원문 보기
Expectations about interest rates and what they might change in the future affect whether you consider this market to be fairly valued, overvalued, or undervalued. I just want to say that earlier today on the Halftime Report, we discussed a note from Michael Hartnett of Bank of America, in which he noted that they don't yet see the current level of interest rates as a threat to anything other than a bull market in bonds and risk assets.
That means that even at 4.7%, approaching five , this is not the level that, in our opinion, will destroy the theses about growth and profits based on market fundamentals. Does this align with what your research and analysis says about the market today? Not a threat yet, but, you know, we're getting closer to it.
We are approaching the interest rate highs we saw when Powell slammed on the brakes to stop much higher inflation three or four years ago. Now I like to look at what's called the TIPS yield, which is the inflation-adjusted yield . These 10-year bonds are going for two and a half percent. When you have a stock market with a P/E ratio of 20, which is almost what we have now after these declines, that gives a real market return of 5%.
So , there is still stock. Two and a half percent. This is slightly below historical rates, but it is still an advantage that I think is important for stocks compared to bonds. So I agree: it's not a threat yet, but if we see real yields continue to rise, it will certainly have an impact. This is the other side of the “ scissors”: you have a denominator in the form of interest rates and a numerator in the form of profits, and if rates rise, then whatever profits are, it may not be enough to compensate.
The company operates primarily through two key business segments: hardware and cloud services. Hardware, which accounted for 30% of total revenue, focuses on the development, manufacturing, and sale of massive wafer processors and local AI systems designed to accelerate workloads without relying on scarce components such as high-bandwidth memory .
Meanwhile, cloud services, which accounted for just under 70% of total GAAP revenue, operate specialized AI data center infrastructure to provide high-speed data inference as a service. Data inference means when a previously trained AI model uses learned patterns to process new information and provide new answers.
This segment also includes support for enterprise agent workflows, tools for programming and deploying large language models. As we mentioned, Cerebras is an unusual player in this industry. There are no direct competitors that produce 8.5-inch square computer chips . However, the main competitor in the hardware space is Nvidia, which is the market leader in artificial intelligence training and inference.
Other notable rivals include AI chip startup Groq and AMD. In the field of cloud technologies and hyperscalers, the main competition is Google's own tensor processor systems, designed for internal cloud workloads. Other players in this space include Amazon's AWS, which makes dedicated chips for training and inference, as well as Microsoft and Meta, which build their own in-house accelerators under the names Maia and MTIA, respectively.
Cerebras is unique in its chip production methods. Other manufacturers typically cut their chips from a silicon wafer —a very thin and flat circular slice of extremely pure crystalline silicon. This serves as the fundamental basis for computer chips and other types of electronic circuits. However, Cerebras manufactures its chip from an entire wafer, thus creating the largest processors ever created.
The Cerebras Wafer Scale Engine 3 chip is about 58 times larger than the Nvidia Blackwell B200 chip. So, imagine something the size of a dinner plate having 4 trillion transistors versus the 208 billion transistors in the B200. In addition to the processing speed advantages of such large chips, the size of Cerebras' products provides enough physical space to embed RAM directly into the chip itself.
Other companies are forced to use external storage, which is slower because the data has to travel a greater distance. Cerebras claims that these unusual processors allow their systems to run significantly faster than traditional supercomputer clusters. Cerebras reported second-quarter results that missed Wall Street expectations.
Total GAAP revenue was $180 million , which was below analysts' estimates of $194 million. Cloud and services revenue for the quarter was $126 million, showing year-over-year growth of 281% . Additionally, equipment sales actually fell 23% year-over-year to $51.1 million. Co-founder and CEO Andrew Feldman said the demand for fast data extraction is huge, and Cerebras is ready to meet it.
Feldman said the company aims to triple its revenue by 2027. Let's look at some of the more positive points for Cerebras. The company raised $6.4 billion in gross proceeds through the IPO and opened a revolving credit line of up to $850 million to accelerate the pace of data center acquisitions. As of June 30, 2026, it also has $25.4 billion in outstanding contract obligations.
Cerebras also has separate partnerships with both AMD and Amazon's AWS to implement faster forms of AI inference. The partnership with AMD, for example, aims to deliver high speeds for Cerebras, increasing throughput by five times. The AI infrastructure market is entering a very interesting phase.
And one of the biggest questions right now is not whether the demand for AI is real. He is definitely real. The bigger question is how long can this actually last? Because if current spending levels continue for a few more years, some semiconductor companies that were recently considered cyclical could enter a completely different phase of their business.
And NVIDIA could be at the heart of this process. Let's start with the most important question. Nvidia's cycle could be much longer. One of the main arguments of the bulls is that the market still underestimates the length of the AI infrastructure cycle. The reason is simple. Computing power is already extremely limited, and TSMC remains one of the most important “ bottlenecks” in the entire semiconductor supply chain.
This creates an unusual situation. Demand is already strong enough to fully utilize existing capacity. While companies are simultaneously making huge plans for additional AI infrastructure. AI companies continue to request more computing resources. And data center operators sign long-term contracts to provide this capacity.
So the argument is not simply that the demand for AI is strong today. That's true. The supply chain is already struggling to meet today's demand. While customers are preparing for much higher demand tomorrow. That is why some analysts believe that the cycle could last until 2027-2028 and possibly even longer.
And if that happens, the current valuation of several semiconductor companies will become much more interesting. Why is Nvidia stock still not moving in line with fundamentals? This is where it gets interesting. Nvidia has a huge influence on AI. Hyperscaler costs continue to rise , and the company generates huge free cash flow.
And yet, stocks are forced to spend a significant amount of time consolidating, rather than just skyrocketing. Why? Because there are several reasons. First, Nvidia is now a huge company . To significantly increase the value of a multi-trillion dollar company requires an extraordinary amount of new capital and earnings.
Second, investors are concerned about competition. Google has its TPUs, Amazon has its own specialized processors, and other hyperscalers are developing their own accelerators. And investors are naturally wondering whether Nvidia's market share in AI output will decline over time. But there is another factor that may be even more interesting, and that is funding.
Nvidia doesn't just sell its GPUs anymore. The company is becoming increasingly involved in the infrastructure ecosystem around these GPUs, and this could create a whole new competitive advantage. Nvidia's new financing model. Think about the AI infrastructure problem. The company wants to build huge data centers.
This requires billions of dollars, and the GPUs themselves alone can represent a huge investment. Then you add power supply, buildings, networks, cooling, and other infrastructure. At some point, funding becomes a bottleneck, and Nvidia appears to be taking on that funding.
Comments 0