大漠哥

A Bitcoin mining company has already written the words “bankruptcy” into its SEC filings.

Vulcan Infrastructure & Power still has a debt of $33.1 million due on October 31.

As of the end of June, it had only about $3.2 million in cash, plus about $6 million in digital assets—so the total available resources are roughly $9.2 million.

──── ✦ ────

Now it is pinning its hopes on a $39.4 million PIPE financing, planning to use most of it to repay the debt.

The problem is timing.

The PIPE must be completed no later than October 10, while the debt matures on October 31—just 21 days in between.

If the financing isn’t closed, during those three weeks it will have to raise money again, negotiate an extension, restructure the debt, or sell assets.

And with its current cash levels, directly repaying $33.1 million leaves very little room.

──── ✦ ────

In its SEC filing, the company has already spelled out the worst-case scenario clearly:
If the financing fails and no other solution can be found, it may face debt default, a restructuring, or even seek bankruptcy protection.

This is also one of the reasons I pay particular attention when looking at mining companies.

BTC’s up-and-down movements matter, of course. But when the debt is due, how much cash is still on hand, and whether the next round of financing can be lined up in time—sometimes those factors determine a company’s survival faster than hash-rate growth.

Mining machines can keep running, but creditors won’t wait forever. #大漠茶馆 $BTC

And there’s also SanDisk 😄

US stocks went crazy again 😄

Today, in a WeChat group, I saw a subject leader share a passage. I read it and felt deeply moved, so I wanted to share it with everyone.

He said that when facing a completely unfamiliar field, it’s hard at first to genuinely develop deep interest. At this point, you need a bit of “belief”: since so many smart people are willing to spend their whole lives studying it, there must be something worth understanding and uncovering within it.

I think this sentence is especially suitable for the AI era right now.

AI is lowering the barriers to acquiring knowledge more and more, making it far easier to enter an unfamiliar field than ever before. But what ultimately determines how far someone can go is still curiosity and focus.

Besides, interdisciplinary ability will become increasingly important.

People who understand technology should learn about finance, economics, and psychology. Investors should study AI, industries, and history. Those who work in AI should further understand business and people’s needs—many problems will then reveal new perspectives.

So I’m becoming more convinced that when you’re young, you should try to get exposed to as many different fields as possible. Don’t worry too much at first about whether you can go deep—just go and look, go and understand, and gradually find the direction that makes you willing to keep investing your time.

No matter what field you go into deeply, you’ll eventually encounter boredom and difficulties. It’s hard to stick with it for many years just by forcing yourself. Interest is the most lasting motivation.

AI can help us learn faster.

But always stay curious—know what you enjoy researching, and at the same time have the ability to continuously step into new, unfamiliar areas. I think that’s the truly scarce capability in the AI era.

This passage comes from a subject leader’s share in a WeChat group. It was very inspiring, so I organized my own understanding and wanted to share it with everyone. #大漠茶馆

Tokenized stocks—this segment has recently started to accelerate noticeably.

Over the course of one month, the number of holders doubled directly to 1.31 million. Monthly transfers surged 179%, reaching $23.13 billion. Active addresses also grew by 34.62%, approaching 572,000.

──── ✦ ────

What’s even more worth watching is the asset distribution scale—it only increased by 5.9% and is currently about $2.38 billion.
In other words, the user base and trading activity are running much faster than the asset size. Big money hasn’t flowed in at large scale yet, but “people” and “trading habits” are already starting to migrate.

──── ✦ ────

Among the major players, Ondo currently leads with $872 million, xStocks has $557.8 million, and bStocks has already reached $521.8 million.

bStocks launched only in June; within two months, its scale is almost catching up to xStocks. This pace shows that exchange traffic and ready-made user entry points are very important in this race for tokenized stocks.

──── ✦ ────

I increasingly feel that the real incremental growth in tokenized stocks may have just begun.
The 1.31 million holders already prove that there are people willing to trade stocks on-chain. Next, what we need to see is whether the $23.1 billion trading activity can continue to convert into a truly settled—i.e., sustained—asset base. #大漠茶馆 $ONDO

SEC has recently filed a case that is a textbook example of an “acquaintance-circle” investment scam. Over 87 investors were involved, with roughly $47 million in funds at issue. The victims were primarily drawn from the same ultra-Orthodox Jewish community. The SEC alleges that more than $11 million was diverted for personal use, ultimately causing investment losses exceeding $25 million.

──── ✦ ────

What is most alarming about this case is that trust replaced due diligence. If a stranger asks you to invest several hundred thousand dollars, you would very likely check the company, the assets, and the flow of funds. But if it’s a friend, a neighbor, or someone from the same circle who introduces it, many people’s defenses naturally drop.

──── ✦ ────

What makes family- or kinship-based scams so terrifying is exactly this: “He’s one of us” is often more persuasive than any promise of returns. In the future, when you encounter an investment introduced by an acquaintance, ask yourself first: if you remove the introducer’s name, would I still dare to invest in this project?

The case is still in the SEC accusation stage; the final outcome will be determined by the court. #大漠茶馆

Why does it feel like this is good news? If he’s involved and gets mixed in, it probably isn’t a good thing... #大漠茶馆 #特朗普立场

Orders exceeding $100 billion—why does CoreWeave still need money?

CoreWeave’s Q2 revenue hit $2.58 billion, up 112% year over year, with a backlog of orders reaching $104.2 billion. After entering Q3, the company also secured new customer commitments of more than $25 billion—leaving near-term computing capacity almost in short supply.

At the same time, however, CoreWeave has raised its 2026 capital expenditure forecast to $35–39 billion. The reason is simple: customers are buying computing power for the coming years, but the company has to prepare chips, servers, data centers, and power in advance.

The more orders there are, the more money must be invested upfront.

This also shows that AI competition is moving into the next phase. It used to be about competing on models and chips; now it’s also about competing on financing capability. Whoever can secure cheaper long-term capital, who can withstand massive depreciation and interest, will have the chance to turn orders into real profits.

Next, when judging AI infrastructure companies, don’t just look at orders—also consider debt costs, capital expenditures, and free cash flow. Orders answer “whether anyone will use it,” while financing answers “who pays first.”

Not investment advice.#大漠茶馆

At a meeting with industry executives, Trump announced something: he would invest $3 billion in key U.S. domestic mineral and mining industries to reduce reliance on China’s supply chain.

It’s a lot of money, but there’s one detail I find even more interesting than the figure itself. In attendance was US Antimony. After the meeting, its chairman and CEO, Gary Evans, in front of reporters, told Bloomberg how much he hopes the government will put into his company.

A CEO being so eager to state the exact number he wants usually indicates one thing: how that $3 billion is going to be allocated hasn’t been decided yet. Whoever speaks first gets in line first.

As for what US Antimony actually does, and how this batch of funds ultimately ends up with specific companies—those things are still unclear. We’ll have to wait until the subsequent allocation plan is released to find out.

Not investment advice. #大漠茶馆

There’s a big whale on-chain who sold today—address bc1qdj.

Within 6 hours, this address split 1,274 BTC to three OTC desks: Cumberland, FalconX, and Galaxy Digital, totaling roughly $81.5 million.

I noticed one detail worth mentioning. It’s an OTC sale, not a direct order on an exchange to dump and crash the market. OTC orders aren’t displayed on the order book, so it’s less obvious from the order book that someone is selling. This kind of selling typically means the seller wants to clear out, but without drawing too much attention.

As for who’s behind this address, and why they’re selling—on-chain data alone can only show these facts so far. We can’t see anything else. More conclusions will have to wait for subsequent address flows or news to come out.

Not investment advice. #大漠茶馆 $BTC

Is anyone else stuck on writing tweets like I am?

There’s a mountain of materials. When it’s time to write a tweet, you have to quickly find the parts you want. Later on, I tried ChatGPT and Gemini—reasoning was decent—but the “AI flavor” was too strong. No matter how you ask it to change, it keeps popping out lines like “No… but…” or “No… it’s…”—that classic scene you just can’t get rid of no matter how you train it. Reading it doesn’t feel like something a real person wrote.

But if you write everything by hand, efficiency is just too low. Other people can publish more than a dozen pieces a day; I can only write one or two. That’s kind of awkward.

Another thing, too. Have you ever thought about filming a decent-looking video? Now AI webtoon animations are everywhere. There are all kinds of tools, and you don’t know which one to use.

Let me teach you a method first—how to make AI webtoon animations.

Step one: you need a core positioning. What direction does this webtoon take? What’s the central idea? You have to write the story clearly first. Basically, it’s just like making a film: you need a script, then break the script into several story arcs— or call them segments—and complete them one by one.

Speaking of each segment, you must generate the images first. Just like cartoons: first create the images, then assemble them—one segment at a time—until you finally piece everything together into a complete whole. Along the way, of course, you’ll need all kinds of adjustments and revisions, including subtitles, narration, and so on, which all require careful polishing before it can take shape.

After talking for so long, I’ll recommend the tool CreaoAI to everyone. I’ve used it for a little over a month, and I created two AI agents.

One is dedicated to handling copywriting—helping me write tweets. The more you use it, the better it gets. In the end, it feels like man and sword become one. The other is an AI agent for video production. The Binance video I made and the videos for the Bored Ape community were all made with it.

In the comments, I’ll drop the invite code. If you’re interested, you can go to my profile and find it there.

Alright, that’s it—wishing everyone a great weekend.

How should I put it? I joined the Square in 2024.

As for earnings? At the beginning, I participated in all kinds of writing/essay contest activities. Almost every issue I joined, and there were basically rewards—sometimes as much as 800u, and other times around 200u.

Later, when the Square launched the Creators Task Board, I kept participating. One of the projects I took part in had a reward of about 3000u.

There were also later creator tips of 200 BNB, and each winner of the crayfish competition received one BNB.

Roughly speaking, all in all, 10,000u is definitely there.

Now the features on Binance Square have been pushed to the extreme: live stream tipping (the Square doesn’t take a cut), and content mining. Every now and then, I get surprised—there are all kinds of rewards for live stream deal-promotion too. Come on in—it's right here waiting for you. @CZ @Yi He

In the past few days, I’ve seen several AI news stories in a row, and I realized they’re actually all about the same thing.

Google and Amazon are continuing to ramp up AI capital expenditures; Anthropic has confirmed that it’s forming an internal chip team; ByteDance has released a native audio-and-video full-duplex model; Unitree Technology has kicked off an IPO. They may seem unrelated on the surface, but underneath, they point to the same trend.

After Google and Amazon released their earnings reports, many people focused on revenue and profit.
But what I care about more is that both companies haven’t slowed down their AI investment. Google continues to build out Gemini while also investing in Anthropic; Amazon continues to expand AWS and develop its Trainium chips. Different betting strategies, but both are betting that demand for AI compute power will keep growing in the future.

──── ✦ ────

If what cloud providers are fighting over is compute power, then what model companies are starting to compete for is control over the “initiative” behind compute.

When Anthropic sets up an internal chip team, I think the significance isn’t just about lowering costs. From renting GPUs to starting to research its own chips, what it truly wants to secure is the most critical foundational capability for future AI.

──── ✦ ────

ByteDance’s release of SeedRealtime follows the same logic.
Compared with solutions that chain multiple models together, a full-duplex model integrates more capabilities into one system, reducing latency and improving efficiency. It optimizes not only the model, but the entire underlying architecture.

Next, look at Unitree Technology.
Choosing this timing for an IPO is, in essence, also about reserving resources for the next phase of competition. Whether it’s financing and expanding production or continuing to develop core technologies, all of it is aimed at strengthening its core competitiveness.

──── ✦ ────

So the more I think about it, the more I feel the AI industry has entered a new stage of competition.
In the past, everyone competed based on model capability; now, more and more companies are extending both up- and downstream. Cloud providers are locking down compute, model companies are locking down chips—the competition is no longer just models, but the entire AI industry chain.

In the first phase, AI companies compete on models; in the second phase, they compete on the industry chain.
What truly determines an AI company’s value may no longer be just the model itself, but rather how many chips, how much compute, and how much infrastructure it can control.

Only represents my personal opinion and does not constitute any investment advice.#大漠茶馆 $GOOGLB $AMZNB

After listening to the earnings call for Google and Amazon recently, I got a feeling: they’re actually betting on the same future—just with completely different ways of placing the bet.

Google is investing in Gemini while also funding Anthropic. For it, if Gemini becomes a leading model, that’s obviously the best outcome; but if Anthropic continues to grow, that investment will also pay off. It turns part of the competition into an investment, and at the same time keeps a backup path for itself.

──── ✦ ────

Amazon takes a different route.

It keeps doubling down on AWS, building its own Trainium chips, and investing in AI infrastructure. On the earnings call, Andy Jassy was very direct: as long as cutting-edge models (Frontier Models) can continue to break through, AI demand in the future will keep growing. The data centers being built today aren’t just costs—they’re locking in future compute capacity in advance.

──── ✦ ────

On the surface, one is betting on models, while the other is betting on infrastructure.

But what they’re truly betting on is actually the same thing—whether Frontier Models can keep pushing forward. If Claude, Gemini, and GPT can continue improving their capabilities, enterprise demand for AI will keep growing, and the compute capacity being invested today will become the most scarce resource in the future.
If model capability starts to stall and enterprise demand slows down, then capital expenditures of several hundred billion dollars could turn from strategic assets into depreciation pressure on the balance sheet.

──── ✦ ────

So I think the real thing worth looking at in these two earnings reports isn’t who earned more, but the fact that they’re both expressing the same judgment: the development of AI is far from over.

Google bets on the model ecosystem; Amazon bets on AI infrastructure. Different ways of placing the bet, but the wager is exactly the same.

Capital expenditures have never been proof of confidence; they’re pricing the future technical boundaries. What ultimately determines how much these data centers are worth isn’t today’s earnings report, but how far the next generation of Frontier Models can still go.

The above is for personal opinions only and does not constitute investment advice.#大漠茶馆 $GOOGLB $AMZNB $NVDAB

Computing power vs Bitcoin positions · MSTR series, continued

I saw a news item, and it connects to what we talked about last time regarding Strategy possibly selling coins.

This time, Morgan Stanley has taken the lead in arranging a $15 billion loan to Anthropic for its campus in Texas, with Google offering equity as collateral. The money will be used to build data centers. The depreciation cycle for this kind of asset starts at around ten years.

At the same time, Trump Media sold 2,628 bitcoins for $165 million in cash. Combined with the partial sell-downs over the past few months, it has now sold 7,281 bitcoins in total and is down roughly $545 million overall. Its holdings have shrunk by 63%, leaving just 4,261 bitcoins.

What’s interesting is this—on the same day, Michael Saylor reaffirmed that Strategy will remain on the side of net buying Bitcoin for the long term.

//

Put these two things together, and you can see that the “bullish” narrative in both Wall Street and the crypto world has entirely different underlying capital logic. The people buying server rooms are betting on capacity. They don’t get that money back for ten years or more—they’re aiming for long-term cash flow. The people buying coins are betting on liquidity: if the market turns, they sell, and they leave none behind.

The bitcoins held by Trump Media are, in essence, more like a cash pool that can be monetized at any time—not a long-term faith position. What Saylor says about “net buying,” lines up with exactly the opposite of what’s happening with Trump Media’s holdings—these are two completely different types of capital actions.

In one sentence: even though they’re both talking about being optimistic about tech and crypto, the time horizon for the money is totally different. Don’t use the same scoreboard to compare them.

Not investment advice.#大漠茶馆

Recently someone wrote a tool called humanizer-cli. Let me tell you about it.

This tool can detect text written by AI. It has 33 different detection methods built in. After you write a paragraph and run the tool, suspicious sentences will be highlighted. Then you can edit them to sound like real human language.

This tool doesn’t judge whether what you wrote is true or false. It only looks at whether your writing sounds like AI. After you finish a first draft, checking it with this tool is pretty useful. The tool is here: https://github.com/0xwilliamortiz/humanizer-cli

I think there’s another even more worth mentioning point. In the past, it was AI learning how to write like humans. Now it’s the other way around. Humans also need to check whether what they write is human enough. Writing used to be a very natural thing. Now it has turned into a detection checkpoint you have to pass. This is definitely something worth thinking about. #大漠茶馆