Samsung’s $200B AI bet raises the stakes for Nvidia suppliers

Samsung Electronics and Broadcom have entered into a memorandum of understanding (MOU) in order to strengthen their collaboration in the fields of memory, foundry manufacturing and advanced packaging. They expect to spend more than $200 billion as a result of their partnership till 2030. The agreement comes at a time when many AI chipmakers are trying to secure their supplies of high-bandwidth memory (HBM), one of the key challenges in the industry.

The agreement was made public at an AI summit held in San Francisco where South Korean President Lee Jae Myung and representatives of prominent AI firms took part. The South Korean presidential office regarded the Broadcom collaboration as one of the initiatives to enhance the local AI semiconductor industry.

While Samsung acknowledged the anticipated financial projections and scale of the partnership, the firm and the President’s office of South Korea have declined to specify how many billions they would supply each year, what the prices would be, when shipments would be scheduled, or how the amount would be split across memory, foundry, and advanced packaging. Due to the lack of any order or buying agreement, the estimated value of $200 billion would represent a mere long-term strategy, rather than already determined revenues.

Why Broadcom buying memory matters beyond the two companies

Broadcom does not manufacture memory chips, but it has become one of the world’s largest designers of custom AI accelerators. Samsung said the broader partnership will support Broadcom’s next-generation AI chips through advances in HBM, foundry manufacturing and advanced packaging. Broadcom already develops Google’s Tensor Processing Units (TPUs) and Meta’s MTIA accelerators, both of which require large amounts of HBM.

This is the reason why the agreement has the potential to create waves in the AI sector. HBM continues to be one of the most constrained components in terms of availability in AI hardware. The Samsung-Broadcom deal can secure an entire manufacturing capability that other chip developers may want to access as well.

It is already clear that demand is on the rise. According to Seoul Economic Daily, SK Group Chairman Chey Tae-won has said Broadcom has been continuously asking for what he calls an astonishing amount of memory chips, emphasizing the way hyperscale AI clients are working hard to make sure they are able to buy enough supplies for the future.

Data from the industry supports this claim. TrendForce’s first quarter 2026 report on HBM technology showed SK hynix still firmly at the head of the market, while Samsung registered the fastest recovery due to increasing HBM3E shipments followed by the launch of HBM4 commercialization. An update from TrendForce, issued later, reported that Samsung had succeeded in validating HBM4 and started shipping it ahead of any other supplier, which reinforces its position against a rising demand for AI.

The deal arrives as memory supplies remain tight

Timing may be just as crucial as the agreement.

The cost of memory has experienced growth over the course of the year, given the high demand for AI servers which exceeds the supply of such servers, and experts do not expect this situation to change any time soon.

On July 3, the forecasting company TrendForce predicted that the contracts for DRAM chips in the third quarter of 2026 would rise by 13-18% compared to the previous quarter, while NAND flash chip prices are expected to hop by 10-15%, stressing the fact that the market for DRAM chips is very tight.

According to SemiAnalysis, the industry has entered a so-called “silicon shortage phase,” during which time production of advanced logic and memory is unable to keep pace with the funding of AI infrastructure. In this scenario, firms that are prepared to make long-term commitments secure more certainty with regard to future supply while manufacturers find themselves locked into long-term contracts with customers for their extremely limited capacities.

Samsung’s bid to regain ground in AI memory

Samsung formed the agreement with Broadcom amidst efforts to bolster market position in AI memory after losing a lead in HBM to SK hynix. The company stated that the partnership with Broadcom brings together its expertise in memory, foundry production, and advanced packaging to create a comprehensive platform for AI customers.

Samsung had revealed a few months ago that it had started mass production of HBM4, and then, on May 29, released what it claimed was the first 12-layer HBM4E samples available for large customers. Samsung asserted that the latest addition to its memory lineup could achieve transfer speeds of 16 gigabits per second, as well as 3.6 terabytes per second bandwidth per stack, targeting future AI accelerators.

Its foundry business has also gained momentum. SemiAnalysis reported that Samsung Foundry has secured Tesla’s AI5 and AI6 programs alongside TSMC and has entered Nvidia’s data-center supply chain.

If the Broadcom memorandum eventually turns into firm production orders, Samsung would add another marquee AI customer to its expanding semiconductor portfolio. The South Korean Presidential Office also said Korean and global technology companies announced about $950 billion in semiconductor cooperation during the AI Summit, making the Broadcom agreement one of the event’s largest strategic commitments.

If you're reading this, you’re already ahead. Stay there with our newsletter.