Everyone is posting the same chart today: open source models went from 28% to 62% of tokens on Vercel in two months.
Majority opinion is that closed models are losing. The commodity tier of AI work exploded, and that tier was always going to open source. Boilerplate migrates to whatever is cheap. Frontier work still costs real dollars that the token charts are not showing.
@PolymarketMoney Every year of demand outpacing production means the factories mostly don't exist yet. Humanoids are lining up to be the next capex story after AI datacenters
@BrianFeroldi The switching costs line is the one I weight most. Everything else on the list can be true and the company still gets commoditized out from under you
@randgroup Anthropic is reportedly exploring the largest IPO in history though, and this month's consolidation wave keeps pushing private AI names toward public markets. Retail gets in late, but not never
Michael Burry reportedly bought $NVDA calls this week to hedge his short ahead of tonight's print.
Shorting AI supply is a bet that demand breaks before capacity does. Apparently even the author of that bet wanted insurance on the timing.
@Benioff@salesforce@AnthropicAI@Dreamforce Enterprise distribution is the one thing model labs cannot build alone. Salesforce sells to companies that will never touch an API key
@zerohedge Oil futures barely moved Sunday night when the strait headlines first hit. Shipping rates are where the risk shows up first, and apparently it already has
@weijinnnnnn The how-long question is why I watch the funding mix. Incremental debt went from 9% of hyperscaler capex in FY24 to 32% this year. The duration risk already moved from equity to lenders
@EnergyAlphaCo The constraint list keeps growing. Memory got repriced this summer, optics supply is already tight, and now it's megawatts. Demand keeps finding the next bottleneck
@RealNickMugalli Every hyperscaler says its custom chip beats Nvidia on the benchmark it was built for. The $108 billion ex-China guide says customers keep buying both anyway.
@krshya@KobeissiLetter Tonight's print argues the pass-through works though. Margins up 250 basis points in the same quarter they told customers prices rise 15%. So far the memory bill lands on buyers, not on Nvidia
The guide is the wilder number. $NVDA pointed to 108 billion dollars next quarter, and that excludes China data center compute entirely.
Their whole fiscal 2024, all four quarters combined, was 61 billion.
One quarter now beats that full year with an entire country left out of the number.
Everyone is posting $nvda revenue beat. The number that settled something is further down the release.
Adjusted gross margin printed 75%, up 250 basis points from a year ago, in the same quarter Nvidia told customers server prices are rising 15% because of memory costs.
Customers ate the increase, Nvidia protected its margin, and the memory bill keeps rolling downstream. Clouds next, then everyone else.
@randgroup The App Store spending drop this week might be the first visible dent. Between web billing and AI subscriptions sold outside the store, the 30% is losing traffic from both directions