Masa’s Son
Non-Consensus Long/Short Technology Investor
NVIDIA's GPU value pool faces architectural competition and shorter useful lives than ten-year assumptions imply. Etched and other accelerator designs may challenge performance within five years, while power limits favor relentless gains in performance per watt; past GPU generations improved more than 2-3x, and industry experts reportedly used four-year lives before GPT. The range of outcomes remains wide. x.com x.com x.com
NVIDIA insiders selling looks like a warning for the AI infrastructure trade. A board member of three decades has cut their stake by 20% since June, following heavy discretionary selling by the CEOs of CoreWeave and Caterpillar and Larry Ellison's flip-flop at Oracle. Insiders dumping on the supposed verge of AGI undercuts the story they are selling. x.com
NVIDIA faces rival chips from every Mag 7 customer, which has an incentive to cut GPU spending and improve margins on its AI buildout. If they can make a GPU 85% as good for 50% less, it is worth using instead of NVIDIA's; the shift will take time. NVIDIA makes a GPU for $5,000 and sells it to those customers for $50,000, leaving them an incentive to save 90% by building their own. x.com x.com x.com x.com
AMD could see much more accelerator demand as OpenAI and Anthropic allegedly increase their orders by multiple gigawatts. Nvidia's only real edge over AMD may soon be its size and resources; Lisa Su is a master operator. x.com
AI software and hardware advances make GPUs and power more valuable, creating huge long-term tailwinds for compute providers and AI model and app businesses. Each new software generation increases a GPU's value; each hardware generation increases the value of a gigawatt of power. That favors Nvidia, Broadcom, Intel and AppLovin. x.com
AI's physical buildout looks set to last to 2030 and beyond because financing, memory supply and hyperscaler spending still point to sustained demand. Nebius has $3.5B in financing, IREN $2.4B at 9% and SPCX $40B for GPU equipment. SK hynix expects shortages through at least 2030 and Samsung says HBM4 is sold out; memory companies are making money with no sign of slowing. Google expects 2027 capex to rise significantly. Pullbacks can happen; be cautious and buy quality. x.com
Scarce compute prices Nebius's and CoreWeave's GPU rentals by marginal-token value, not chip cost. Better models lift installed watts' value faster than new GPUs devalue old ones. SpaceX says contracts renew at higher dollars per watt; CoreWeave's A100 deal runs to 2029, and Nebius's Blackwell auction cleared 15% above its prior high. The 2-3 year GPU-life bear case keeps getting set back. If supply catches demand, older GPUs would be repriced by performance per watt, but that looks far off. … Show more
Nvidia GPUs may retain high residual values while memory, power, data centers and chips remain scarce. Ornn Data's chart shows B200 rentals in October still more than twice their January level, with older H200 rentals higher still. This may last longer than expected, but normal depreciation returns once newer generations ramp and supply catches up. x.com x.com
SpaceX's exclusive use of Nvidia GPUs and other supply constraints make Musk's 13-20 GW operational compute forecast for end-2027 too high. A substantial buildout is still incredibly enticing, but reaching that scale likely requires cheap capacity from struggling neoclouds, mostly through capacity deals and some acquisitions. Its attractive $40B off-balance-sheet GPU lease deal is only a drop in the bucket of the debt needed. Capacity can still grow quickly over time. x.com x.com x.com x.com … Show more
Older GPUs retaining value after years of rental income is a positive for IREN's GPU-cloud economics. Per @wallstengine, an A100 system with 9,600 GPUs and servers is listed for $59.7 million, about $6,200 per GPU. Against an estimated original rack cost of $25,000 per GPU, that suggests a 25% residual after six to seven years; the GPUs also rented for $2-10 an hour for six years. The residual is based on an asking price, not a sale. x.com x.com x.com
Nvidia is winning fast inference: OpenAI shipped its flagship speed tier, GPT-6 Astra Ultrafast, on Nvidia GPUs, gaining a second source and eroding the exclusivity Cerebras was priced on, though Cerebras is still faster and capacity likely played a role. Operators keep extending GPU lives, Amazon to 5-6 years, Microsoft and Google to 6, Meta to 5.5, because old chips keep earning on inference. More lifetime revenue per chip justifies customers' next Nvidia order. x.com x.com x.com x.com
CoreWeave may run out of financing before it can access GPUs under contract in about four years. If it cannot pay Nvidia, share sales are its remaining option until investors see no path to payment and the stock collapses. GPU-backed loans are no longer available even with customer backing; bankruptcy is forecast for Q3 2027 based on its books. x.com x.com x.com
