lets start with the DCF because thats where this whole thing falls apart immediately. nvidia is currently trading at somewhere around 35-40x forward revenue. not earnings. revenue. to justify that in a discounted cash flow model you need to assume data center revenue grows at 40%+ annually for the next 8-10 years, that gross margins stay above 70% despite competition, and that the discount rate stays relatively benign. change any one of those assumptions even slightly and the fair value collapses hard. i ran it with 20% long term growth (which is still an extremely aggressive assumption for a semiconductor company at this scale) and a 10% discount rate. you get somewhere in the $90-105 range depending on terminal assumptions. $96.66 is not a random number, its roughly the midpoint of a realistic range under non-euphoric assumptions.
revenue picture is genuinely impressive, ill give them that. data center segment went from like $15B annually to over $80B in two years. but heres the problem with hypergrowth at that velocity - it almost never sustains. the hyperscalers buying H100s and B200s are doing so in a capex supercycle that is itself driven by competitive panic, everyone buying because everyone else is buying. microsoft, google, amazon, meta are all spending hundreds of billions on AI infrastructure and a huge chunk flows to nvidia. but capex cycles turn. they always turn. when amazon decides they have enough GPUs for 18 months, what does nvda's quarterly look like. nobody wants to model that scenario right now because the stock goes to $97 in that model and that makes people uncomfortable.
enterprise value is currently sitting north of $2.5 trillion. to put that in context thats more than the entire german stock market. for one semiconductor company. the EV/EBITDA is somewhere around 45-50x. TSMC, which literally manufactures the chips nvidia designs and has irreplaceable strategic value, trades at like 15x. the premium nvidia commands over its own foundry partner is extraordinary and requires permanent dominance to justify.
the 1999-2001 comparison writes itself here. cisco at peak 2000 was priced for permanent internet infrastructure dominance. they were the picks and shovels play, the "safe" AI bet of that era, everyone said you couldnt lose because the internet was real and cisco made the routers. cisco revenue was real. cisco growth was real. stock still dropped 86% and has never in 25 years returned to its 2000 high. nvidia is the cisco of the current cycle. the underlying technology is real, AI is real, but the price already assumes nvidia wins everything forever with no competition, no margin pressure, no customer diversification away from third party silicon.
which brings me to the competition point that gets hand-waved constantly. google has TPUs. amazon has Trainium and Inferentia. microsoft is designing custom silicon. meta is building MTIA chips. these are not startups, these are the exact customers nvidia depends on for 40%+ of revenue, all vertically integrating away from nvidia dependency. AMD exists. startups like Cerebras and Groq exist. the moat is real today but the timeline on which it stays this wide is not 10 years, its probably 3-4.
balance sheet is actually strong, cash position around $35B, minimal debt, FCF is real and growing. this is not a zombie company, not a fraud, not even close. which again makes it worse from a valuation standpoint because you cant blame the price on financial engineering. the business is genuinely excellent and the stock is still expensive by a factor of 3-4x. FCF yield at current prices is under 2%. you are paying for 30 years of perfect execution upfront.
my fair value is $96.66. i think we get there either through a demand air pocket in the data center cycle, a broader multiple compression in AI names, or just the slow grinding realization that $150B+ quarterly revenue implied by current prices is not actually coming. probably some combination of all three.
not shorting it. the momentum and index buying make timing impossible. but i wouldnt touch it as a long either and i think anyone buying at $130+ is essentially paying 2000-era cisco prices