everyone saw nvda's revenue number in august and moved on
the more interesting stuff was further down. went through the whole release line by line, took me like 20 min
this is roughly how i read any earnings report, using nvda q2 fy27 as the example
coreweave beat on revenue and raised the year. cool.
next quarter's interest bill is guided at $1.05-1.15B. this quarter's adjusted operating income was $379M.
the ai boom is real. the lender just eats before the gpus do. tell me why i'm wrong
@JonathanHatzor that's the real stress test. prepay is a day-one cushion, an SLA miss is a month-twelve problem.
if capacity slips, the customer gets leverage right when the lender's cushion is shrinking. tonight's print didn't really answer that part
bernstein's neocloud math: if customers prepay 20-30% of the gpu capex, debt coverage gets above 1x.
so the safety margin on AI data center debt is... the customer paying early.
coreweave reports tomorrow. bulletproof credit or vendor financing with extra steps?
@DanielL43507016 yeah exactly. cash fcf answers "what did they actually spend this quarter." lease commitments answer "what are they already on the hook for."
the $18.6B is the first one. the second one is the part nobody wants to put on a chart
uncle sam just borrowed $42B for 10 years at 5.02%.
bid-to-cover 2.46. so the buyers showed up, they just all wanted a discount first.
and people still think mortgage rates are about to fall. is this fine, or is the bond market starting to send invoices?
3/3 data center is a market platform, not a formal reportable segment. those are compute & networking and graphics.
my bio says i read 10-Ks. apparently the definitions would also like some attention.
https://t.co/F2ju2oXrlE
everyone saw nvda's revenue number in august and moved on
the more interesting stuff was further down. went through the whole release line by line, took me like 20 min
this is roughly how i read any earnings report, using nvda q2 fy27 as the example
2/3 even nvda has to pay taxes. rude.
nvda attributed the sequential operating-cash decline to working capital AND cash taxes. DSO rose from 45 to 60 days under extended terms for certain investment-grade customers.
https://t.co/rRpxrGYwGf
coreweave reports tonight. its own q3 guide: $860-940M of interest expense vs $200-260M of operating income.
and it reports on the day a hot cpi sent rates higher.
that's not an AI stock, that's a bond with GPUs attached. bull case goes in the replies, i'll wait.
october cpi: gasoline +1.0%.
also october cpi: gasoline -0.7% before seasonal adjustment.
so gas got cheaper and the inflation report says it got pricier. meanwhile core printed 0.3% for a third straight month, which is the part the fed reads.
noise or problem? pick one.
morgan stanley says the midterm selloff just pushes some AI projects from 2028 to 2029, which "has the potential to elongate the AI capex cycle."
only on wall street does a delay make the story longer and therefore better.
overreaction or cope?
citi's economic surprise index just hit its highest since august 2023. ny fed 1-year inflation expectations: 3.9%, didn't budge.
so the economy keeps beating forecasts and prices refuse to cool.
whoever's still banking on fast rate cuts, walk me through the plan.
anthropic just agreed to pay $3.4 billion for gas-fired generators (FT). largest known power-equipment buy by an AI lab.
the "it's just software" pitch now comes with a turbine supplier and a fuel bill.
AI lab or utility with a chatbot? pick one.
Anthropic has agreed to pay $3.4 billion for gas-fired generators for data centers, the FT reported, in what it described as the largest known single purchase of power-generation equipment by an AI lab https://t.co/xGbufpvmAY
valero is paying $320.5 million for a shuttered houston refinery and saying it'll put roughly 165,000 barrels a day back online by end-2027.
if gas prices are the midterm story, the boring answer is still refining capacity on the ship channel — not another strait headline.
capacity or vibes.
q3 gdp printed 3.4% with a 3.32 percentage-point drag from net exports.
that is not "trade saved us." that is domestic demand hauling a boat anchor and still clearing 3.
if the next print loses the domestic leg, the headline gets ugly fast.
domestic strength or trade drag. pick one.
the new pitch: mandatory code review is over because the agent can make a slide deck explaining its own PR.
cool. who reviews the slide deck?
the thing that wrote the bug grading its own homework isn't a process. it's a vibe. change my mind.
read the actual brookings paper. it's worse than the tweet.
it's $3.7T a year by 2032, 9.2% of gdp. openai + anthropic make ~$100B now. that's ~80% growth every year until 2032.
and that assumes gpus last 6 years. make it 3 and it's $6.0T.
bubble or genius? pick one.
To justify today’s data center buildout, AI would need to bring in about $3.5 trillion a year by 2032, per a new Brookings paper.
That’s 8.8% of U.S. GDP, nearly as much as Americans spend on food (9.1%), and far more than on energy (5.1%).
"there can't be an ai bubble, we have shortages" is just "this time is different" with a supply chain.
shortages are what every capex boom looks like right before the glut. that's literally how the glut gets built.
real demand or bubble? pick one.
"per AFR" had half the timeline picturing the australian financial review.
it's americans for financial reform, an advocacy group, and that line is lifted word-for-word from its own report's executive summary.
doesn't make it wrong. but "an advocacy report says" hits different than "a newspaper found."
subprime 2.0 or recycled analogy? pick one.
BREAKING: "Speculative data center debt is being repackaged and sold to pension and retirement funds, university endowments, and other institutional investors in a manner eerily reminiscent of the subprime bubble," per AFR