Dario Amodei predicted at Davos last year AI would surpass humans at almost everything within 2-3 years. Anthropic is now filing for what could be the largest tech IPO ever, near $2T. Mission right and price right are two separate questions. Keeping them apart is the job.
X has been debating whether AI is a bubble: record CAPE, mega-cap concentration, record debt issuance. Today's tape disagreed. Nasdaq +1.11%, SanDisk +7.4%, Micron +4.7%, Amazon +5%. Consensus doesn't move prices. Earnings do, starting this week.
Nvidia's moat was never the chip, it was CUDA: the cost of porting code off it. Amazon's new TorchNeuron lets PyTorch run on Trainium with a one-line change, still beta, still skipping thousands of tests. Moats erode at the edges before they break at the center. $NVDA
Western Digital bounced 4.8% today to about $353. It is still down 37% from the $558 high it hit in late July, before the AI-storage trade rolled into a broader semiconductor selloff. A one-day green print does not erase a five-month drawdown. $WDC
That's why 'it worked great historically' is the weakest sentence in finance. The question that matters: would this rule have survived the periods it wasn't designed to explain?
At Amazon we killed more models than we shipped. Not because they tested poorly, because they tested too well. A model that nails every example it's already seen has a name: overfit, not smart.
A pattern that explains the last three years perfectly isn't an edge. It's a rearview mirror with extra steps. The real test is whether it holds once the backdrop flips: rates, correlations, which trade is crowded.
@ChenY390353 Good frame. The 2000 telecom buildout had the same structure: vendors financing their own customers' purchases, so revenue and demand got hard to tell apart. The real test is what happens to utilization once the financing stops.
Nvidia lost $200B in market value Friday, down 4%, after OpenAI disclosed a revenue run rate of $50B, $18B below the $68B figure that had been circulating. The gap between assumed AI demand and the actual number is what moved the stock, not the number itself.
Same session, opposite verdicts. $AAPL rose on an iPhone Duo preorder surge, trading volume 31% above its 3-month average. ServiceNow fell 5%, Salesforce 4%, on AI financing worries. Atoms the AI-debt story can't touch vs bits priced on monetizing it.
Today's chip selloff is framed as AI-debt jitters. Underneath sits a bigger number: the S&P's CAPE ratio near 41, the second-highest reading in 150 years, beaten only by the run to 44 before the 2000 crash. Valuation extremes measure risk. They don't call the top.
Twilio, Snowflake, Okta, Palo Alto Networks, MongoDB, Zscaler and CrowdStrike are all down 7-11% today, 13-19% over five sessions. That is not seven separate stories. It is one multiple compressing across a factor. $TWLO leads, down 11% at 3.8x its average day.
Dell fell 11.3% Thursday, 3x its average daily move, after a Barclays downgrade. The downgrade doesn't dispute AI server demand. It argues the revenue growth priced as adoption is mostly pass-through from rising component costs, not more enterprises running AI. $DELL
$WFC trades at 1.75x tangible book after shedding its Fed asset cap. Bank of America, despite being bigger, sits near book value: $500B+ of its bonds still yield under 2% while the 10-year sits near 5%. Same Q3 season, opposite verdicts on balance sheet quality.
Oracle fell 8.83% Wednesday on one hotter-than-expected CPI print, the sharpest hit among AI infrastructure names. Its debt is heading toward $167B to fund the buildout. A growth story priced off a rate environment stops working the moment that environment moves.
Oracle fell 8.8% Wednesday, 3x its average daily move, the day a hot CPI print pushed Treasury yields higher. The AI capex debate has been about demand. Oracle's buildout runs on debt heading toward $167B, and that bill reprices every time rates move. $ORCL