In the midst of the RAM memory crisis, Samsung takes a leap with its HBM4 memory. It does not imply good news for the pocket

We are in full RAM price crisis. The industry is a cake that three large producers share and the data centers and the artificial intelligence They want to eat the whole cake. Samsung is one of the companies that manufactures memory for consumption and data centersand will soon begin mass production of its latest broadband memory chips: the HBM4.

Don’t throw the bells in the air too soon.

HBM4. This technology represents a crucial advance in stacked memories. Its density allows double the bandwidth, key to transmitting more data per second, but they are also up to 40% more energy efficient than HBM3. In short: they consume less energy and have fewer bottlenecks, which translates into an improvement in data processing.

Industry sources point out that Samsung will use the 10-nanometer D1c manufacturing process for the matrix of these HBM4 memorieswith an internal structure of 4 nm. It’s a more advanced process than the 12-nanometer D1b from its main rival, SK Hynix. In addition, it will achieve a data transfer speed of 11.7 Gbps compared to 9-10 Gbps of the current standard.

Hello Nvidia. South Korean media they point that these new Samsung HBM4 modules they would have passed Nvidia certification testing and will be in february when the company starts mass manufacturing them. Where will they end up? Some to Nvidia’s new AI acceleration system, called Vera Rubinothers at the heart of Google’s seventh-generation TPUs.

After these reports, the company’s shares they went up 5.3% in the Seoul market.

The enemy at home. In statements To South Korean media, Samsung representatives have commented that they feel quite confident with a new product that will clear up doubts about the company’s ability to supply the demanding needs of data centers. The fifth-generation HBM3E memories were a bottleneck for the company, so major players in the AI ​​industry looked next door: SK Hynix.

Also South Korean, she is the second leg of memory chip manufacturing. The third is the American Micron Technology, a considerable distance from the two South Koreans. A year ago we already told that SK Hynix had achieved enormous efficiency in the DRAM stacking process to create these HBM memories, which allowed it to be 8.8 times more efficient than Samsung or Micron and, therefore, produce more modules for an industry that never stops asking.

Meanwhile, the two South Koreans were in a race for the development of the new generation HBM4, and Samsung seems to have struck the first blow. Of course, it is estimated that Hynix will also begin mass production of these new memories on the same dates.

And the consumer… what? Well nothing. If you were expecting good news related to the price of RAM, it must be said that no improvements are expected. These HBM4 modules will go to Nvidia, but we recently commented that OpenAI had reached an agreement with Samsung and SK Hynix to supply with 900,000 wafers per month. It is the volume equivalent to 39% of the estimated global capacity… and only for one company.

Translation? Bottleneck in the market, a manufacturing speed that may not meet that demand and more bad news for the user. We have seen that Micron has abandoned its Crucial brand for consumers in favor of RAM for data centers, and that Samsung and SK Hynix are focused on HBM4 memories en masse, although they are not used in consumer devices, implies that this is where they will focus on this lucrative AI market.

In short: Samsung may be dominating the new generation of memories, but 2026 seems difficult for anyone who wants to build a PC, expand RAM of yours, buy a new mobile or even wait for good news from the Steam Machine.

Image | TSMC, Google

In Xataka | RAM has become so, so expensive that there are manufacturers selling computers in an unprecedented way: “pre-assembled”