Key Takeaway
AI infrastructure is still expanding across cloud, memory, custom silicon, power, cooling, and networking. At the same time, the market is testing the trade: crowded positioning, capex discipline, deleveraging, and rate uncertainty are all back in the conversation.
The signal remains broad, but the tape is not simple. Durable AI demand is on one side. Fragile positioning is on the other.
The AI Infrastructure Signal Is Still Broadening
The strongest signal came from the AI infrastructure stack.
Microsoft reported revenue of $90.0B, adjusted EPS of $4.74, and operating income of $40.6B. Azure & Other Cloud revenue grew 43% year over year, while Microsoft Cloud revenue reached $59.3B.
Satya Nadella said Azure revenue surpassed $100 billion for the first time, and Microsoft 365 Copilot reached over 30 million paid seats.
His framing was direct: Microsoft is “advancing the frontier on the cost-to-outcome curve,” helping customers turn tokens into business results.
Memory Is Becoming Central to the AI Trade
That demand is also showing up in memory.
Samsung’s Q2 showed revenue of $118.1B versus $119.2B expected, up 130% year over year. Operating profit was $61.7B versus $60.8B expected, up 1,814% year over year. Samsung expects strong H2 memory demand led by continued AI infrastructure spending and broader agentic AI adoption.
The DRAM market structure is part of the story. Shay Boloor’s market-share note put Samsung at roughly 39%, SK hynix at 29%, Micron at 22%, and CXMT at 8%.
With Samsung, SK hynix, and Micron controlling about 90% of the market, the argument is that the industry now has more pricing discipline as capacity shifts toward higher-margin HBM through take-or-pay contracts and price floors.
CXMT is still a real commodity DRAM competitor. It is China’s most valuable publicly traded company at a $500B valuation. But the key caveat is that it lacks meaningful HBM exposure.
That same memory thesis showed up across multiple Micron-focused posts. One framed AI as not just a compute story, but “a memory story,” arguing that every AI model, LLM, bot, robot, computer, CPU, and GPU has the same mission: compute faster, which requires more memory.
Another claimed $MU will make more money than $META, $TSLA, $AMZN, and possibly more than $GOOG, $GOOGL, and $AAPL. Those are aggressive claims, but they fit the broader market narrative that memory is becoming a bottleneck, not an afterthought.
The Stack Goes Beyond GPUs
The AI data center buildout lens widened the trade even further. One brief argued AI-driven data center demand could nearly quadruple to 194 GW by 2035, with beneficiaries across $TSM, $VRT, $NVDA, $AMD, $MU, $SKHY, $AVGO, $MRVL, $CRDO, $ALAB, $OKLO, $SMR, $BE, $CEG, $VST, $LITE, $COHR, $AAOI, $ASML, $AMAT, $LRCX, and $KLAC.
Daniel’s Broadcom post added the custom-silicon angle. OpenAI and $AVGO reportedly unveiled Jalapeño, OpenAI’s first custom AI chip, designed end-to-end in 9 months with help from OpenAI’s own AI models. Hock Tan’s view was that every frontier AI company will eventually design custom silicon with Broadcom.
So the AI infrastructure story is not only GPUs. It is cloud revenue, memory supply, HBM, custom chips, data centers, power, cooling, networking, and equipment.
Earnings Were Strong, But Not Every Stock Got a Clean Pass
Earnings strength did not mean every stock got a clean pass.
Meta reported Q2 revenue of $60.8B, up 28% year over year. Advertising revenue was $59.4B, up 27% year over year. Daily active people reached 3.60B, up 3% year over year. Q3 revenue guidance was $61B to $64B.
But EPS fell 13% year over year to $6.18. Reality Labs lost $4.62B. FY26 capex guidance rose to $130B to $145B.
Mark Zuckerberg said AI is accelerating Meta’s core business and opening new enterprise opportunities, but the market focused on the EPS miss and continued spending.
Charlie Bilello sharpened that concern: Reality Labs has now lost $87 billion cumulatively since 2020.
Robinhood’s Report Was Cleaner
Robinhood’s report was cleaner.
$HOOD posted revenue of $1.31B, up 32% year over year. Net income was $573M, up 48% year over year. Adjusted EBITDA was $741M, up 35% year over year.
Platform assets reached $369B, and net deposits were $21.7B, an all-time high. Gold subscribers reached 4.8M. ARPU rose 24% year over year to $187. The company bought back 4.4M shares.
Vlad Tenev’s quote centered on “making everyone an owner,” with product velocity across Robinhood Chain, Robinhood Ventures, and Trump Accounts.
Macro and Positioning Are Testing the Tape
The earnings picture had real growth, but the market was also dealing with macro pressure.
The Fed held rates unchanged for the fifth straight meeting, voting 9-3 to keep the benchmark rate in the 3.50%-3.75% zone. Hammack, Kashkari, and Logan dissented in favor of a hike. Kobeissi called it the longest Fed pause since the 2008 cycle.
The broader market recap showed fragility underneath the surface. Retail investors sold a net $243M of single stocks, the largest one-day outflow since the COVID crash.
JPMorgan said Korea’s KOSPI is down nearly 40% from its June 22 peak after deleveraging, with leveraged ETF unwinds complete and hedge funds roughly 90% done.
Volatility Is Still Part of the Trade
That is the tape Tom Lee was responding to. He argued that once corrections in memory chip companies, AI stocks, and the S&P 500/Nasdaq-100 end, markets will hit “MUCH MUCH HIGHER NEW ALL TIME HIGHS.”
He cited ETFs including $DRAM, $SMH, $QQQM, $VOO, $SPMO, $XLK, $CHAT, $AIS, $AIPO, $SOXX, and stocks including $MU, $SNDK, $NVDA, $AMD, $LRCX, $GOOG, $MRVL, $TSM, $AVGO, and $NBIS.
Heisenberg’s $NBIS note captured the other side of that volatility: down 30% peak to trough over two days, with the question, “Buying opp soon?”
Outside AI
Outside AI, the SpaceX-related items were a risk-management reminder.
One $SPCX article argued there are “no worries” adding on dips, citing Starlink advertising, Grok, XMoney, Starship, and “pure Elon exposure.” The author currently holds about 85% $TSLA and 15% $SPCX, with a plan to trim Tesla and move closer to 50/50 if $SPCX falls below $100.
The value-investing SpaceX post was more skeptical. It said SpaceX is trading below its IPO price and summarized the lesson this way: “Wall Street is not your friend.” The IPO was described as four times oversubscribed, with lockups starting next month and possible “even more selling.”
The bottom line was: “Use Wall Street wisely. Don’t be used by Wall Street.”
Oil was a separate cyclical bet. Var Energi has a 115 billion Norwegian crowns market cap, about $12 billion, and cash flow from operations of 2.1 billion. At oil at 100, the speaker said companies like this could pay roughly $2 billion in dividends per year, about a 17% return, but that could quickly fall to $1 billion at lower prices.
The conclusion was that it is not yet interesting from a cyclical value perspective until oversupply and “peak oil” concerns are priced in.
The risk watch also included geopolitics and the Fed.
The geopolitical item from @geopoliticaleconomyreport argued that the U.S. paused the Iran war because of depleted Patriot interceptors, rising casualties, and energy disruption around Hormuz, Yemen, the Red Sea, and Saudi oil exports. It claimed the U.S. began with around 2,500 Patriot interceptors, used at least 1,500 since February 28, has less than 40% left, and spent about $6 billion on interceptors priced at $4 million to $5 million each.
The Fed risk item from @clearvaluetax9382 argued the press conference offered “nothing” new: no rate change, no forward guidance, and continued money printing. It highlighted Kevin Worsh saying the Fed is “on the job” and “will deliver,” while also saying June CPI mattered “not much” and that the Fed is sticking with its 2% PCE objective while watching broader inflation data.
Bottom Line
AI infrastructure demand still looks strong across Microsoft, Samsung, Meta, memory, custom silicon, and data-center commentary. At the same time, the market is testing valuation, leverage, spending discipline, and crowded trades.
That is the tension to track: durable AI demand on one side, fragile positioning on the other.
This is analysis and context, not financial advice.