Smart Money Splinters on Nvidia as AI Capex Cycle Approaches $600 Billion

Rarely has the marquee AI trade attracted such openly contradictory institutional positioning. David Tepper added to Nvidia, Dan Loeb liquidated the entire holding and George Soros's vehicle purchased puts — three incompatible views on the same cash-flow stream. The bull case is anchored in scale: Amazon, Alphabet and Microsoft are collectively directing nearly $600 billion toward capital expenditure, an outlay that flows disproportionately to accelerator and networking suppliers. The bear case is one of duration and depreciation — whether hyperscalers can sustain that intensity once return-on-invested-capital scrutiny intensifies, and whether custom silicon erodes pricing power at the margin. Adjacent developments, including reported collaboration between Elon Musk's and Jensen Huang's ventures, reinforce the vertical-integration thesis but also concentrate ecosystem risk. For allocators, the practical takeaway is that Nvidia has ceased to be a directional AI proxy and has become a valuation-and-cycle-timing debate, with hedging costs rather than conviction increasingly setting position sizes.