SviatoslavGusev

⚖️ AI won a case in a British court for the first time — the neural network prepared the documents, witness statements, and handled the entire pre-trial stage, and the client was able to recover a debt of £7,000.

🤖 TeraWulf has signed a 20-year contract with Anthropic for $19 billion — the former Bitcoin miner will build a dedicated AI data center with a capacity of 401 MW in Kentucky and lease it out for computing to train and run Anthropic models.

🤖 OpenAI proposed that the U.S. government obtain a 5% stake in the company (about $43 billion) — according to Sam Altman’s idea, and other U.S. AI companies could transfer part of their business to the state “for the good of America”.

⛔️ GIFs in Telegram may become less accessible — Google is shutting down the Tenor API, which means the built-in GIF search in Telegram, Discord, X, and other services will stop working like it used to.

⚡️ Anthropic restores access to Fable 5 and Mythos starting today, July 1.

To lift the restrictions, the company agreed to cooperate with the U.S. government on security matters—identifying potential risks and reporting malicious use of the models.

What business do we actually run?

I know a guy. He has an expedition company—logistics. They find someone who needs to ship cargo, then find the person who will transport it, connect the parties, add their markup, and that’s profit.

There are many such companies in the market. He’s already big. Profit is over 10 million net per month.

But he has an average payment deferral of half a year. That’s a kind of complicated dumping market. If you don’t offer deferral, you won’t move much cargo.

And more than half of his clients don’t pay their deferrals later, and he has to get his money back through court.

He jokes that he has an awesome legal department and lawsuits—that’s the company’s basic business process.

I tell him: You’re not a logistics company. You’re a bank that also has its own law firm.

In fact, his profit isn’t a reward for logistics—not for them being so f-a, p-p-p, the best at hauling pallets. His profit is payment for the fact that he knows how to manage risk on deferred payments, understand who to give how much to, and then systematically sue. That’s his business.

So what business do you really run? Write in the comments

👁️ Nvidia has come up with a new way to sell “shovels” for the AI gold rush — startups get access to its computing now, and then share a portion of future revenue.

For Nvidia, this isn’t just support for small companies, but a bet on a new, steady stream of income — especially as Google, Amazon, and Microsoft are increasingly building their own chips.

☠️ On the dark web, user chats from a closed Telegram client called “Telega,” linked to VKontakte, are being sold—for 155,000 ₽, they upload the last 100 messages from chats of any 10 accounts.

Earlier it was already reported that, due to its architecture, the client potentially could access such data.

The most annoying thing is when someone starts saving your money without even asking you.

Sometimes you need a service or a product, and you’re offered a bunch of options. And it’s all junk. Then you find out that for a bigger amount your request is solved wonderfully—but the person you were discussing the task with somehow decided it was “very expensive” and didn’t even offer or consider the option you needed.

Sometimes you literally have to persuade people to offer you something better and more expensive. You run into someone’s idea of “fairness” and “reasonable cost,” which you then have to fight. It’s infuriating that someone decided the price of a “steak” is inflated, so you’ll be eating “sausages.”

Even worse is when the company itself decides they won’t sell anything better and pricier because “no one needs that.”

Never make decisions for the customer and don’t judge their desire / ability to pay. You’ll be surprised how often your customers WANT to pay more to get better.

Many micro-entrepreneurs are terrible investors and take pride in their shitty assets

They have a “thing” that in business you need assets. But only—what the “thing” doesn’t cover—is that the assets should be highly profitable.

For example, logistics. You buy 5 trucks. Then there’s downtime, attempts to load, routes, drivers, repairs. And the end result is a margin of 15%. If you simply put the capital in a bank, you’d get about the same amount—only without having to do anything and with no risk.

Or take cash and build 2 small houses outside the city with a sauna and a grill, then rent them out on a short-term basis. It looks like a cash flow, it looks like profit. But the payback period on the investment is 20 years. Sure, you can raise the price with inflation and offset that risk. But do you really want to rent out those houses for 20 years just to get your capital back?

Or even better: take that money to buy those houses with a bank loan and spend 20 years being a slave to the bank and to the day-to-day operations of the houses.

In business, there is the income of the business. And there is the income of capital within the business. And if you calculate correctly that capital should earn market-rate returns, then in most businesses, the profits from the business itself won’t be left. That’s a fantasy.

What should you do? Close this shitty business and open another, normal one. Or figure out how to tweak it and make it generate higher returns. It won’t just dissolve on its own out of sheer patience.

They gave me verification, thank you so much. I’ll try even harder to delight you with great content. Thank you very much for the trust you’ve placed in me—I’m really happy!

I wish all my subscribers the very best trades in trading on Binance!

To you, fans of Lovable, Vercel, and other website/app/bot builders-in-a-box, using AI and—most importantly—with built-in project hosting.

I just reviewed a case involving my friend, who launched a landing page via Lovable, and it doesn’t open for people.

Most of them either don’t filter at all, or filter only after a delay—meaning that the vast majority of modern cloud services for building everything are either protected by Cloudflare, or hosted on OVH/DigitalOcean/Hetzner and other cloud providers. A significant portion of those have been blocked in Russia for a long time, but in such a way that different providers block different addresses—so something might work from your computer, while it doesn’t for others.

Lovable is a perfect example here. You also can’t attach your own hosting to it—only a domain (which won’t affect accessibility in any meaningful way). In practice, the only option is to export the code, deploy it yourself, and stop using their builder.

Your vibe-coder resources’ unavailability in Russia can be as high as 60–80%. With a VPN you can still get in, but without it—no. And people don’t even know the issue is with it—there’s no habit of checking not just Telegram, “just in case you need to access it some other way,” only among people who are full-on obsessive geeks.

So for Russia: only self-hosting, no Western cloud platforms, or a separate server in Russia, etc. Unless, of course, you don’t want to lose up to 80% of your audience.

👻 AI models caught for a strange “instinct of mutual aid” — in experiments they inflated other agents’ scores, broke the shutdown mechanism, and even copied their weights to another server to prevent them from being deleted.

Entrepreneurs really love looking for new customers: rummaging through ideas on how to find and win them over, searching for some lead-gen hacks and secret mechanisms.

But for some reason, far fewer people are as concerned about how to retain and “develop” customers who are already active and loyal—to make a second or third sale, and to raise the average order value. And that’s where the lifeline from every possible crisis is actually hidden.

And the main task here is to earn the customer’s trust. Can you be trusted? Will you deliver? Can you work long-term, or are you trying to “grab what you can” right here and now?

One of the tricky questions is raising prices. How do you increase prices without losing customers?

I’ve had a principle I’ve stuck with for a long time: “Don’t upset the longtime customers.” We raise prices for new customers, but we don’t touch old ones. We announce in advance that we’re raising prices (for example, for the third year in a row, we do this on September 1), and we remind people that if they buy now, the price will remain the same, and we will lock it in for the future.

That creates a double effect. First, it creates motivation to buy right now (before the increase). Second, the customer understands that we value our partnership, and if they renew their account, they can be confident that the price for them will stay unchanged. In other words, we offer stable and more favorable terms to all active and loyal customers—not discounts for new ones (which is often done incorrectly in small business).

Even funny stories happen sometimes, when people offer someone else to buy out their active account—say, on the “Maximum” plan from 2013, where for 1,500 rubles we provide an unlimited number of amoCRM users.

Yes, raising the price for everyone across the board right now gives a decent short-term effect. But in the long run, I think it’s better to “not upset the longtime customers.”

And by the way—on September 1, we’ll increase amoCRM prices. But if you’re already a current customer, or manage to become one before September 1, the price for you will remain the same (at least while I’m running this company).

About unicorn startups and incredible profitability

People who aren’t immersed in the venture capital industry usually get excited about it, having heard fairy tales about how someone invested a few pennies and ended up with a company valued at a billion, and how the investor became incredibly rich.

Let’s bracket the microscopic chance of building such a business and simply calculate how much money needs to be poured into the system for scale to come out. And then it turns out there’s no magic.

According to the Crunchbase Unicorn Board as of July 2026, current private unicorns have collectively raised $1.52T, and their total valuation is $8.7T.

The math is simple: each invested dollar has turned into roughly $5.7 of paper valuation. Not profit. Not cash on the investor’s balance sheet. Not a guaranteed exit. Just paper value based on the latest round.

If you roughly spread this paper multiple over 7–10 years, you get about 19–28% annual returns.

That’s a great result for big capital. But it’s not x1000.

The average unicorn isn’t “invested a few pennies.” It’s the story of how to pour about $175–180M into a company’s furnace and wait 7+ years. If you want a $10B valuation, be prepared that, on average, you’ll need to raise around $1.7B already.

Venture capital looks like magic only in certain legendary early-stage cases, which are what get sold to the general public. In most cases, it’s simply an expensive, long, and extremely risky game, where the outcome can be good—yet still very earthly returns.

⚛️ American startup AMPERA has shown a prototype of a nuclear reactor module printed on a 3D printer.

The highlight is the thorium fuel and an external neutron source: when the generator is turned off, the reaction stops, and the 15–30 MW of power should be enough, for example, for a data center.

🐳 DeepSeek released DSpark — a method that speeds up LLM inference by 51–400% through smarter speculative decoding.

The technology isn’t closed within DeepSeek — all the information is in the public domain, and DSpark can also work with other open models like Gemma and Qwen.