MINIMAX’s $117M Revenue Is a Mirage: The 17.8% Margin Tells the Real Story

The chart says growth. The volume says pain. Over the past six months, MINIMAX—China’s darling of AI video generation—posted revenue of $117 million, up 283.1% year-over-year. On paper, that's explosive. On my screen, it reads like a warning flare. Let me be clear about what I'm looking at. This is not another textbook SaaS curve. This is a company burning $358 million in half a year while telling the market that everything is fine because losses narrowed 11%. I've watched this movie before. It was called DeFi Summer, and it ended with a lot of people holding bags while the founders held press releases. Here's the number nobody's talking about: gross margin. The filing shows gross profit of $20.8 million on $117 million in revenue. Do the math. That's 17.8%. For context, a healthy software company runs at 70% or higher. Even a cloud provider scraping by on infrastructure sits at 30%. MINIMAX is generating video at a cost structure that makes mining Dogecoin look efficient. This is the classic crypto winter pattern transplanted onto the AI industry. In 2021, we saw protocols with $2 billion in TVL and no revenue. Now we see AI companies with $117 million in revenue and no margin. The chart lies. The volume speaks. The revenue jump is real. I'm not disputing that. Gross profit grew 464.8% year-over-year, which tells me the unit economics are improving. But when your gross margin is 17.8%, the trajectory doesn't matter yet. What matters is that MINIMAX is spending $3 for every $1 it makes. That's not a business. That's a mission statement. Let me dig into the architecture. MINIMAX doesn't compete head-to-head with OpenAI on text. It's an angle I've watched since the Paris hackathon days—when I learned to spot the gap between what a white paper promises and what the demo actually does. MINIMAX is building multi-modal, particularly video generation. The Hailuo series. The video generation models. The voice and text models that round out the stack. That's the right play. Video is where the demand is. The global content industry is adopting AI tools at a pace that scares every producer I talk to. And MINIMAX has captured a piece of that demand. But the cost structure is brutal. Video inference is expensive. Generating a minute of high-definition video requires thousands of GPU inference passes. That's not a software problem. That's a physics problem. The compute is the cost. The GPU is the bottleneck. And MINIMAX, as a Chinese company, faces an additional constraint: the export controls. They can't just buy every H100 they want. They have to rely on domestic chips or negotiate with cloud providers. Here's where I want to push back on the optimistic narrative. The revenue growth is impressive, but it's driven by a strategy of buying market share with cash. In crypto, we call it liquidity mining. You pay people to use your product, inflate the usage metrics, and then hope you can raise another round before the truth catches up. MINIMAX is doing the same thing. They're pricing aggressively, offering free tiers, and subsidizing subscriptions to build a user base before the technology gap closes. This works when your investors believe in the story. And the story here is solid: China's Sora killer. The multi-modal AI leader. The next thing in generative media. But stories have a shelf life. Let me focus on the margin issue more because it's the part that will get overlooked in the mainstream press. In the AI industry, gross margin is the clearest signal of technological efficiency. For text models, the compute cost is falling. For video, it's still stuck. The architecture is the cost. There is no magic upgrade that makes video generation cheap; it's a physical constraint of the hardware and the algorithmic efficiency. MINIMAX has 17.8%. This means that 82.2% of every dollar they earn goes directly to compute, bandwidth, and storage. That leaves very little to pay for sales, research, and admin. And yet they're spending $358 million a year. That's a serious gap. From my time at the Paris Hackathon, I remember the pattern: a team demoes a beautiful front end, but the smart contract has a reentrancy bug. The visual is smooth. The foundation is cracked. MINIMAX has a smooth product. But the foundation—the cost structure—is cracked. Let me look at the competitive landscape for a moment. In video generation, MINIMAX competes with players like ByteDance's Jimeng and Kuaishou's Kling. These companies have distribution channels that MINIMAX doesn't. They have app stores. They have social media platforms. They have millions of users who can generate videos without leaving their ecosystem. MINIMAX has to rely on the quality of its model to win. That's a dangerous game. In crypto, we've seen how fast a technical lead can be erased. One model release from a competitor and your edge is gone. What is the contrarian angle here? The thing that nobody is talking about? The article I read focused on the revenue number, but the actual strategic story is the margin. The company is spending $358 million a year to build a moat in video generation, and the unit economics are not improving fast enough to justify the burn. Here's the counter-intuitive insight: maybe the real value of MINIMAX isn't in their model, but in their infrastructure. Every AI company is essentially a compute reseller. They buy GPUs, they rent them to users, and they hope the inference costs fall faster than the prices they can charge. The company that owns the most efficient compute pipeline will win. It's the same as mining. The winners are the ones with the cheapest power. But MINIMAX doesn't own its compute. It rents it from cloud providers. That means its margin is subject to the whims of Alibaba Cloud, Tencent Cloud, or whoever they're using. The moment those providers raise prices, MINIMAX's margin will get even thinner. So the question is: can MINIMAX reach a point where the margin improves enough to justify the current losses? That's a question of technology. Can they develop a better model architecture that uses fewer tokens? Can they optimize inference so that each video costs less? Or can they build a pricing structure that allows them to charge more for their output? The answer to that question will determine whether MINIMAX is the next great company or the next cautionary tale. The timeline is critical. They have about 18 months to show meaningful improvement in their margin before the market loses patience. The next 12 months will be defined by the availability of compute. If MINIMAX can secure more compute at lower prices, it can scale. If the GPU supply is constrained, it will be forced to raise prices or reduce the quality of its service. The chart lies. The volume speaks. The volume of MINIMAX's revenue is real, but the volume of its losses is even louder. I'm not saying the company is doomed. I'm saying the current narrative is incomplete. The market sees a growth story, but it's missing the real story: a company with a cost structure that doesn't yet support its ambitions. Alpha doesn't wait for permission. I'm not waiting for the next earnings call to make a judgment. I'm watching the margin. I'm watching the compute costs. I'm watching the next round of funding. What should you watch? The next quarterly report. The change in gross margin. The announcement of new enterprise partnerships. The cost of Hailuo's video generation. These are the signals that matter. Because at the end of the day, a company that burns $358 million on $117 million in revenue is not a business. It's a pilot program. The question is whether it will become a business before the capital runs out. That's the question that will define MINIMAX. And that's the question I'll keep watching.