MINIMAX's 283% Revenue Surge Hides a Structural Problem: The Math of AI Video Generation Doesn't Close

The market is not celebrating a victory; it is pricing in a survival story. Over the past fiscal period, MINIMAX, the Chinese AI video generation powerhouse, reported revenue of $117 million, a staggering 283.1% year-over-year increase. Gross profit jumped 464.8% to $20.8 million. On the surface, this is the kind of growth that launches a thousand headlines. But the macro view reveals what the micro hides. Strip away the top-line enthusiasm, and you find a company burning $358 million in losses—down only 11% from the prior year—against a gross margin of roughly 17.8%. This is not a technology company. This is a capital-intensive infrastructure play wearing a software disguise. In my years analyzing cross-border payment rails and DeFi liquidity pools, I have seen this exact financial silhouette before. It resembles a Layer-2 network that generates massive transaction volume but leaks value to the underlying Layer-1 for security and data availability. The revenue is real. The unit economics, however, are still broken. The question is not whether MINIMAX can grow; it is whether the growth mechanism can ever overcome the structural cost of its own inputs. The core insight here is mathematical, not narrative. A gross margin of 17.8% means that for every dollar of revenue, over 82 cents is consumed by direct costs—predominantly GPU compute, electricity, and network infrastructure. Video generation is not text inference. Generating a single minute of high-definition video requires thousands of GPU inferences, each consuming energy and time. This is the fundamental constraint that dictates the company's trajectory. Unlike a SaaS company that reaches 70% gross margins at scale, MINIMAX is structurally bound to its hardware suppliers. My experience auditing Terra's algorithmic stability in 2022 taught me to look for feedback loops that appear sustainable in the short term but mathematically guarantee collapse under stress. MINIMAX faces a similar, albeit less dramatic, loop. High growth attracts capital. Capital buys compute. Compute enables model training and inference. Better models attract more users. But every incremental user adds marginal compute cost that directly erodes the already thin gross margin. The company is essentially renting its growth from NVIDIA and cloud providers. Regulation is the new liquidity engine, but so is hardware access. The contrarian angle is uncomfortable for the AI narrative crowd. Everyone is focused on the 283% revenue growth as proof of product-market fit. But the real story is the 17.8% gross margin, which suggests that MINIMAX's competitive moat is not its model architecture or its brand—it is its access to capital and subsidized compute. If a competitor with deeper pockets, such as ByteDance or Kuaishou, can match the model quality and offer lower prices, the growth rate will stall. Strategy prevails where sentiment fails, and the strategy here is a race to achieve scale before the capital window closes. Let me be specific about the cost structure. In 2025, I led a cross-border stablecoin pilot on Polygon, and I learned that infrastructure costs do not scale linearly. They scale with complexity. For MINIMAX, the complexity is in the inference stack. The $358 million loss is not merely a function of high research and development spend; it is a direct consequence of the astronomical cost of serving video models to millions of users. The 11% loss reduction is encouraging, but it is likely driven by revenue growth dilution rather than genuine operational efficiency. If the company were truly optimizing its cost structure, the gross margin would be expanding far faster than it is. Mapping the chaos, one block at a time, requires us to look at the geopolitical dimension. MINIMAX is a Chinese company. Access to the most advanced GPUs, such as NVIDIA's H100 or H200, is restricted by US export controls. This forces reliance on domestic chips like Huawei's Ascend series or indirect access through cloud providers. This is a structural bottleneck that Western competitors do not face. It increases the cost per FLOP and reduces the efficiency of the training pipeline. This is not a minor issue; it is a fundamental constraint on the company's ability to close the gap with SOTA models like Sora or Veo. The industry impact is significant. MINIMAX's growth validates the commercial demand for AI video generation, but it also exposes the brutal economics of the sector. The company is a proof point for the entire Chinese AI ecosystem, demonstrating that domestic players can achieve global-scale revenue. However, it also serves as a warning: the path to profitability in AI video is not through better models alone, but through owning the compute infrastructure or achieving unprecedented algorithmic efficiency. From a competitive standpoint, MINIMAX has carved a niche by focusing on multimodal generation, specifically video, rather than competing head-on with OpenAI on text. This is a smart tactical move. But the moat is shallow. The barrier to entry for a well-funded competitor is not model knowledge; it is the capital to burn on compute and marketing. The company is likely employing a 'space-for-time' strategy—using aggressive pricing and marketing to build a user base before the technology gap narrows. This works only if the company can achieve a scale that makes switching costs prohibitive for users. The investment case is a classic growth-at-any-price scenario. Using a price-to-sales multiple of 10-20x, MINIMAX's valuation could range from $1.17 billion to $2.34 billion. But this assumes the growth rate persists. With a burn rate of approximately $700 million annually, the company needs a cash reserve of at least $1.4 billion to survive two years without new funding. If the capital markets tighten, the dilution risk is substantial. Trust is verified, never assumed, and in this case, the market is betting on a future that is far from guaranteed. So what is the takeaway for the macro observer? The convergence of AI and crypto is inevitable; timing is tactical. MINIMAX's financials are a microcosm of a broader trend: the AI industry is becoming a compute-constrained, capital-intensive sector where only the most efficiently financed players survive. For investors, the signal to watch is not the revenue growth but the gross margin trajectory. If MINIMAX can push gross margins above 40% within the next four quarters, the business model becomes credible. If it remains in the high teens, the company is merely a high-volume intermediary for GPU vendors. The next 12 to 18 months will determine whether this is a business or a burn.