@Compute_King: The Revelation of Samsung Becoming the World's Most Profitable Company! Samsung Electronics' Q2 results are indeed staggering. According to preliminary earnings, Samsung expects an operating profit of 89.4 trillion won (about $58.4 billion) for the second quarter of 2026, a year-on-year surge of about 19 times; revenue is expected to reach 171 trillion won, also significantly above market expectations. Even more interesting is...

The Revelation of Samsung Becoming the World’s Most Profitable Company!

Samsung Electronics’ performance this quarter is indeed remarkable. According to preliminary results, Samsung expects an operating profit of 89.4 trillion Korean won (approximately $58.4 billion) for the second quarter of 2026, a surge of about 19 times year-on-year. Revenue for the same period is expected to reach 171 trillion Korean won, significantly exceeding market expectations.

More interestingly, Samsung’s quarterly operating profit has surpassed NVIDIA’s previous quarter’s operating profit of approximately $53.5 billion. In other words, on this quarter’s profit sheet, Samsung has temporarily overtaken NVIDIA to become one of the world’s most profitable companies.

This also means that AI infrastructure has entered its second phase.

Over the past two years, everyone focused on GPUs. Whoever had GPUs, training clusters, and inference clusters held the power to dictate computing power. So H200, B200, and GB300 became the hard currency of the AI industry.

Now, the entire pricing system for memory has been reassessed by AI demand.

People have realized that what truly determines the ceiling of AI infrastructure is an entire system: GPUs, HBM, DRAM, NAND, networking, packaging, cooling, data centers, software stacks, scheduling systems, all the way down to components, PCBs, transformers, PDUs, and so on.

Today’s surge in Samsung’s profits is actually a reminder for the AI industry: Computing power isn’t just GPUs—computing power is the supply chain.

The cost curve of AI infrastructure is being repriced.

Previously, industry practitioners thought that as long as GPU prices stabilized, AI inference costs would continue to decline.

But now, things are far more complicated: ASICs may be cheaper than GPUs, but HBM prices have increased; the unit price of computing power may drop, but memory and storage prices have risen; model architectures may become more efficient, but user usage is growing exponentially; inference optimization is advancing, but context windows, multi-agent calls, and multimodal inputs are consuming all optimization gains.

This is the real contradiction the AI industry will face next: technology reduces costs, but demand devours cost reductions.

So Samsung’s earnings report is more like a new bill presented to the market by the entire AI supply chain:

In the past, everyone paid for GPUs. Now, everyone pays for GPUs + memory + storage. In the future, everyone will inevitably pay for power, networking, scheduling, and delivery capabilities.

That’s why I have always believed that AI companies must not—and cannot—only tell a model story. The AI companies that truly survive must ultimately understand three things:

First, understand models; Second, understand engineering; Third, understand the supply chain.

So, folks, the question is: among all these model companies, which ones fit this profile?

Haha, personally, I think DS and Meituan are quite interesting.

Compute King (@Compute_King): Memory giants counting cash non-stop? Samsung, SK Hynix, SanDisk ignite an absolute super cycle for storage

Editor’s note: After Jensen launched Vera Rubin at CES, the market’s expectations for strong demand for memory, NAND flash, and SSDs in AI inference, training, and edge AI scenarios continue to soar. Despite high volatility, the market still holds high growth expectations for the storage sector amid tight supply-demand conditions.