Three negative headlines in the last 24 hours: Amazon servers waiting in warehouses, a giant Korean IPO plunges 32%, and an AI model that threatens software companies again ๐จ
We sat down to analyze what actually happened overnight, what is fact, what is an estimate, and what is a rumor,
and above all, what the true meaning of these things is!
$AMZN $NVDA $DELL $CRWV $NBIS $ADSK
Three negative headlines arrived overnight, from Seoul to San Francisco,
At first glance, there is no connection between them.
But we connected the dots, and 2 of them tell the same story: the market is no longer asking if there is demand for AI, it is asking who pays, when, and how long the money is stuck before it starts generating revenue. ๐ต
And the third headline asks a different question: what AI is doing to niche software companies. ๐จ
Let's start from the beginning:
1๏ธโฃ Amazon: there are servers, there is space, there is no electricity ๐จ
Overnight, Business Insider revealed internal Amazon emails from August,
According to one email, Amazon is expected to receive server racks with Nvidia chips worth about $5 billion in August, and then another $3 billion each month, "for the foreseeable future".
So where exactly is the problem? ๐ฑ
According to people familiar with the matter, some AWS sites have space for the racks, but not enough electricity to power them. โ ๏ธ
So the equipment waits in intermediate warehouses until the site is ready.
And then comes the problematic sentence:
At least one supply chain team at AWS did not have a "clear predictable path" for bringing the racks into the data centers. โก๏ธ
And now to the detail that most headlines missed: who pays during the waiting time?
โ Usually Amazon pays for a rack only when it enters the data center, until then, the suppliers bear the cost of the equipment in transit.
โ But the emails describe a new process:
Amazon takes delivery of the racks earlier, before installation. ๐งฐ
โ The meaning, according to the report: Amazon spends cash earlier, and any delay in connecting to electricity postpones the moment this equipment starts generating money for it. ๐ต
Why is this sensitive right now? ๐จ
According to Amazon's Q2 report, its free cash flow for the last 12 months turned negative: an outflow of $7.6 billion, compared to an inflow of $18.2 billion a year earlier. ๐ก
The reason?
Purchases of equipment and property rose by $66.1 billion in a year, mainly for AI.
Amazon, for its part, said that storing ready equipment near the sites until there is electricity is a standard operating procedure, and rejected the claim that it has no installation plan. ๐
2๏ธโฃ Elice: the biggest IPO of the year in Korea, and -32% on the first day โ๏ธ
Elice is a Korean "neo-cloud": a company that builds data centers and rents out AI computing power, like CoreWeave in the US. ๐บ๐ธ
On Tuesday morning, Seoul time, it began trading on the Kosdaq exchange, after raising about 201 billion won, the largest IPO on this exchange this year. ๐ต
And look what exactly happened here!
โ IPO price: 90,500 won, the top end of the range.
โ Demand: institutional bodies requested 388 times the offered amount, and the public 408 times, with more than 10 trillion won in orders.
โ And then, in trading, a drop of up to 32%, to 61,400 won.
A value of about 1 trillion won dropped in the first hours to about 700 billion.
Meaning, despite the huge demand, in actual trading the sellers dominated right from the open...
3๏ธโฃ SKY-001 model: an AI model targeting the software applications niche ๐จ
Overnight, Sam Pasupalak, CEO of Skyfall AI, introduced SKY-001, which the company calls "the first engineering world model". ๐ก
The post gained about 2.5 million views.
The team came from Maluuba, the deep learning lab that Microsoft acquired. ๐ฑ
The model receives an image or sketch and returns all the construction stages of the product in design software:
parts, components, and operations; in structures, it can turn a building outline into apartments and rooms. ๐
In short, a revolution. โก๏ธ
And the numbers, according to the company: matching accuracy of 79.7% in 3.9 seconds and at half a cent per task, compared to 79.1% in 199 seconds and at 75 cents for GPT-6 Astra. ๐ฑ
Why does this interest Wall Street?
Software is a very profitable niche; if one engineer with AI does the work of several, investors ask: how many licenses will be sold in the future? ๐ฉ
The headlines sound very bad, but here is the other side of the coin:
โ Amazon is still ordering $3 billion a month in Nvidia racks.
A full warehouse is a sign of a delay, not of demand disappearing,
and in Q2, AWS revenue rose 37% to $42.2 billion. ๐ต
โ Elice itself is profitable: in the first half of 2026, revenue of 34.4 billion won and operating profit of 10.3 billion; the drop was on the first day, during trading, and not the closing price. ๐ต
โ The SKY-001 model is a company claim, not a tested product; moving from a working tool to a replacing tool takes years, and customers do not replace software in one day either. ๐ต