into smart city and national-level intelligent computing center projects, this strategic positioning value far exceeds the financing act itself. Once it holds this listing license, it can participate in national lifeline industries that are extremely sensitive about data security in a completely legitimate and lawful manner.
But this in no way means the A-share IPO channel is an easy bone to chew. The CSRC's review of tech companies seeking listings has always been sharp and cold-blooded in penetrating the underlying interest chains. MiniMax must provide the issuance review committee with extremely detailed and watertight supporting materials, presenting real revenue flows that can withstand the most stringent financial audits. They must prove that their C-end consumer products have established a working logic for user top-ups and subscription monetization, or that their B-end private deployment has formed sustainable high repurchase rates and positive operating cash flow.
More critically, the A-share review system has a highly self-consistent set of evaluation standards for a company's technological attributes. They will examine with a magnifying glass whether your core technology is truly independently controllable and whether there is excessive dependence on overseas open-source agreements in the key underlying architecture. If a large model is merely packaged as a faster search engine without the vertical integration capability to be deeply embedded into core business systems such as industrial production lines or financial risk-control hubs, then even the most dazzling star team is highly likely to be exposed under the relentless string of tough questions from the issuance review committee.
MiniMax's recent series of extremely aggressive capital operations and organizational expansion has, in effect, sounded the early end-of-era horn for the rough-and-tumble first phase of China's large model startups. The next stage of the battle will be a bloody millstone defined by heavy-asset investments of tens of billions of real money, industrial-scale armies of hundreds or thousands of people, and brutally ruthless quarterly commercialization assessments. In this cruel track where the ultimate truth of life and death is defined by computing power density, data security compliance, and real cash-generation efficiency, any romantic fantasy about technology changing the world must bow to the cold financial statements. In this survival race with no retreat, the only way to win a sliver of hope is to abandon illusions of asset-light miracles, and use the heaviest asset barriers and the muddiest, most grounded commercial closed loops to harvest the remnants of the market.
