Why Is RAM So Expensive? Samsung's AI Chip Boom Just Made It the World's Most Profitable Tech Company

The number that beat Nvidia

Samsung Electronics reported preliminary operating profit of 89.4 trillion won, about $58.4 billion, for the second quarter of 2026. That is a 19-fold increase from the 4.7 trillion won the company earned in the same quarter last year, and it edges out Nvidia's most recent quarterly operating profit of $53.54 billion.

For a few months, Samsung is the most profitable technology company on earth by that measure. The result beat analyst consensus, which had clustered around 84.6 trillion won ($55.1 billion), by roughly 6%. Full segment-level results, including how much of that profit came from chips versus phones, are due July 30.

The driver is not a mystery. Samsung's semiconductor division builds the DRAM, NAND, and high-bandwidth memory that go into every AI server rack being built right now, and demand for that memory has outrun supply for three straight quarters.

Why is RAM so expensive right now

If you have priced out a new PC, a graphics card, or even a phone upgrade recently, this is the same story from the other side. Samsung raised commodity DRAM prices by roughly 90% in the first quarter of 2026 compared to late 2025 pricing, then pushed through another 50% to 60% increase in the second quarter. The company is now negotiating a further 20% hike for the third quarter.

Contract prices for 12GB LPDDR5X modules, the type used in flagship phones and laptops, have climbed to around $145, up sharply from about $120 earlier this year and roughly triple where they sat in early 2025. NAND flash pricing moved almost as fast, up about 53% quarter over quarter by Citi Research's estimate.

None of this is a temporary spike tied to one product cycle. AI data center operators are buying memory in volumes the industry has not built capacity for, and Samsung, SK hynix, and Micron, the only three companies that make HBM at scale, are the ones setting the price.

Samsung's HBM4 comeback

A year ago, Samsung was the memory maker playing catch-up. It struggled to get Nvidia's approval for its HBM3E chips over performance issues while SK hynix locked up the bulk of Nvidia's business. That has flipped. Samsung has since cleared HBM4 qualification with both Nvidia and AMD, and its DS division president told employees at a July 3 town hall that 2026 chip profit alone will exceed everything the division has earned in roughly 40 years in the semiconductor business.

Samsung is now shipping HBM4 to Nvidia for the Vera Rubin AI platform and signed a supply agreement with AMD covering the Instinct MI455X accelerator and next-generation EPYC server DRAM. Samsung's HBM4 reportedly runs at pin speeds up to 13 Gbps, ahead of the 10 Gbps floor Nvidia and AMD required, and the company is pricing it in line with SK hynix rather than discounting to win share, a sign of how tight supply really is.

Analysts polled across brokerages expect Samsung's full 2026 operating profit to land near 300 trillion won, somewhere between $196 billion and $217 billion depending on the won's exchange rate. That would eclipse the company's entire cumulative semiconductor profit from 1985 through 2025.

The other side of the ledger

Beating Nvidia on paper did not translate into a market celebration. Samsung shares fell more than 10% in Seoul the day the preliminary numbers came out, dragging the benchmark Kospi index down hard enough to trigger a brief circuit-breaker suspension. The stock had already run up more than 150% this year on AI optimism, so a beat that was only 6% ahead of consensus read as a disappointment to investors pricing in something bigger.

The profit is also unevenly distributed inside the company. Memory is carrying nearly the entire result while Samsung's mobile division absorbs higher component costs on its own phones, and its foundry business, which competes with TSMC for advanced chip orders, has been losing ground and contributing little to the headline number. Chip workers are getting a cut too: Samsung's profit-sharing plan pays DS division staff up to 10.5% of operating profit as stock, worth as much as $26.6 billion this year, following a last-minute union deal.

How long the memory shortage lasts

For US builders and operators, the practical question is when memory prices come back down, and the honest answer is not soon. Samsung and SK hynix are jointly building an $800 billion memory fab cluster in Korea, but meaningful new production volume is not expected before 2027 at the earliest, and Samsung has told customers to plan for tight supply through at least that point. SK hynix reports its own Q2 results on July 29, and analysts expect the two companies combined to post close to 150 trillion won in quarterly operating profit.

For US hardware buyers and enterprise IT teams, the near-term impact is straightforward: server memory, GPUs with onboard HBM, and consumer devices built around DRAM or NAND are structurally more expensive as long as AI data center demand keeps outpacing fab capacity. Samsung's full Q2 breakdown on July 30 will show exactly how much of this profit is memory versus everything else, and whether the mobile and foundry drag is getting worse.

The numbers to remember

• Samsung's preliminary Q2 2026 operating profit hit 89.4 trillion won ($58.4 billion), a 19-fold jump from Q2 2025 and higher than Nvidia's most recent quarter.
• Commodity DRAM prices rose about 90% in Q1 2026 and another 50% to 60% in Q2, with a further 20% hike sought for Q3.
• Full-year 2026 operating profit could reach 300 trillion won, more than Samsung's entire cumulative chip profit since 1985.
• Samsung shares still fell more than 10% on the news, since investors had priced in even bigger numbers.
• New memory fab capacity from the Samsung-SK hynix $800 billion cluster is not expected before 2027, so tight supply and high prices likely persist.

Sources