Revenue hit $117 million. Up 283.1% year-over-year. Gross profit exploded 464.8%.
Every headline screams growth. Every press release frames this as validation of the AI video generation gold rush.
Then you look at the gross margin.
17.8%.
That's not a technology company. That's a logistics business. And nobody in the echo chamber wants to do that math.
Let me walk you through what the press release doesn't say, because the ledger remembers everything.
Context: The Numbers Behind The Narrative
MiniMax, the Chinese AI startup known for its Hailuo video generation models, filed financials that show a company in hyper-growth mode. Six-month revenue of $117 million. Gross profit of $20.8 million. Net loss of $358 million—which actually narrowed 11% year-over-year.
The story being sold: AI video generation has crossed the chasm. Real customers are paying real money. The unit economics are improving.
The story the data tells: This is a capital-intensive business burning cash at an alarming rate while the core product economics remain fundamentally broken.
Here's what matters. Revenue grew 283%. Gross profit grew 464%. That's the positive signal. The gap between those two numbers suggests genuine operational improvement—better inference optimization, improved model architecture, smarter cost controls.
But 17.8% gross margin means for every dollar MiniMax brings in, it spends $0.82 on direct costs. For context, mature SaaS companies run 70-80% gross margins. Even cloud infrastructure providers—which carry heavy physical costs—typically clear 50%.
MiniMax's margin profile places it closer to a commodity reseller than a defensible technology platform.
Core Analysis: The Unit Economics of AI Video
Let me break down what 17.8% actually tells us about the business.
Video generation is computationally brutal. Generating one minute of high-definition video can require thousands of GPU inference calls. Compare that to text generation, where a single prompt-response cycle consumes fractions of a cent in compute. The cost structure is fundamentally different.
Based on my experience auditing token projects and analyzing on-chain infrastructure costs, I've seen this pattern before. When direct costs consume over 80% of revenue, one of three things is happening:
The company is underpricing to capture market share. The inference pipeline is inefficient. Or the underlying compute costs are simply too high for the current pricing model.
In MiniMax's case, it's likely all three.
The 464.8% gross profit growth suggests rapid efficiency gains. But here's the uncomfortable question: is that improvement coming from genuine technical optimization, or from aggressive pricing that's still losing money on every transaction?
The loss of $358 million is roughly three times revenue. That's not a company approaching profitability. That's a company still in the "buy growth at any cost" phase—and the cost is enormous.
Let's put this in perspective. If MiniMax continues at this burn rate, annualized losses approach $700 million. The company will need continuous capital infusion. And in a market where investors are increasingly demanding a path to profitability, that's a fragile position.
The gross margin trajectory is the single most important metric to watch. If MiniMax can push from 17.8% toward 30-40% over the next two quarters, the story changes. If it stays stuck below 25%, the fundamental economics don't work—regardless of revenue growth.
The Contrarian Angle: Growth Is Not A Moat
The 283% revenue growth is real. That's not in dispute.
But let me challenge the narrative that this validates the AI video generation business model.
What this actually validates is the demand side. Yes, customers want AI video generation. Yes, they'll pay for it. But MiniMax's financials reveal something uncomfortable: the supply side is brutal.
The company is effectively functioning as a pass-through for GPU costs. Every dollar of revenue carries with it $0.82 of direct costs. The value creation—the model intelligence, the user experience—is being squeezed between expensive compute on one side and competitive pricing pressure on the other.
This is the "pick and shovel" problem in reverse. Normally, selling picks and shovels during a gold rush is the smart play. But in AI, the shovel makers (GPU providers like NVIDIA) capture most of the value. The miners—companies like MiniMax—are left competing on razor-thin margins while bearing all the risk.
Here's what worries me more. The competitive landscape is about to get dramatically worse. OpenAI's Sora, Google's Veo, ByteDance's Jimeng, Kuaishou's Kling—all are pushing into the same space. All have deeper pockets and existing distribution channels.
MiniMax's current growth is partly a function of first-mover advantage. The question is whether that advantage persists when competitors with 10x the resources enter the market with comparable technology.
History suggests it doesn't. On-chain data doesn't lie, and neither do financial statements. Every market that sees this kind of capital influx eventually experiences margin compression and consolidation. AI video generation will be no different.
The China Factor And Compute Constraints
The other elephant in the room: compute access. MiniMax is a Chinese company subject to US export controls on advanced semiconductors. This creates two problems.
First, the company may not have access to the most efficient GPUs (H100, H200, and beyond). That means higher compute costs per unit of output—a structural disadvantage that will be difficult to overcome through software optimization alone.
Second, the company is likely dependent on domestic alternatives like Huawei's Ascend chips or on indirect access through cloud providers. Both options carry performance and cost implications.
This explains, at least partially, the depressed gross margins. It's not just about pricing strategy or model efficiency. It's about being locked out of the best hardware while your competitors have unrestricted access.
Follow the TVL, not the tweets. In this case, follow the compute, not the press releases.
The strategic question for MiniMax: how do you build a sustainable competitive advantage when your cost structure is structurally higher than your global competitors? The answer, so far, appears to be "grow fast and figure it out later." That's a risky strategy in a capital-intensive business.
What Actually Matters Now
Three metrics will tell you whether MiniMax's story holds:
Gross margin trajectory. If it stays flat or declines, the business model is broken. If it climbs above 30%, there's hope.
Customer concentration. Is this a few large enterprise contracts or broad-based demand? Concentrated revenue is fragile revenue.
Model iteration speed. MiniMax needs to stay ahead of Sora and Veo on quality-per-compute. That's a hard race to win when you're operating under hardware constraints.
The current financials are a snapshot of a company buying market share at a substantial discount. Whether that's a smart investment or a value trap depends entirely on the next two quarters.
Smart contracts have no mercy. Neither do financial markets. The market will eventually price MiniMax based on its ability to convert growth into sustainable unit economics—not on the headline revenue number.
The revenue is real. The growth is real. But the margin profile suggests this is a business that needs to fundamentally restructure its cost base to survive. Until that happens, the 283% growth story is a distraction from the uncomfortable truth: this company is spending $3 for every $1 it earns, and the path to profitability remains unclear.
I've seen this pattern before in crypto—projects with impressive usage metrics and terrible economics. The market eventually figured it out. It always does.
The question for MiniMax is whether it can fix the economics before the market forces the issue.
That's the only question that matters.