Three percent of every dollar traded in the US stock market this year passed through Nvidia alone. Micron sits second at 2.8%, Tesla third at 2.1% and SanDisk fourth at 1.5%. Put the ten most traded names together and they average 16.4% of notional volume year to date. Why a memory maker ended up here Memory was treated as a cheap component until last year. TrendForce put the rise in conventional DRAM contract…
Why it matters. Heavy trading and easy trading are not the same thing. When money crowds into names with wide spreads and thin resting orders, the cost of getting out in a shakeout runs higher than most people assume.
What keeps a $500 billion loan book safe is not the GPUs. It is the credit of the companies renting them. Bloomberg reported that the six firms designed the structure around separate entities able to issue tens of billions of dollars of bonds at a time, with Goldman Sachs running the public bond sales. Why a supermarket chain belongs in this story Meru reads the structure as a sale-and-leaseback. You sell a build…
Why it matters. What turns this lending sour is not a fall in GPU prices but a wobble in the credit of buyers like OpenAI. The structure asks you to watch the tenant rather than the collateral, so a long contract term on its own explains nothing about safety.
The US Treasury finalized a rule on August 11 that permanently ends beneficial ownership reporting for American companies and American persons. Information already filed by US persons will be deleted from the database. A beneficial owner is the person who actually owns or controls a company even when their name never appears on its paperwork. Stacking companies to hide that name is the standard move in money launde…
Why it matters. The only US register built to trace anonymous companies has been emptied of its subjects by rule rather than by changing the law. The cost of checking does not vanish; it shifts onto banks and financial firms.
The Kobeissi Letter reported that stablecoin card spending hit a record $1.03 billion in July. An independent tracker counting the same month came to $748.7 million. Where the two numbers separate The two counts also disagree on growth. Kobeissi wrote +16% month over month and +200% year over year. Paymentscan put the same month at +19.1% and +144.7%. The gap comes from what gets counted. a16z, working through…
Why it matters. Two counts of the same market in the same month differ by 28%. A record only means something once you know whose count was copied.
Supermicro told the market it expects $65 billion to $72 billion in revenue next fiscal year. The number analysts had been working with was $52.5 billion. CoreWeave and Lumentum reported the same evening, and all three came in above what the market had penciled in. Serenity read the evening as proof that every layer of the AI buildout is running and that it has started to show up in earnings. Where the three comp…
Why it matters. It was the day AI capital spending showed up as booked orders and revenue, but the same releases also showed how much capital it takes to turn those orders into machines.
The number of homes listed for sale in the United States has risen to 1,126,252. That is the July 2026 figure, up 2.1% from June and up 2.1% from a year earlier. The same report contains one more sentence that changes how the million-plus figure reads. That inventory is still 11.6% below typical 2017 to 2019 levels. Homes have not piled up. The shelves that emptied out after the pandemic are about half refilled.…
Why it matters. More listings do not lower the monthly outlay if the rate rises with them. Whether housing has loosened is decided by the mortgage rate, not the inventory count.
Data center lease and equipment purchase commitments at Big Tech have been tallied at $2.6 trillion. The figure combines leases that have not yet started with equipment and services contracts already signed. Adding up the disclosures of five companies, Reuters found more than $1 trillion in leases that have not yet commenced. Put the equipment commitments on top of that and you arrive at the $2.6 trillion. Why it…
Why it matters. Commitments in a footnote do not show up in today's leverage, yet they move onto the balance sheet all at once when the buildings are finished. Reading only the current statements misses that lag.
Word has gone around that ChangXin Memory Technologies, or CXMT, has pushed the yield on its 17nm DDR5 past 90 percent. This is not a company disclosure. It is a Chinese press report being passed along. Yield means how many usable chips come off a single wafer. Past that line, nine out of every ten come off saleable, and at that point a price cut still leaves a margin. In DRAM, yield is the cost structure itself.…
Why it matters. How far Chinese DRAM has reached into which brands and which models decides when Samsung, SK hynix and Micron start losing pricing power.
The Bank of Korea started buying gold again on August 3, 2026. It had not added a single gram since February 2013. The report adds that the bank has also started buying overseas-listed physical gold ETFs. Bullion and listed products are going into the same basket now. What stopped the buying was not the market The bank began in July 2011, taking 25 tonnes at $1,544 an ounce, and filled out 90 tonnes over five pu…
Why it matters. Central bank buying is the least price-sensitive slice of gold demand. Whether that slice is growing or shrinking sets the floor under the price right now.
Samsung Electronics says it will use High-NA EUV, the next generation of lithography tools, in volume production starting at its 1nm process. That pushes back a plan to bring the tool in at 2nm and 1.4nm, and 1nm volume production is set for around 2030. Park Chang-min, a Master at the company, laid out the roadmap on August 11 at a next-generation lithography conference in Suwon. He said Samsung had hoped to apply…
Why it matters. Years sit between buying a tool and selling chips made with it. Samsung contracted High-NA for 2nm and now says volume use starts at A10, while Intel has set 2028 volume manufacturing on the same machine. The race at the leading edge turns less on who owns the tool than on when it goes onto a line.