MiniMax's $117M Revenue Is a Mirage: The Gross Margin Says Everything

The tape doesn't lie. And the tape on MiniMax's latest financials is screaming something most headlines refuse to hear. Revenue hit $117 million, up 283.1% year-over-year. Gross profit surged 464.8%. The narrative writes itself: AI video generation is printing money. But then you look at the gross margin. 17.8%. Let that sink in. This isn't a software company. This is a reseller with extra steps. We didn't need a leak or a whispered source to see this. The numbers are right there, buried in the earnings release that everyone skimmed for the top-line figure. A 17.8% gross margin means that for every dollar MiniMax brings in, over 80 cents is swallowed by the direct cost of delivering the product. For a company that's supposed to be building the future of media creation, that's not a business model—it's a charity for GPU suppliers. Context matters here. MiniMax isn't just any AI startup. They're the Chinese darling of multimodal generation, the team behind the Hailuo video models that have been wowing users with their consistency and quality. They went public in Hong Kong, positioning themselves as the bridge between Western capital and Chinese AI engineering. And in a vacuum, the growth is undeniable. 283% revenue growth in a year is absurd by any standard. The problem is that this growth is being purchased at a price that would make a DeFi yield farmer blush. Let me break down what those numbers actually mean, because this is where the real analysis lives. The gross margin of 17.8% is the single most important data point in this entire release. Compare that to a mature SaaS company that typically runs at 70-80% gross margins. Compare it to a cloud provider like AWS, which hovers around 60-65%. MiniMax is operating at a fraction of that. The implication is brutal: the cost of compute—the GPUs, the electricity, the networking, the inference costs—is eating them alive. Here's the kicker based on my years auditing tokenomics and infrastructure projects: this tells me MiniMax is fundamentally a pass-through entity for Nvidia's margins. They're taking massive capital, converting it into GPU clusters, and then selling the output at a price that barely covers the hardware depreciation. The 464.8% growth in gross profit sounds impressive until you realize it's growing off a tiny base. $20.8 million in gross profit on $117 million in revenue is not a healthy scaling trajectory. It's a warning. The loss picture is equally telling. $358 million in losses, only an 11% narrowing from the prior year. That's not a company approaching breakeven. That's a company that's still in the "burn everything to win" phase. And in a bull market for AI narratives, that might be acceptable. But let's be real about what this means for the sector. Here's the contrarian angle that nobody's talking about: this isn't just a MiniMax problem. This is a systemic flaw in the AI video generation business model. Every player in this space—from OpenAI with Sora to Google with Veo to ByteDance with Jimeng—is facing the same structural math. Video generation is computationally monstrous. A single minute of high-definition AI video requires thousands of GPU inference passes. The cost curve is fundamentally different from text-based models. When ChatGPT runs a query, it costs fractions of a cent. When a video model generates a clip, it costs real money. This is why I've been saying the Layer2 sequencer debate has a parallel here. Everyone wants to talk about the decentralized future, but the centralized infrastructure costs are the real story. MiniMax's gross margin is the proof that video AI is a capital-intensive industrial business, not a software subscription business. The market hasn't fully priced this in yet. They see the revenue growth and assume the unit economics will magically improve. But based on my experience watching infrastructure projects, unit economics don't improve without fundamental architectural breakthroughs—and those take years. What's the hidden play here? MiniMax is likely using aggressive pricing to buy market share. They're deliberately selling below cost to capture enterprise API contracts and consumer subscriptions. It's a land grab. The bet is that once they have the users and the data flywheel spinning, they can raise prices and optimize their inference stack to the point where margins expand. That's a legitimate strategy, but it requires three things: endless capital, relentless technical optimization, and a competitor landscape that stays frozen while they scale. All three are unlikely. We need to talk about the cost structure more deeply. The 3.58x ratio of losses to revenue isn't just R&D spending. It's the cost of staying in the race. Every quarter they don't release a better model, they lose ground. Every quarter they do release a better model, they burn more cash on training runs. The Chinese GPU supply situation makes this even worse. Export controls mean MiniMax is likely paying a premium for less efficient hardware or relying on domestic alternatives that may not match Nvidia's performance. This is a structural handicap that no amount of revenue growth can fix. So what's the takeaway for anyone watching this space? The next 12-18 months are the test. If MiniMax can push gross margins above 30% while maintaining even half their growth rate, they might have a real business. If they can't, and if the loss ratio doesn't compress meaningfully, this becomes a cautionary tale about the difference between revenue and value. The market is going to start asking harder questions. When you're losing $358 million every six months, you're not building a moat—you're digging a hole. The tape doesn't lie. The question is whether investors are ready to read it.