📊 Unprofitable Chinese AI startup valued on par with Gazprom

The first full-scale IPO in this sector, launched on the Hong Kong Stock Exchange by Chinese developer MiniMax, has finally pulled back the curtain of secrecy. For media executives, producers, and content creators who rely daily on video generators like Hailuo or text-based platforms, this financial report offers far more than dry figures. It serves as a roadmap revealing the actual cost of AI production, the scale of price dumping, and the long-term survival prospects of independent AI services facing fierce competition from Big Tech.

Key Metrics of the AI Unicorn

The MiniMax brand is hardly a household name just yet. Founded in 2022, the Chinese company develops foundational generative models. Within the creative community, it is best known for its video model, Hailuo, which confidently secured a spot in the top 10 of Artificial Analysis’s authoritative rankings in 2025.

In January 2026, MiniMax launched its initial public offering (IPO) on the Hong Kong Stock Exchange, offering the AI industry its first opportunity to dissect the real financial inner workings of the generative model business. The trading results stunned the market: since listing, the company’s shares have tripled. Consequently, a startup with a net profit margin of negative 2,368% achieved a valuation of approximately $30 billion to $35 billion, effectively matching the market capitalization of the Russian energy giant Gazprom. Our detailed analysis breaks down how this anomalous economic model operates.

To understand the scale of MiniMax’s business in the first quarter of 2026, consider these confirmed metrics:

• Market Capitalization: Approximately $30 billion to $35 billion.
• IPO Capital Raised: $620 million, with shares skyrocketing 109% on the first day of trading.
• Revenue: $79 million against research and development (R&D) expenditures of $250 million.
• Headcount: 428 employees at the close of 2025.

Business Structure and Architecture

The company scales its AI products across two primary pillars: consumer (B2C) and enterprise (B2B). The evolution of the startup’s tech stack began with large language models for text (the abab 1, abab 5.5, and abab 6.0 lineups based on the MoE architecture) before transitioning into multimodality. This led to the rollout of the Speech-01 voice model, the Music generator, and ultimately, its flagship video model, Hailuo, followed by upgraded versions of its Text-01, M1, M2, and Hailuo-02 generators in late 2025 and early 2026.

Today, the company’s business model rests on two foundational segments.

The first comprises generative AI-native products for the B2C market. This includes user interfaces for all key modalities: the MiniMax M2 text model, the Hailuo video platform, and proprietary audio and music services. Additionally, for the domestic Chinese market, the company operates the highly popular product Talkie (localized as Xingye)—an interactive virtual character chat service that serves as a direct counterpart to the West’s Character.ai. Notably, Talkie’s targeting and interface are tailored heavily toward a younger demographic; the app’s onboarding screen prompts users to select age cohorts where the topmost tier is delicately labeled as “over 26.”

The second segment is the OPEN Platform for the B2B sector. This is an advanced infrastructure for third-party developers and enterprises, providing APIs to integrate text, speech, video, and images into commercial applications ranging from interactive entertainment and smart devices to marketing, healthcare, and online education. The infrastructure layer includes its own platform for large-scale experimentation, model training frameworks, and optimized systems for inference.

Currently, the aggregate audience across the company’s AI services exceeds 212 million users, with the lion’s share—175 million—anchored by Talkie. In the B2B segment, over 100,000 companies and independent developers globally utilize the platform, and in the spring of 2026, the ecosystem was augmented with its own intelligent agent, MaxClaw.

Working “for Peanuts” Compared to Silicon Valley

An analysis of personnel expenses at MiniMax reveals an extraordinary degree of frugality by AI industry standards. As of late 2025, the company employed 428 specialists, with a total compensation fund of $84 million (up from $55 million in 2024). Simple math shows that the average annual compensation for a single engineer or researcher hovers around $200,000—a figure that includes factored-in stock options.

By the standards of American AI giants, where scarce talent commands multi-million-dollar salaries, the Chinese team is essentially working on pure enthusiasm. To put this into perspective: the 2025 compensation packages for the CEO of Warner Bros. Discovery—whose management brought the media conglomerate to the brink of bankruptcy—exceeded the annual payroll of MiniMax’s entire workforce combined by a factor of ten.

Dumping as an Entry Ticket

According to legal disclosures in the prospectus, MiniMax commands roughly 0.3% of the global market for foundational generative models, rounding out the world’s top 10 AI vendors. OpenAI remains the undisputed industry leader with a 30.1% market share ($3.2 billion in 2024 revenue), followed by Google at 16,9% ($1.98 billion) and Microsoft at 8.2% ($0.96 billion). Behind them sit Anthropic (4.7%), Midjourney (2.8%), and Amazon (1.8%). MiniMax ($40 million in revenue for fiscal year 2024) shares its cohort with players like Elon Musk’s xAI, ElevenLabs, Baidu, and Runway.

The global foundational model market, valued at $10.7 billion in 2024, is projected by analysts to clear the $206.5 billion mark by 2029—a twenty-fold increase, with $151.5 billion driven by applied AI applications and $55 billion by the Model-as-a-Service (MaaS) segment.

MiniMax’s positioning in this race is unique. Independent quality benchmarks (specifically from the Artificial Analysis platform) indicate that the company’s products do not occupy the top slots. In video generation, the Hailuo 2.30 model slipped to 25th place on the global leaderboard by spring 2026, trailing technological flagships such as Alibaba (HappyHorse-10), ByteDance (Dreamina Seedance 2.0), and Kuaishou’s Kling 3.0. In the real-world practice of AI video production, Hailuo consistently ranks in the top 10 for popularity but hovers near the bottom of that tier. In the text intelligence category, the flagship MiniMax-M2.7 maintains a respectable 13th place, sitting alongside Qwen3.6 Plus and Grok 4.20.

However, MiniMax’s ultimate competitive advantage is aggressive price dumping. The company’s cost for token and media generation is 3 to 10 times lower than that of its primary rivals. As of April 2026, generating one minute of video on the Hailuo 2.3 platform costs clients a mere $2.80. For comparison, a minute of video production on the Kling 3.0 Pro engine runs $13.44, while Google’s Veo 3.1 costs $12.00. The MiniMax-M2.7 text model, priced at a blended rate of $0.53 per million tokens, is 3 times cheaper than its comparable open-source competitor, Kimi K2.6 ($1.71), and nearly 20 times cheaper than the premium Claude Opus 4.6 ($10).

The current market consensus suggests that tier-two models still lag behind technology leaders in contextual comprehension and execution quality, even if the score gaps appear minor. The strategic calculation of investors backing MiniMax is straightforward: over time, the performance gap between AI models will inevitably narrow, and by leveraging ultra-low operational overhead and modest R&D spending, the Chinese startup will achieve actual profitability far faster than its competitors.

The Enterprise Sector Pays More Willingly

The conversion of MiniMax’s massive user base into tangible cash flow is in its infancy, averaging roughly 1%.

The breakdown of paying accounts relative to total users reflects this dynamic:

• MiniMax LLM (Text): 0.05%
• Hailuo AI (Video): 0.7%
• Talkie (Social Chat): 0.9%
• MiniMax Audio (Sound): 1.6%
• Open Platform (B2B): 1.9%

In line with standard SaaS market logic, enterprise clients demonstrate a significantly higher willingness to pay. The number of active B2B clients (entities spending at least $50 per billing period on APIs) is growing exponentially: rising from 100 organizations in 2023 to 700 in 2024, and reaching 2,500 clients based on the first 9 months of 2025.

On the retail side, the average ticket size for paying users in the Hailuo segment stood at $56 for the first 9 months of 2025 (approximately $6 per month). Total revenue for the Hailuo video model during this period reached $17.5 million, with 80% ($14.1 million) generated via fixed user subscriptions and the remaining 20% coming from situational token top-ups.

The Financial Balancing Act

Two overarching trends stand out within MiniMax’s revenue architecture. First is a pronounced wave of globalization: 70% of revenue is generated outside of mainland China, and the international share continues to expand. Second is a pivot toward enterprise servicing: while B2C accounted for 67% of revenue in 2025 compared to B2B’s 33%, commercial sales growth in the B2B segment is now significantly outstripping the consumer sector.

This directly mirrors the trajectories of global giants like OpenAI and Anthropic: the true commercial goldmine for AI lies within corporate integration. B2C revenues also feature distinct structural nuances; while the Hailuo video service monetizes through subscriptions, the entertainment-focused Talkie draws its primary income from advertising.

On the expenditure side, AI developers are traditionally weighed down by two massive cost centers: inference costs (the computational power required to keep live models running) and R&D investments (researching and training new networks). While legacy tech firms rarely see R&D costs exceed 5% to 10% of turnover, companies in the current phase of the AI cycle are forced to spend multiples of what they actually earn.

MiniMax’s core operating expenses reveal the following dynamics:

1. Inference and Direct Computation: The cost of keeping models operational comprises over 90% of the Cost of Goods Sold (COGS). MiniMax’s engineers have achieved remarkable technical progress here: by optimizing inference algorithms, COGS relative to revenue fell from 125% in 2023 to 88% in 2024, dropping further to 77% in 2025. Consequently, the company’s consolidated gross margin rose from 12% to a positive 25%. However, margins vary drastically by segment: B2C grosses balance on the razor’s edge of profitability at 4% (following a prolonged stint deep in the red), whereas the B2B segment reliably holds at an excellent 70%. These figures neatly explain why America’s OpenAI is rapidly shifting its gaze from mass-market B2C to enterprise clients while ruthlessly shelving structurally unprofitable consumer projects like its Sora video generator.
2. Marketing: During aggressive market-share acquisition phases, tech startups typically flood promotional channels with capital. MiniMax, however, bucked this trend by cutting marketing expenses by 40% year-on-year. Management attributes this to organic brand recognition for Hailuo and Talkie, bolstered by highly effective referral programs.
3. Research and Development (R&D): Investments in training next-generation architectures rose from just under $200 million to $250 million by the end of 2025, a sum three times greater than the startup’s current revenue. Nonetheless, the company’s capital cushion appears resilient: its existing financial runway provides enough leverage to subsidize and sustain developments for at least another four years. The IPO prospectus projects R&D spending to remain flat through 2029, with initial operating profits already appearing on the horizon. For a venture-backed AI startup, a four-year runway is an exceptional stability indicator.
4. Fair Value Loss: It was this exotic accounting metric—climbing to a negative $1.6 billion (twenty times the annual revenue)—that dragged the company’s net profit margin down to a shocking negative 2,368%. According to clarifications provided by the startup, this net loss is purely technical and paper-driven. It stems from the mandatory revaluation of the company’s preferred shares, whose balance sheet valuation experienced an avalanche-like surge immediately prior to the IPO due to intense investor demand.

If one strips this technical anomaly from the financial statements, the company’s real, adjusted operating margin still remains deeply negative at minus 316%, driven entirely by front-loaded R&D investments.

Takeaways and Core Questions For the Model

Analysts still harbor numerous reservations regarding the long-term viability of MiniMax’s business model, as well as the broader economics of foundational AI models. The anomalous post-IPO stock surge was largely driven by speculative factors and an acute shortage of “pure-play” AI assets on international stock exchanges (aside from MiniMax, only one other comparable Chinese AI developer, Zhipu, ventured an IPO early in the year).

The defining challenges for MiniMax over the coming years will center on:

• The Clash with Global Leaders: Will a strategy of “slightly lower quality at a 10x discount” guarantee long-term retention of enterprise clients in the competitive B2B segment?
• Regulatory Risks and Black Swans: Can the Chinese startup preserve and scale its international footprint (which yields 70% of its income) in the face of mounting geopolitical friction and Western sanctions targeting the PRC’s technology sector?