Samsung is expected to post an 18-fold jump in quarterly profit, a figure that would have been implausible before the AI training boom rewrote memory economics. The Korean chipmaker's earnings surge tracks directly to surging demand for high-bandwidth memory, the specialized DRAM stacks that sit next to every Nvidia and AMD accelerator training frontier models. The result marks one of the sharpest year-over-year swings in the company's history.
The magnitude of the jump reflects how brutal the prior comparison base was and how hot the current cycle has become. Samsung's memory business spent much of 2023 and early 2024 in a cyclical trough, with commodity DRAM and NAND prices depressed and factory utilization cut. AI training clusters flipped the picture: HBM3E and HBM4 are now the highest-margin memory products in the industry, priced at multiples of standard DRAM per bit, and every hyperscaler wants more.
Nvidia's H100, H200, B100, and B200 accelerators each ship with tens of gigabytes of HBM soldered directly onto the package. AMD's MI300 and MI325 lines do the same. When Meta, Microsoft, Google, and Oracle order accelerators in six- and seven-figure quantities, they are also ordering HBM stacks in proportional volume — and Samsung, SK Hynix, and Micron are the only three suppliers on the planet who can make them.
Key facts
- 01Samsung is expected to post an 18-fold year-over-year jump in quarterly profit driven by AI memory demand.
- 02The surge is tied to high-bandwidth memory shipments feeding Nvidia and AMD AI accelerators.
- 03Samsung is separately in talks with Anthropic on a custom AI chip, per prior AI Chat Daily reporting.
- 04The result marks one of the sharpest earnings inflections in Samsung's history, tied directly to the AI training cycle.
Samsung has spent much of 2024 and 2025 chasing SK Hynix, its Korean rival, which locked in Nvidia's HBM3E qualification earlier and captured the majority of that generation's premium volume. Samsung's HBM4 ramp and its qualification progress with Nvidia have been the single biggest swing factors in the company's earnings recovery. The expected 18-fold jump suggests that ramp is now delivering revenue at scale.
The memory division's rebound also lifts Samsung's broader semiconductor story. Foundry — the contract chip-manufacturing business that competes with TSMC — has been under pressure, and consumer electronics carries thin margins. Memory is doing the heavy lifting, and within memory, HBM is doing the heavy lifting within that. The AI capex cycle has, in effect, financed Samsung's entire chip franchise.
The company is also positioning further up the AI stack. Samsung is in talks with Anthropic on a custom AI chip, part of a broader push by frontier labs to secure alternative silicon paths outside of Nvidia's roadmap. If any of those design engagements convert into high-volume foundry work, Samsung's earnings mix shifts again — this time toward logic manufacturing rather than pure memory.
The counterweight is cyclicality. Memory has always been a boom-bust business, and the last decade offers multiple examples of 12–18 month windows where prices doubled and then collapsed as capacity caught up. HBM is more supply-constrained than commodity DRAM because it requires advanced packaging and stacking capacity that takes years to build, but the same logic that produced the 18-fold jump on the way up can produce a sharp reversion if AI capex slows or if Micron and SK Hynix add capacity faster than demand grows. An 18-fold comparison also flatters any recovery off a low base — the absolute profit number matters more than the multiple.
The read-through for the AI market is that memory is now the tightest link in the accelerator supply chain, and the pricing power sits with the three HBM makers rather than with the GPU vendors buying from them. Samsung's earnings recovery is not a Samsung story so much as a proof point that every dollar of AI training spend flows through a narrow set of memory suppliers before it reaches a model. As long as frontier labs keep scaling parameter counts and context windows, that flow does not slow — and Samsung's next several quarters look structurally different from anything the company has posted in the past decade.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
