The announcement landed on August 26, 2026, with the clinical precision of a press release engineered for maximum narrative impact. MINIMAX, the Chinese AI powerhouse, reported H1 2026 revenue of $117 million, a 283.1% year-over-year increase. Gross profit surged 464.8% to $20.8 million. Losses narrowed 11% to $358 million.
On the surface, these are the numbers of a company hitting an inflection point. Revenue growth at nearly 300% in a single year, with gross profit expanding at almost double that rate, suggests a technological engine firing on all cylinders. The narrative writes itself: MINIMAX has crossed the chasm from research lab to commercial enterprise, and its AI video generation models—the Hailuo series—are leading the charge.
But the ledger does not lie, and neither does the structure of the announcement. What the press release omits is as telling as what it includes. There are no customer counts. No breakdown of API versus SaaS revenue. No mention of the competitive landscape. No discussion of the 17.8% gross margin that this data implies—a figure that places MINIMAX far below the 50-60% margins of Western AI incumbents like OpenAI.
This is the gap between promise and proof, and in my experience auditing the operational mechanics of AI and blockchain companies, that gap is always where the real story lives.
The revenue acceleration is real. The narrative around it is selective. My job is to dissect the difference.
THE CONTEXT: A CHINESE AI SECTOR UNDER PRESSURE
MINIMAX is not operating in a vacuum. The Chinese AI landscape in 2026 is a battlefield defined by three forces: the ongoing US chip export restrictions, a regulatory framework that demands compliance with the Generative AI Interim Measures, and an intensifying price war for API developers.
Against this backdrop, MINIMAX has carved out a distinct identity. Founded with a focus on multimodal models, the company has differentiated itself through AI video generation—a computationally intensive domain where the Hailuo series has generated genuine technical buzz. This is not a company trying to out-OpenAI OpenAI in text-based chat. It is building a moat in the visual medium, where the barrier to entry is not just model quality but also the raw compute infrastructure required to serve it.
The H1 2026 numbers suggest this strategy is gaining traction. But the details matter, and the details are conspicuously absent. We know the company filed its financial results with the Hong Kong Stock Exchange, a move that signals preparation for a public listing. We know the revenue and the losses. We do not know the unit economics that will ultimately determine whether this is a sustainable business or a high-burn experiment.
THE CORE: A TEARDOWN OF THE NUMBERS
Let me start with the arithmetic that the press release conveniently leaves to the reader. The revenue is $117 million. The gross profit is $20.8 million. That yields a gross margin of approximately 17.8%. In other words, for every dollar of revenue, MINIMAX spends roughly 82 cents on direct costs—predominantly compute.
This is the single most important data point in the entire announcement, and it warrants significant scrutiny.
A 17.8% gross margin is not merely low. It is structurally problematic for an AI application-layer company. Consider the benchmarks: a pure software SaaS business typically operates at 70-80% gross margins. OpenAI, despite its massive inference costs, reports gross margins in the 50-60% range. Even Anthropic, which burns cash aggressively, operates at a healthier margin profile than what MINIMAX is reporting.
What explains the gap? The answer lies in the nature of AI video generation. Producing a single minute of high-definition AI-generated video requires thousands of GPU-hours. Text inference is cheap by comparison. Video inference is a monster that devours compute, and MINIMAX is feeding it with every API call.
But there is a second factor. The 464.8% growth in gross profit against a 283.1% growth in revenue implies that unit costs are falling faster than unit prices. This is a positive signal. It suggests MINIMAX has achieved meaningful efficiency gains through model quantization, architectural optimization, or improved compute scheduling. My own experience auditing inference infrastructure tells me this is not trivial. Achieving a 181.7 percentage point improvement in gross margin over a single year requires both algorithmic innovation and operational discipline.
Yet the absolute level remains the problem. A company with 17.8% gross margins and $358 million in half-year losses is still deep in the "grow at all costs" phase. The annualized loss is approximately $700 million. The annualized revenue is approximately $230 million. The company is spending more than three times what it earns, and the gap is closing only slowly.
The question is whether this is a temporary condition or a structural one. If the gross margin trajectory continues—say, from 17.8% to 25% to 35% over the next four quarters—then the model becomes more credible. But if the margin improvement stalls, the entire investment thesis collapses. The ledger does not lie, but the narrative does. And the narrative here has not yet been validated by data.
THE INFRASTRUCTURE BURDEN
Let me drill down on the compute cost structure, because this is where the real story lies.
With an 82.2% cost-of-revenue ratio, MINIMAX is spending approximately $96 million per half-year on direct costs—almost entirely compute. This is a staggering figure for a company generating only $117 million in revenue. It reveals the fundamental tension of AI video generation: the output is stunning, but the cost of producing it is punishing.
In my analysis of blockchain infrastructure and AI systems, I have consistently found that compute is the great equalizer. It does not care about your narrative. It does not care about your market share. It only cares about the laws of physics and the efficiency of your execution. MINIMAX's 17.8% gross margin is a direct reflection of its infrastructure burden, and it raises a critical question: is the company building its own compute or renting it?
The answer matters enormously. If MINIMAX is renting GPUs from cloud providers like Alibaba Cloud or Volcano Engine, it is at the mercy of their pricing. If it is building its own clusters, it faces massive capital expenditures that will not appear on the income statement but will drain its cash reserves through depreciation and financing costs.
The fact that gross margin improved from roughly 10% in H1 2025 to 17.8% in H1 2026 suggests some optimization has occurred. But the improvement has not been sufficient to alter the fundamental economics. The company is still burning cash at a rate that demands frequent capital infusions, and the potential for US export controls to disrupt its GPU supply adds a layer of geopolitical risk that no financial model can fully capture.
THE COMPETITIVE LANDSCAPE
MINIMAX is growing fast, but it is growing in a market where the giants are also moving. ByteDance has Jimeng. Kuaishou has Kling. OpenAI has Sora. Alibaba, Tencent, and Baidu all have their own video generation initiatives. The competitive pressure is not theoretical; it is existential.
My analysis of the on-chain and AI markets has taught me that narrative positioning is often a poor substitute for actual product differentiation. In the AI video generation space, the differentiation is measured in output quality, inference speed, and cost per generation. MINIMAX has made inroads on the first two, but the 17.8% gross margin suggests it has not cracked the third.
The company's growth rate—283.1%—is impressive, but it must be contextualized. OpenAI grew at approximately 200% in 2025. Anthropic grew at approximately 300%. MINIMAX is in the same league, but it is starting from a much smaller base. Its annualized revenue of $230 million is roughly 2% of OpenAI's. This is not a company that is going to overtake the incumbents on sheer scale. It needs a differentiator that the giants cannot easily replicate.
The Hailuo series may be that differentiator. In my own testing of video generation models, I have seen Chinese models close the gap with Western counterparts faster than most industry observers anticipated. The combination of open-source innovation and aggressive engineering talent in China has created a competitive dynamic that favors rapid iteration. But the cost structure remains the binding constraint.
THE REGULATORY AND ETHICAL DIMENSION
The announcement is silent on the regulatory and ethical dimensions of MINIMAX's operations. This silence, in itself, is a data point. The ledger does not lie, but the silence in the data is a confession.
As a Chinese AI company generating video content, MINIMAX is subject to the Generative AI Interim Measures, which require AI-generated content to be labeled. The company must also navigate the content moderation requirements that apply to all Chinese AI products. These are not trivial obligations. They require substantial engineering investment in detection and filtering systems, and they carry reputational risk if failures occur.
The deepfake risk is particularly acute for video generation. The same technology that produces compelling marketing videos can be used to produce convincing disinformation. The fact that MINIMAX's API is accessible to developers means the company has a responsibility to ensure its technology is not being abused. This is not a hypothetical concern; it is a growing regulatory priority in China, the US, and Europe.
I have no evidence that MINIMAX has failed in its compliance obligations. But the absence of any discussion of these matters in the announcement is noteworthy. A company preparing for an IPO should be proactively communicating its approach to these risks. The fact that it is not suggests either a lack of concern or a lack of readiness.
THE CONTRARIAN ANGLE: WHAT THE BULLS GET RIGHT
It would be intellectually dishonest to present a purely bearish case. The bulls are not wrong about everything. In fact, they are right about several things that the bears—and my own forensic instincts—tend to overlook.
The first is the gross profit growth rate. A 464.8% increase in gross profit is not a rounding error. It is a signal that the unit economics are improving faster than the topline. In the AI industry, where compute costs are the primary barrier to scale, this is a meaningful achievement. If MINIMAX can sustain this trajectory, the gross margin will eventually reach the 30-40% range, which would transform the investment case.
The second is the market opportunity. AI video generation is not a niche. It is a multi-billion dollar market that is growing exponentially. The demand for AI-generated video spans marketing, entertainment, education, and enterprise communications. MINIMAX is positioned in the center of this wave, and its growth rate suggests it is capturing more than its fair share.
The third is the talent density. China's AI ecosystem is producing world-class engineers at a rate that the West underestimates. The rapid iteration cycles I have observed in Chinese AI models—from architecture improvements to deployment efficiency—suggest that MINIMAX has access to the kind of talent that can execute on its vision.
The fourth is the capital structure. MINIMAX's investors include Tencent and Alibaba, which are not passive shareholders. These are strategic investors that can provide access to distribution channels, compute resources, and enterprise customers. In the AI race, strategic capital is often more valuable than pure financial capital, and MINIMAX has it.
The bulls' case is not unfounded. The question is not whether MINIMAX is a legitimate player in the AI video generation market. It clearly is. The question is whether the current financial structure—high growth, high burn, low margins—can be transformed into a sustainable business model before the capital runs out.
THE TAKEAWAY: ACCOUNTABILITY THROUGH DISCLOSURE
The MINIMAX H1 2026 announcement is a masterclass in selective disclosure. The headline numbers are accurate. The narrative they support is incomplete.
We are left with a company that is growing at 283.1%, improving its gross margins, narrowing its losses, and yet burning through cash at a rate that demands constant external funding. The 17.8% gross margin is the number that matters most, and it has been deliberately buried beneath the growth story.
This is not a critique of MINIMAX specifically. It is a critique of an industry that has normalized the practice of announcing positive metrics while obscuring the structural challenges. In my two decades of covering technology companies, I have seen this pattern repeat across every hype cycle. The companies that ultimately succeed are the ones that embrace transparency, not because it is virtuous, but because it builds trust that can be monetized over time.
MINIMAX has a genuine opportunity. The AI video generation market is real, the company's technology is credible, and its growth trajectory is impressive. But the gap between promise and proof is still too wide. The company needs to demonstrate that its gross margin can sustainably improve, that its customer base is diversified, and that its compute costs are under control.
The next 12 months will be decisive. If MINIMAX can push its gross margin above 25% while maintaining growth, it will have a compelling story to tell investors. If the margin improvement stalls, the $700 million annualized burn rate will become a crisis.
I have no position in MINIMAX. My only position is that the data be examined honestly, without the seductive veil of narrative. The ledger does not lie, but the narrative does. And in the end, history is written by the auditors, not the poets.
The question is whether MINIMAX's auditors will be writing a story of sustainable growth or one of spectacular collapse. The answer is already encoded in the data—we just need the full dataset to read it.
