The 17.8% Margin That Explains Everything About AI Video Generation

In the quiet of the bear, we count the coins. But this is not a bear market—it is a bull market for AI narratives, and the numbers coming out of MiniMax's H1 2026 financial disclosure demand a closer look than the headline-grabbing 283.1% revenue growth suggests. The real story is not the top line; it is the gross margin. At 17.8%, this is a company spending $0.82 of every dollar earned on direct costs, primarily compute. That is not a technology company. That is a commodity processing plant with a GPU rental bill. And yet, the market is pricing it like a software monopoly. Let me explain why this discrepancy is the single most important data point in the entire AI application layer right now. To understand the context, we have to map the global liquidity environment. We are in a period where the Federal Reserve's balance sheet is expanding again, M2 money supply is ticking upward, and risk assets are responding accordingly. In this macro regime, capital flows to the highest-growth stories, regardless of unit economics. This is the same pattern we saw in 2020 with SaaS companies and in 2017 with ICOs. The market is not rewarding profitability; it is rewarding narrative velocity. MiniMax's 283.1% revenue growth fits this narrative perfectly. But my experience mapping capital flows during the ICO era taught me that when the narrative outpaces the fundamentals, the correction is brutal. The question is not whether MiniMax is growing—it is whether the growth is sustainable enough to survive the next liquidity contraction. Let me break down the core mechanics. MiniMax reported H1 2026 revenue of $117 million, up 283.1% year-over-year. Gross profit came in at $20.8 million, up 464.8%. That is a massive improvement in efficiency—the gross margin expanded from roughly 10% in H1 2025 to 17.8% now. On the surface, this is a positive signal: the company is getting better at controlling compute costs. But let me put this in perspective. OpenAI operates at a 50-60% gross margin. Anthropic is in a similar range. Even the most capital-intensive AI infrastructure companies, like CoreWeave, manage to eke out better margins than this. A 17.8% gross margin means MiniMax is essentially a pass-through entity for GPU costs. The company is generating $117 million in revenue but spending $96 million just on direct compute and data costs. That leaves very little room for R&D, sales, or administrative expenses—which explains the $358 million net loss for the half-year. Now, here is where my institutional due diligence background kicks in. When I was preparing risk assessments for the Spot Bitcoin ETF applications, we looked at custody solutions and market manipulation surveillance. The key was to identify structural vulnerabilities before they became systemic. MiniMax has a structural vulnerability: its business model is entirely dependent on the cost of AI inference, which is itself dependent on the availability of high-end GPUs. For a Chinese company, this is doubly problematic. The US export controls on NVIDIA H100 and H200 chips mean MiniMax is likely running on a mix of older NVIDIA chips (H800/A800) and domestic alternatives like Huawei Ascend. This creates a cost disadvantage that is not going away. The 17.8% gross margin is not a temporary inefficiency; it is a structural ceiling imposed by the geopolitical environment. The contrarian angle here is the decoupling thesis. The market narrative is that MiniMax is a high-growth AI company that will eventually achieve OpenAI-like margins as it scales. I am skeptical. The alpha hides in the variance others ignore, and the variance here is between the revenue growth rate and the gross margin trajectory. If MiniMax were truly achieving technological breakthroughs in inference efficiency, we would see gross margins approaching 30-40%. Instead, we see a company that is growing revenue by 283% but only improving gross margin by 7.8 percentage points. This suggests the growth is being bought, not earned. The company is likely underpricing its API access to attract developers, a classic land-grab strategy. But in a capital-intensive industry like AI video generation, underpricing is a death sentence unless you have an infinite capital runway. MiniMax's $358 million half-year loss implies an annual burn rate of over $700 million. At that pace, the company needs to raise significant capital within the next 12-18 months, or it will face a liquidity crisis. Let me talk about the specific business line, because this is where the real insight lies. MiniMax's growth is almost certainly driven by its AI video generation models, particularly the Hailuo series. Video generation is the most compute-intensive application in AI—a single minute of high-quality video generation can cost more than 100,000 text-based API calls. This explains the low gross margin. But it also explains the revenue growth. The AI video generation market is at an inflection point, with players like ByteDance's Jimeng, Kuaishou's Kling, and OpenAI's Sora all vying for dominance. MiniMax is positioning itself as a differentiated player in this space, focusing on multimodal content generation rather than general-purpose chatbots. This is a smart strategic move—it avoids direct competition with OpenAI and Anthropic in the text domain. But it also means MiniMax is competing in a market where the cost structure is brutal and the incumbents have deeper pockets. ByteDance can subsidize Jimeng indefinitely. Kuaishou has a massive user base to cross-sell. OpenAI has the brand and the compute partnerships. MiniMax has neither. We do not predict the storm; we build the hull. This is the principle I apply to every investment decision, and it is the principle that should guide any analysis of MiniMax. The company's financials reveal a classic growth-at-all-costs strategy. The revenue growth is real, the market opportunity is real, but the unit economics are not yet viable. The question is whether MiniMax can improve its gross margin from 17.8% to 30%+ within the next 12-24 months. This would require a combination of inference optimization, model quantization, and scale economies. It is possible, but it is not guaranteed. The company's ability to raise capital at favorable valuations will be the determining factor. If MiniMax can secure a $5-7 billion valuation in an IPO on the Hong Kong Stock Exchange, it will have the runway to continue its land-grab strategy. If the IPO market turns cold, the company will be forced to make a choice between growth and survival. Looking at the broader picture, MiniMax's financials are a microcosm of the entire AI application layer. We are seeing a wave of companies growing at 200-300% annually while burning through cash at alarming rates. This is not sustainable. The market will eventually demand profitability, and when it does, companies with gross margins below 20% will be the first to be punished. The takeaway is not to short MiniMax—the stock could easily double in the next six months on IPO momentum. The takeaway is to understand the structural dynamics at play. The AI video generation market is a capital-intensive, compute-hungry, brutally competitive space. The winners will be those who can achieve technological breakthroughs in inference efficiency, not those who simply buy growth with underpriced APIs. In the quiet of the bear, we count the coins. In the noise of the bull, we count the costs. MiniMax's costs are still too high, and that is the only number that matters.