Nebius 值得买入：第二季度营收 5.82 亿美元，同比增长 454%，ARR 从 19 亿美元增至约 30 亿美元，并预计在 400 亿美元承诺额的支持下达到 70 亿至 90 亿美元。需求超过供给；产能售价比此前 Blackwell 高点高出 15%，收购旨在提升利润率更高的软件业务，但存在执行/资本开支风险。

My company, however, is much smaller; it's Nebeus. Some of you have been following me for a while, and we know that Nebius is a name we admire a lot, and we have been covering his news for some time .

The Nebius team is the same team that built Yandex, which is similar to Google in Russia. Since we're talking about Google, it's worth noting that they had to separate their businesses or abandon them entirely.

So, they built the Nebius company headquarters in Amsterdam.

It can be considered a quintessential European success story. It may be the first large European publicly traded company to reach a market capitalization of $100 billion in the past decade.

Initially, Nebius's work was very simple; they were renting out the computing power of artificial intelligence. They sold their stake for years to come, but they did make some acquisitions.

They are working to ensure that their work is not limited to selling physical servers to companies like Microsoft and Meta. We'll talk about that later, but the goal is to ensure they have a robust and sustainable software platform that generates more profit for them than for every GPU they own, operate, and rent out .

By the way , if we look at the second quarter figures, we find that Nvidia's value is around $60 billion . As for the revenues for the second quarter, they amount to $582 million.

Yes, the stock is trading at a very high premium, but revenues have increased by 454% year-over-year , and by 46% compared to the first quarter. The AI cloud computing sector alone grew by 514% year-on-year.

According to their calculations, annual recurring revenue (ARR) – calculated from last month by multiplying it by 12 – amounts to $3 billion , compared to $1.9 billion just three months ago.

The company is profitable after adjustment, with the group's adjusted EBITDA of $236 million and a profit margin of approximately 41%. They also expect, of course, that annual recurring revenue (ARR) will grow very rapidly to between $7 billion and $9 billion .

And of course, how can all this growth be achieved? This requires a huge expenditure. Capital expenditures (CapEx) are also expected to be very high, ranging between $20 billion and $25 billion.

My initial projections for 2027 indicate that capital expenditures (CapEx) will be significantly larger than in that year. But year-end recurring annual revenue (ARR), to me, could be $30 or $40 billion, which means multiplying last month's revenue by 12, which could represent next year's revenue , or in this case, 2028 revenue.

But that's not entirely accurate; It's just a simplified way of expressing it, because last month, for example, you might make $100 million, but two months later you might make $70 or maybe $150 million .

Therefore, multiplying one month's revenue by 12 and considering that to be the whole picture does not seem logical.

At the same time, to accelerate growth and ensure the availability of funds to spend on this expansion, they have backlogged contracts worth approximately $40 billion. Microsoft and Meta are a major part of these contracts.

We don't have all the details of those deals right now, but they are certainly not high-margin deals for them, and that's a good thing because they use them to strengthen their core business, ensuring they can spend to grow their business at a faster pace than others.

The future of this sector does not lie in signing huge contracts with companies like Microsoft or Meta, and perhaps Google, or others, in the foreseeable future with very low profit margins.

No, the future of this sector is to try to create another platform like AWS or GCP that attracts more enterprise customers with much higher profit margins.

It is true that the headline may not say: “They got a $20 billion contract from Microsoft or Meta,” but a more appropriate headline would be: They are generating tens of billions of dollars in revenue with high profit margins . This is the title it should be.

Now , we know that demand far exceeds supply in all sectors at the moment. The CEO stated that Nips could sell its entire planned 2027 production capacity today on basic terms, but is holding onto a portion of it due to high prices.

The company has already auctioned off its production capacity, achieving a price 15% higher than the highest price it has ever charged for Nvidia Blackwell chips. I think they have raised prices in general since then .

So, if you can generate between $10 and $15 million per megawatt, and then that figure rises to between $20 and $25 million , and you can sell part of it for $40 million , it makes sense that you would n't want to sell your entire production capacity.

Because if you can charge a higher price for a specific period of time, it will help you accelerate the growth of your business, as this sector is capital-intensive. Therefore, you need this liquidity early on to secure the company's future.

It should be noted that Nvidia is a company that has made many acquisitions, especially recently. Many of these acquisitions are made for one main product, called the " Token Factory".

It is a managed inference platform, meaning that companies pay to run AI models in a production environment rather than simply renting raw chips.

The company acquired Tavily earlier this year, which enables AI agents to conduct direct web searches. An agent who books flights or conducts research needs up-to- date information, not just what the model learned during training.

Now, the developer gets computing and research all in one place. Therefore, they have one less resource and one more reason to stay.

The next company is Eigen AI. This company is working on speeding up the operation of the same model on the same devices. Therefore, faster models mean more answers per GPU per hour, which also means more profits that Nebius can generate.

The next company is Clarifai. This company is working on improving the entire system around the model, and how tasks are scheduled across thousands of graphics processing units .

It also improves engine performance, and Clarifai, which was the other acquisition, adjusts the volume of data traffic.

Then there's a newer company, Infrarise, which was founded this week, and which didn't exist a year ago. Intel acquired it for approximately $100 to $150 million. I believe the company's founders previously worked at Granulate, a company that was acquired by Intel and then shut down due to cost-cutting pressures.

So, what does Infrarise offer? In short, its goal is to ensure that Token Factory is the primary product for attracting customers , while ensuring they continue to use the platform.

Having an unused graphics processing unit (GPU) between processing operations means a financial loss. There is something called cold starting, just like with vehicles. To be honest, I didn't know that this term was used in the field of graphics processing units and data centers, but it does exist .

Therefore, if you can reduce the cold start time to no more than 20 minutes, let's say one minute, you will save the customer a lot of time; Therefore, this saves a lot of time for Nebius, which means, once again, that Nebius can ensure that the utilization rate of each graphics processing unit (GPU) is consistently high.

If this rate increases, the company will generate significantly higher revenues than most of its competitors. It can also be rented out at better prices due to the high utilization rate, which enhances its competitive advantage.

Of course, all of this is great, but execution is key. We have heard about many problems related to data center construction permits, from permit delays to lawsuits, etc. Therefore, all of that must be taken into account.

This is not a problem specific to Nebius, but perhaps a general problem, but I think Nebius has a great diversity in the locations and number of its data centers.

The United States is a huge market, the Middle East is expanding, and so is Europe. Therefore, I firmly believe , and I believe their head of revenue recently stated, that they have a plan to reach $100 billion in revenue in the near future.

Of course, execution is key, but given the previous execution and trajectory, the fact that the majority of the engineers there have previously built a hugely successful company, Yandex, is a tremendous competitive advantage, especially in this field where speed of action, availability of capital, and most importantly, having engineers working in perfect harmony is crucial to me, and that is why I included Nebius on this list .

Yes, it was a really interesting company, and as you know, Nvidia was a major investment partner of theirs. I believe they have invested around two billion dollars in the company so far.

Their accumulated business volume amounts to approximately $40 billion in customer commitments, spanning several years. Therefore, it is undoubtedly one of the most ambitious projects in this field, and in my opinion, it has become an increasingly important major player.