Sentiment Has Turned
Softbank, Thiel, Burry...
This Selling In Not Mechanical
Mispriced Risk
Labor Data For Real?
Bear Market Rallies & 6666
Softbank, Thiel, Burry...
You saw the headlines by now - the big SI whales are selling. And yet NVDA stock is still above key $184 heading into earnings Wednesday night, and then it is still enjoying a weekly breakout level above key $173 - that it must get/stay below to risk reversing it’s strong bullish trend and with it the semi sector.
Regardless, the AI narrative is being poked and prodded from lots of angles.
Here’s Goldman’s latest worries on AI as software:
1. Capital is now the bottleneck: ORCL debt/CDS stress highlights that capex is hitting limits; if spend slows, “picks & shovels” look mispriced, with leveraged/peripheral names (NEO, multi-cloud) getting hit hardest.
2. Crowding is extreme: The AI trade is everywhere across sectors, so any wobble triggers outsized rotations and sentiment shocks.
3. Kimi K2 spooked the market: A near-GPT model at a fraction of the cost reignites doubts about the ROI of massive datacenter spend.
4. Tencent cut capex: Even if blamed on chip constraints, lower spend reinforces fears that the hyperscaler ROI math may not hold.
(GS, Privorotsky)
Then there is the focus on AI as Infrastructure, especially as compare to China (also GS, but different department):
1. US holds 44% of global data center capacity but faces power bottlenecks in 8 of 13 regional markets.
2. US power spare capacity is declining, approaching critical thresholds (15% threshold) by 2030.
3. China projects 400 GW power spare capacity by 2030, exceeding global data center demand (120 GW).
4. China is expanding power supply across renewables, coal, gas, and nuclear to support AI growth.
5. US power constraints could temporarily slow AI progress by 2030 despite current leadership.
6. China’s grid size (twice US) ensures sufficient spare capacity for data centers and other industries.
7. US power market tightness has already triggered spikes in real-time power prices and capacity costs.
The take-away?
Best line I’ve read on synthesizing these points comes from Dave Friedman:
“One is inclined to say that Goldman doesn’t understand the underlying tech. One is also inclined to say that Silicon Valley doesn’t understand the underlying finance….”
He continues:
“Market 1: may be underpriced, because most investors cannot model compounding inference demand.
Market 2: may be overpriced, not because AI will disappoint, but because the financing horizon is longer than the technological horizon.”
And the 5 Trillion dollar question:
“… is whether the downstream cash flows justify all of Nvidia’s sales. No evidence yet that that is the case.”
Nvidia earnings Wednesday certainly would delight if Jensen addressed this question to investor satisfaction. More likely, he will continue to talk his book: that there is insatiable demand, so keep buying GPUs, keep building more datacenters... because that’s the only way to win.
But is that the best investment bet? Again, Dave Friedman:
“Markets do not crash when narratives change. They crash when funding durations and technological durations diverge.”