I don’t own the neoclouds because the whole story comes down to GPU useful life and selling us the wrong metric EBITDA.
$CRWV , $NBIS, $IREN all look much better on adjusted EBITDA because D&A and interest sit below the headline. https://t.co/n6mVCIOyem
I analyzed what would need to happen for Reddit to reach a $300B market cap by 2030.
For $RDDT to get there, the math requires:
• ~$15B revenue
• ~50% adjusted EBITDA margins
• ~$6B free cash flow
• The market still paying ~20x EV/Sales
https://t.co/BdI0W5kiXF
$MSFT at 21x forward earnings with 19% revenue growth and $625 billion in contracted obligations. The CapEx panic has compressed the multiple to levels not seen since 2016. Copilot is 3.3% penetrated across 450 million seats. I see huge upside in MSFT if Copilot seats grow a bit
@Ritholtz Expectation of higher rates is bad for gold. If real rates, led by higher expected interest rates increase, the opportunity cost of holding gold increases
There is a number that explains the entire copper bull-bear debate: 12 vs 27 tonnes of copper per megawatt of data center capacity. One counts what is inside the building. The other counts what it takes to power it. The gap between them is 450,000-840,000 tonnes per year of "hidden" demand. https://t.co/m777BSLTrE
The EIA’s March oil forecast is $21/barrel higher than February’s — and it already assumes the Strait of Hormuz reopens by April. The market at $94 disagrees. Traced what that means for hyperscaler margins, the Fed, and your portfolio:
https://t.co/CrMHniZrrX
If hyperscalers cut CAPEX 10-20%, the hyperscaler stocks go up. And the infrastructure supply chain. $CRDO, $ALAB, $NVDA, $AMD, $MU goes down hard.
The companies funding the buildout benefit from cutting it. The companies feeding the buildout have nowhere to hide. https://t.co/wJA8XyfXk3
$FCX made $25.9B in revenue last year at an average copper price of $4.75/lb, while its biggest mine was literally buried under mud.
The Grasberg mud rush in September 2025 knocked out 20% of FCX's copper production. Output fell from 4.2B lbs to 3.4B lbs. And revenue still grew.
Now Grasberg is restarting. Copper is at $5.90/lb. And nobody seems to be doing the math.
Production was flat for three years, then dropped 20%. Revenue didn't care because the copper price kept climbing.
The gap between today's spot ($5.90) and last year's realized price ($4.75) is $1.15/lb. On normalized production, that's roughly $3.8B in incremental annual cash flow that hasn't shown up in a single quarterly earnings print yet.
Grasberg at full capacity is the lowest-cost copper operation on the planet. It also produces 1.3 million ounces of gold per year. At $4,000 gold, that's another $5.2B in revenue from a single asset.
Sometimes the most interesting setup is just a company getting back to normal.
$CRWD spent years being told Microsoft would eat their lunch by bundling security "for free."
This week Satya Nadella and George Kurtz spoke to CrowdStrike's sales team together. Falcon is now on the Azure Marketplace.
The most dangerous competitor became a distribution channel.
$RIO2 doubled in 6 months and trades at 4.0x forward EBITDA on consensus gold/copper. Fairly valued? At consensus, sure. But consensus has gold at $4,742 while spot is $5,293. https://t.co/ryUgUuLjxP
That 14% Agentforce deployment rate is the number to watch. Klarna tried the "rip out Salesforce" playbook and ended up replacing it with other SaaS + an AI layer on top. If deployment catches bookings, the AI-as-interface-layer thesis wins. If not, margin compression eats everything.
@KobeissiLetter Hedge funds are record-short software but the Klarna case shows something weirder, they didn't replace SaaS with AI, they replaced expensive SaaS with cheaper SaaS. AI became the glue, not the replacement. Are shorts pricing in displacement that isn't actually happening?
The gap between "AI can answer a question" and "AI can run a process that requires mid-stream judgment in a regulated environment" is measured in years, not quarters. Most CFOs are not going to bet their payroll system on closing that gap early. https://t.co/SL0O0ZGUnb
@BenEsmael1 Honestly, if I'm the CTO? Per seat, every time. Per token pricing is a nightmare for budget planning. You're essentially signing up for a variable cost that scales with usage you can't fully predict.
$NOW The underappreciated problem with SaaS companies selling AI features: every AI action has a real inference cost paid to hyperscalers. Old model was zero marginal cost per user. New model has variable costs the company doesn't control. That's a permanent change in business quality, not a temporary margin headwind. https://t.co/y9WtDIxnBH
@BenEsmael1 Per seat to per token is the right direction but look at who actually captures the margin on the token side. ServiceNow runs on Azure and uses Anthropic/OpenAI models, they're a middleman in the inference value chain now. Revenue up, gross margin guided down to 82%.