Michael Burry Shorts Nvidia, Buys Calls: The Hedge That Reveals the Real Bet

The options flow hit my terminal at 14:32 EST. Someone bought 5,000 contracts of Nvidia calls at the $200 strike, mid-range, paying single-digit premiums. The same entity had just disclosed a massive short position. Michael Burry. The man who called the housing bubble now straddles the AI trade like a cat on a hot tin roof. You don't short the most valuable company on Earth and then buy calls as an afterthought. That's not conviction. That's risk management wearing a disguise. Burry's Scion Asset Management filed the 13F in February 2025. The position: puts on Nvidia, calls as a hedge. The market read it as bearish. I read it as something else entirely. A pure short would be clean. This is a barbell. And barbells tell you more about the lifter's uncertainty than their strength. Let's establish the battlefield. Nvidia holds roughly 90% of the AI data center GPU market. The H100 sells for $25,000 to $30,000. Gross margins sit above 70%. The company's data center segment contributes over 85% of revenue. CUDA, the software ecosystem, has over 4 million developers locked into its paradigm. This is not a company under siege. This is a fortress with a moat full of code. Burry's thesis, as I parse it from the filings and his public statements, hinges on three pillars. First, the monopoly is temporary. AMD's MI300 series approaches H100 performance in select benchmarks. Google's TPU v5p competes in training. Amazon's Trainium and Microsoft's Maia are in the pipeline. Second, the valuation is stretched. The trailing P/E hovers around 60-70x. Third, capital expenditures will crush profitability. Nvidia is spending billions on data centers and R&D. Burry calls this 'entering and passing through the top of the capex bubble.' Here's where my forensic instincts kick in. I've audited ZK-proof circuits under load. I've watched DeFi liquidity pools drain in real-time. I know what a structural flaw looks like. Burry's thesis has a structural flaw: it underestimates the switching costs embedded in CUDA. Let me walk you through the math. A developer who has spent five years optimizing PyTorch pipelines on CUDA doesn't just switch to AMD's ROCm because the hardware is 10% cheaper. The migration cost is measured in engineering months, not dollars. I've seen this play out in my own work. When I audited StarkWare's circuits in 2019, the gas optimization I found saved 14% on verification time. But that optimization was worthless outside the specific constraint system. The same principle applies here. CUDA is not just a compiler. It's a gravitational field. Now, the contrarian angle. Everyone is focused on the short. I'm focused on the calls. Burry bought call options as a hedge. That's textbook protective put strategy. But the strike price matters. $200 mid-range, single-digit premium. That's not a hedge against a catastrophic drop. That's a hedge against a squeeze. If Nvidia rallies on earnings, the short bleeds. The calls cap that bleed. This tells me Burry expects volatility, not necessarily a collapse. Here's the insight nobody is talking about. Burry's historical record with this exact strategy is mixed. He's admitted it himself. In past earnings plays, the hedge has sometimes cost more than the short gained. This is not a man with a crystal ball. This is a man running a probabilistic model with asymmetric downside protection. The real question is whether the AI capex cycle is peaking. Look at the hyperscalers. Microsoft, Meta, Amazon, Google. They're spending billions on data centers. Some of that spend is going to Nvidia. Some of it is going to custom silicon. The custom silicon threat is real but slow. Trainium and Maia are years away from scale. By the time they arrive, Nvidia will have shipped Blackwell and Rubin. The architectural lead is roughly two years. That's an eternity in this market. Let me give you a concrete data point from my own trading. In 2021, I ran a Uniswap V3 vs SushiSwap arbitrage script. 450 micro-trades in a day. Netted $28,000. The lesson wasn't about arbitrage. It was about efficiency. The market prices in what it can see. What it can't see is the 15-minute lag between OTC desk sales and ETF spot purchases. I found that lag in January 2024, right after the Bitcoin ETF approvals. Institutional mechanics create supply shocks that retail sentiment can't predict. The same dynamic applies to Nvidia. The options market is pricing in a binary outcome. The underlying reality is a gradient. So what's the takeaway? Burry is not wrong about the long-term trajectory. Competition will erode Nvidia's margins. Custom silicon will gain share. The question is timing. Short-term, the CUDA moat holds. The system-level solutions—DGX, NVLink, InfiniBand—create a barrier that single-chip competitors can't breach. The software subscription model, CUDA-X at $4,500 per GPU per year, is a recurring revenue stream that the market hasn't fully priced. I'm not telling you to buy Nvidia. I'm telling you to understand the structure of the bet. Burry is hedging his own uncertainty. That's the signal. When a legendary short seller buys calls, he's telling you the downside is real but the upside is unpredictable. The market is a machine for processing information. The options chain is the exhaust pipe. Read the exhaust. Watch the $200 strike. If Nvidia breaks above it on volume, the short squeeze accelerates. If it fails, the puts gain traction. Either way, volatility is the only certainty. Position accordingly. And remember: arbitrage is just efficiency with a heartbeat. Burry is arbitraging his own reputation against the market's collective delusion. The question is who blinks first.