The value of semiconductor companies is soaring as the shortage of artificial intelligence (AI) chips intensifies. Big Tech companies are launching unprecedented-scale investments to expand data centers, triggering a flood of “love calls” to the semiconductor industry.

NVIDIA announced on the 17th (local time) that it has signed a contract to supply Meta with millions of AI chips, including Blackwell and Rubin, over the next several years. It did not disclose the exact contract size. However, according to a recent estimate by market research firm IDC, NVIDIA’s AI chips cost an average of USD 16,061 (about KRW 23.26 million) each, meaning that purchasing 1 million units would require more than KRW 23 trillion.

The industry is paying attention to the fact that Meta, one of NVIDIA’s major customers, has once again signed a large-scale AI chip supply contract. At the end of last year, Meta was reportedly pushing to introduce Google’s AI chip, the Tensor Processing Unit (TPU), as a replacement for NVIDIA chips. However, considering Meta’s technology ecosystem, which has so far been optimized for NVIDIA chips, as well as the excellent performance of the latest Rubin chips, it is seen as ultimately having chosen NVIDIA again.

Meta stated that it will not only purchase GPUs such as Rubin among NVIDIA’s AI chips, but also acquire Grace central processing units (CPUs) for use in its data centers. Among major North American Big Tech companies, Meta is the first to adopt NVIDIA CPUs for data center use. In the AI industry, data center CPUs have mainly been supplied by Intel or AMD. An industry official said, “As NVIDIA’s AI ecosystem expands from GPU-centric to include CPUs, technology companies’ dependence on NVIDIA is deepening.”

Jensen Huang, NVIDIA’s chief executive officer (CEO), said, “NVIDIA will provide all of its technologies to help Meta build next-generation AI infrastructure across CPUs, GPUs, networking, and software.”

As NVIDIA secures a large-scale AI chip supply contract, expectations are rising that South Korea’s semiconductor industry, including Samsung Electronics and SK hynix, will also benefit. This is because NVIDIA’s AI chips are equipped with advanced memory semiconductors from Samsung and SK.

Recently, AI-driven demand has surged, sending memory semiconductor prices sharply higher. According to industry sources, the price of sixth-generation high bandwidth memory (HBM4), which Samsung and SK are beginning to mass-produce and ship this year, has reportedly increased by a double-digit percentage compared with the previous generation (HBM3E). In addition, market research firm TrendForce analyzed that contract prices for DRAM in the first quarter of this year (January–March) rose by up to 60% from the previous quarter. HBM is an advanced memory semiconductor created by vertically stacking multiple DRAM chips.

This AI chip shortage is expected to worsen as Big Tech companies further increase their data center investments this year. Alphabet, Google’s parent company, and Amazon plan to spend USD 180 billion and USD 200 billion, respectively, on capital expenditures (CAPEX), including data center facility investments, this year. Both figures are record highs on an annual basis and represent a 50–100% increase year-on-year.

Bloomberg reported, “There is little likelihood that the memory supply shortage will ease in the near term,” adding, “In the industry, some are even calling this situation ‘RAMmageddon.’” The term is a neologism combining “RAM,” referring to memory semiconductors, and “Armageddon,” meaning the end of the world.

Popular News