NVDA +20% YTD: Greifeld Loves the $500B AI Bet, But Has One Big Fear

$NVDA jumped 3% on Tuesday after announcing half a trillion dollars in funding commitments to drive compute as an asset class, bringing its year-to-date gain to roughly 20%. The YouTube finance crowd is mostly bullish. But former Nasdaq CEO Bob Greifeld, speaking on CNBC, offered the most interesting take in the room: he likes the vendor financing expansion because it spreads risk to third-party lenders who can help set market rates for compute. That part is fine. The part that should make you sweat? Greifeld flagged that LLM compute efficiency could improve 100-fold over the next decade, which would gut the long-term collateral value of the data center hardware underpinning those loans.

Meanwhile, over on the AI infrastructure side, $CRWV (CoreWeave) is having a moment that makes Nvidia's move look almost conservative. The stock surged as much as 23% after reporting Q2 revenue that doubled year-over-year, a backlog exceeding $100 billion, and a raised fiscal year 2026 outlook. CNBC noted CoreWeave already contracted an additional $2 billion in new deals within just the first five weeks after quarter close. For context, $25 billion was roughly CoreWeave's entire backlog one year ago. That is not a typo.

On X, the semiconductor crowd is treating CoreWeave as the cleanest proxy for AI infrastructure demand, pointing out that even older Nvidia GPU generations are running at high utilization through 2029. The Bloomberg morning crew noted the Philadelphia Semiconductor Index surged 3% on the day, with $MU (Micron) up 6.5%. Bob Sloan of S3 Partners, also on CNBC, had called the memory trade peak on June 22nd and now says the DRAM rebound of 8% validates his earlier warning. He is now ringing the bell on SpaceX as the frothiest trade he has ever seen, with short interest dropping from $32 billion to $24 billion in just 60 days.

The bull case for Nvidia is pretty simple: Ed Yardeni raised his S&P 500 target to 8,400 citing an earnings-led melt-up, tech saw its second biggest inflow week ever according to Bank of America, and the AI compute sector is showing an 85.1% earnings beat rate this season. The bear case, in Greifeld's own words, is that you are underwriting loans against hardware that could be worth a fraction of today's value if model efficiency keeps compounding. That is a real structural risk buried inside a very exciting headline number.

If Nvidia's $500B compute financing gamble lands, it redefines the company from chip seller to financial infrastructure. If efficiency curves blow past anyone's models, it becomes the most expensive bet in tech history. Either way, $NVDA is no longer just a semiconductor stock, and the market is only beginning to price what that actually means.