Samsung Stock: Do Cheaper AI Tokens Help or Hurt?

Samsung Electronics has been one of the best-performing large stocks of the past year, up more than 400% on the back of a historic memory-chip boom. So the question a lot of investors are asking is a fair one: if the cost of running AI keeps collapsing — if AI “tokens” get ever cheaper — does that threaten the whole story? The counterintuitive answer, at least on the evidence so far, is that cheaper AI has meant more demand for memory, not less, and analysts have specifically flagged memory makers like Samsung as beneficiaries. But it’s a genuinely two-sided debate, and memory is one of the most cyclical businesses there is. Here’s how the pieces fit — the boom, the mechanism, and both sides of the argument. None of this is investment advice.

Why is Samsung stock soaring?

Samsung is riding what many analysts call one of the biggest upcycles in the history of the memory business. Its profit has hit record highs — first-quarter 2026 operating profit came in around ₩57 trillion, up more than 750% year over year, and on 7 July it pre-announced record second-quarter operating profit of roughly ₩89 trillion, comfortably above the ~₩86 trillion consensus — powered almost entirely by its chip division as AI-driven demand sends memory prices sharply higher. Contract memory prices have jumped sharply, with mainstream DRAM up double digits quarter over quarter and NAND rising even faster, and further increases forecast. The stock has risen more than 400% over twelve months as the market re-rated Samsung from a sleepy electronics conglomerate into a core AI-memory play — though, tellingly, it fell about 6% on that record Q2 print, a first taste of the two-sided debate below.

There’s also a technology story underneath the prices. Samsung spent 2024–25 as the memory laggard, but it has since caught up fast: it cleared qualification at the dominant US AI-chip customer, began shipping its newest HBM4 chips, and shipped the first samples of a next-generation HBM4E — narrowing the gap with the HBM leader. As of 7 July, the shares trade around ₩297,000 — down about 6–7% on the day even as Samsung pre-announced that record Q2 profit, and roughly 20% below their ₩374,500 record. Full segment results follow later in July.

What does “cheaper AI tokens” actually mean — and why it matters for memory

Running AI has become dramatically cheaper. The price to generate a “token” — the basic unit of an AI model’s output — has collapsed by somewhere between roughly 280 and 1,000 times in just two to three years, as models got more efficient and competition intensified. On the surface, that sounds like bad news for anyone selling the expensive hardware AI runs on. But the opposite has happened: even as the unit price collapsed, total AI spending has climbed — enterprise spending on generative AI alone is estimated to have jumped more than 20-fold in two years.

The reason is a more-than-150-year-old economic idea called the Jevons paradox: when something becomes cheaper to use, people don’t spend less on it — they use so much more that total spending climbs. Cheaper tokens made AI economical for a flood of new uses, and modern “agentic” AI workflows burn 100 to 1,000 times more tokens per task than a simple query. Inference — running models to serve users, as opposed to training them — is now roughly two-thirds of all AI compute. And crucially, the part of the supply chain that genuinely can’t expand quickly isn’t GPUs (whose rental prices swing up and down) but the memory and packaging behind them. More AI usage flows straight into more demand for memory.

The chain, step by step:

• Tokens got radically cheaper — roughly 1,000× over a few years.
• Usage exploded — cheaper AI unlocked countless new uses, and agentic workflows devour tokens.
• Inference took over — serving models is now about two-thirds of AI compute, and it runs 24/7 for years.
• Memory is the choke point — GPU rentals fluctuate, but HBM, packaging and grid power can’t scale on a dime, so demand piles into memory.

So do cheaper AI tokens help or hurt Samsung?

On the evidence so far, they’ve helped — and one major brokerage recently made the point explicitly, arguing that cheaper AI models won’t slow AI investment and naming DRAM suppliers as beneficiaries. Since Samsung is the world’s largest DRAM and NAND maker, that lands squarely in its lane. But this is a real debate, not a settled one, so here’s both sides honestly.

The bull case is straightforward: if cheaper tokens keep driving an inference explosion, that means more AI servers, and every one of them needs memory — not just the premium high-bandwidth kind, but conventional DRAM and NAND too, which is exactly where Samsung’s scale is greatest.

The bear case is equally real. Efficiency breakthroughs — new techniques that compress the memory an AI model needs — could, in theory, reduce memory consumption per workload even as usage grows. Cheaper compute could also be read as a sign the frantic capital-spending boom is closer to its peak than its start. And underneath everything sits the memory cycle: this industry has a long history of violent boom and bust, prices are currently at extraordinary highs, and enormous new factory capacity is being built that could eventually turn today’s shortage into a glut. The claim that “AI has ended the memory cycle” is, notably, the kind of claim investors have heard right before previous downturns.

Samsung vs SK Hynix: does the token question hit them differently?

It’s worth knowing how Samsung sits against its arch-rival, because cheaper tokens don’t affect them identically. SK Hynix — which recently filed to list on the Nasdaq — is the premium leader in high-bandwidth memory, the highest-margin product and the one inside the top AI accelerators, with well over half that market. Samsung is the broad-based leader across all memory: number one in conventional DRAM and NAND, but only number two-to-three in HBM, recently edged by Micron. If cheaper tokens drive a wave of mass-market inference, a lot of that demand is for ordinary memory at scale, which plays to Samsung’s breadth; the premium HBM demand rewards SK Hynix’s lead. Both, however, ride the same underlying cycle.

Is Samsung stock a buy? Bull and bear

Short version: this is an explainer, not a recommendation. Here’s the balanced case. As of 7 July the shares sit around ₩297,000, off roughly 20% from their record after a run of more than 400% in a year — and they fell about 6% even as Samsung pre-announced record Q2 profit, a neat illustration of the bear case. The analyst consensus is still bullish (a Strong-Buy tilt, mean target near ₩465,000), with the most optimistic calls ranging up toward ₩850,000.

The honest conclusion is that cheaper AI tokens have, so far, been a tailwind for Samsung rather than a threat — the Jevons paradox turning falling prices into rising demand. What no one can promise is that the demand cycle stays this hot, because in memory it never has forever. That tension — structural AI demand versus an industry that has always eventually turned — is the whole investment case in one sentence.

This is an explainer, not investment advice — Drawpie isn’t a financial adviser. Do your own research and consider a licensed professional before making any investment. For more on this theme, see our take on the rotation into AI-hardware stocks.