SEOUL — Samsung Electronics is heading toward another staggering earnings record as artificial intelligence creates one of the tightest memory-chip markets in years.
But Wall Street and Korean investors are no longer impressed by record profits alone.
They want proof that the boom will last.
Samsung is expected to report preliminary third-quarter operating profit of roughly 106.1 trillion won, or about $79 billion, according to an LSEG SmartEstimate cited by Reuters.
That would represent an almost nine-fold increase from 12.17 trillion won a year earlier and would give Samsung its fourth consecutive quarter of record operating profit.
The extraordinary numbers reflect a global shortage of memory chips caused largely by surging demand from artificial-intelligence data centers.
Companies building AI infrastructure need enormous quantities of:
high-bandwidth memory,
server DRAM,
NAND storage,
and other advanced semiconductor components.
Samsung, SK Hynix and Micron control most of that global memory supply.
For now, that scarcity has handed them extraordinary pricing power.
But the market is beginning to ask a much more uncomfortable question:
What happens when memory prices stop rising this fast?
Samsung Has Already Delivered One Historic Quarter
The scale of Samsung’s earnings rebound became clear in July.
For the second quarter, Samsung estimated operating profit of approximately 89.4 trillion won, compared with just 4.7 trillion won a year earlier.
That was roughly a 19-fold increase.
Revenue rose around 129% to 171 trillion won. Samsung’s own preliminary guidance confirmed the 89.4-trillion-won operating-profit estimate.
Those numbers would normally send a stock sharply higher.
Instead, Samsung shares fell 6.9% on the day, wiping more than $80 billion from its market capitalization.
Why?
Because investors were already expecting exceptional results.
The concern had shifted from:
“Is the AI boom real?”
to:
“How much longer can this pace possibly continue?”
That same question is now hanging over the third-quarter results.
Memory Prices Have Been Doing Much of the Heavy Lifting
Samsung has benefited from unusually powerful pricing conditions.
During the second quarter, Citi estimated average selling prices for conventional DRAM rose about 44% quarter over quarter, while NAND prices jumped roughly 53%.
The reason is partly AI.
Manufacturers have diverted more wafer capacity toward expensive high-bandwidth memory used alongside AI processors.
That leaves less capacity available for ordinary memory chips used in smartphones, PCs and traditional servers.
Scarcity then pushes prices higher across the entire market.
That dynamic turned a specialized AI memory shortage into a broader semiconductor profit boom.
But the price increases are now beginning to moderate.
DRAM Price Growth Is Slowing
TrendForce expects conventional DRAM contract prices to rise about 10% to 15% in the fourth quarter.
That is still strong.
But it is dramatically slower than the approximately 60% jump seen during the second quarter.
The reason is not that demand has suddenly disappeared.
Suppliers are increasingly concerned that pushing prices too high could hurt the very industries buying the chips.
Expensive memory raises the cost of:
smartphones,
PCs,
servers,
and other electronics.
Eventually those higher costs can weaken end-user demand.
So Samsung and its rivals face an unusual balancing act.
They want high prices.
But they do not want prices so high that customers stop buying devices.
Samsung’s Chip Margins May Have Stopped Expanding
This is the number investors may watch even more closely than headline profit.
SK Securities analyst Han Dong-hee estimates Samsung’s memory-chip operating margin will be around 76% in the third quarter.
That is extraordinarily profitable.
But it is roughly unchanged from the previous quarter.
A flat margin after such explosive growth could signal that the easiest part of the earnings surge is over.
Samsung can still make enormous profits.
But if margins have peaked, investors may stop assigning increasingly higher valuations to those profits.
That is one reason analysts have already cut Samsung’s third-quarter profit forecasts by roughly 7.7% since the end of August.
The earnings boom remains enormous.
Expectations are simply becoming harder to beat.
The Won Is Creating Another Headwind
Currency movements are also working against Samsung.
The South Korean won strengthened about 14.3% against the U.S. dollar during the third quarter, its biggest quarterly gain since early 1998, according to Reuters.
Samsung earns substantial revenue overseas.
When the won strengthens, those foreign earnings translate into fewer won when brought back onto the company’s financial statements.
That means even extremely strong semiconductor demand can produce slightly less impressive reported earnings.
It is another reminder that Samsung’s profit trajectory depends on more than AI demand alone.
Samsung Is Finally Closing the Gap in HBM
The most strategically important part of Samsung’s comeback may be high-bandwidth memory.
HBM is one of the most critical components inside modern AI systems.
The technology stacks memory chips vertically, allowing GPUs and accelerators to access enormous quantities of data at extremely high speeds.
For years, Samsung had an uncomfortable problem:
SK Hynix became the HBM leader first.
SK Hynix secured a dominant position supplying Nvidia and became one of the clearest semiconductor winners of the AI boom.
Samsung struggled with qualification delays, particularly around Nvidia-linked products.
That weakness mattered because HBM is far more profitable and strategically valuable than ordinary memory.
But Samsung has made substantial progress.
JPMorgan estimates Samsung’s HBM market share could rise to around 34% in 2026, up from approximately 20% last year.
SK Hynix’s share is expected to decline to about 46% from 60% as Samsung expands shipments of its latest HBM4 products.
If that trend continues, Samsung’s AI story becomes much stronger.
HBM Is Consuming More of the World’s Memory Factories
The supply picture could remain tight for another reason.
Samsung says high-bandwidth memory could consume nearly 30% of global DRAM wafer capacity in 2027, compared with roughly 20% today.
That matters because HBM and conventional DRAM compete for the same underlying wafer-production capacity.
Every wafer used for HBM is one that cannot simultaneously become ordinary PC or smartphone memory.
So even if AI-related HBM production expands, conventional memory could remain scarce.
That is one reason the current shortage may last longer than traditional memory cycles.
SK Hynix Says 2027 Could Be the Tightest Year Ever
Samsung’s biggest rival is even more bullish.
SK Hynix CEO Kwak Noh-jung told Reuters in July that 2027 could be the worst memory shortage in the industry’s history from a supply perspective.
He said customer demand could remain higher than available production capacity even beyond 2030.
That forecast supports the bullish case for Samsung.
If the shortage really lasts for several more years, Samsung may continue enjoying strong pricing and utilization even if quarterly price increases slow.
But investors have seen memory booms before.
And they know how they usually end.
Memory Chips Have Always Been a Brutally Cyclical Business
DRAM and NAND have historically followed boom-and-bust cycles.
Demand rises.
Prices increase.
Chipmakers build more factories.
Eventually supply catches up.
Then prices collapse.
That cycle has repeated for decades.
AI may make the current boom larger and longer.
It does not necessarily abolish the economics of supply and demand.
Reuters Breakingviews cited Bernstein forecasts suggesting Samsung’s average DRAM selling prices could more than halve by 2028 as new capacity finally enters the market.
That is why investors are cautious even while Samsung reports record profits.
Today’s shortage creates tomorrow’s incentive to build too much capacity.
Samsung and SK Hynix Are Spending Enormous Amounts
South Korea is responding to AI demand with one of the largest semiconductor expansion programmes in history.
Samsung and SK Hynix are expected to contribute hundreds of billions of dollars toward a national chip-production buildout intended to substantially expand capacity.
South Korea announced a broader $576 billion semiconductor and AI strategy in June, with Samsung and SK Hynix expected to invest around 800 trillion won in new facilities and related infrastructure.
That spending is necessary if demand keeps growing.
But it introduces a huge risk.
Factories take years to build.
If they finally enter production just as AI spending slows, the market could swing rapidly from shortage to oversupply.
That is the classic memory-industry trap.
Samsung Is Trying to Protect Itself With Long-Term Contracts
The company knows that history.
So it is trying to make the next downturn less painful.
Samsung said in July that it wants roughly two-thirds of its memory output covered by long-term customer agreements.
Those deals can give Samsung greater visibility over demand and pricing.
Some semiconductor contracts increasingly include:
minimum-volume commitments,
price bands,
take-or-pay clauses,
and even upfront customer deposits.
Micron is following a similar strategy.
Its remaining performance obligations under long-term agreements recently jumped to about $150 billion, up from around $100 billion just one quarter earlier.
The industry is essentially trying to turn memory chips into something more predictable.
That would represent a major structural change.
But Contracts Cannot Eliminate the Cycle
Long-term deals provide a buffer.
They do not eliminate risk.
Customers can still renegotiate.
Contracts eventually expire.
And only part of industry output is covered.
Reuters Breakingviews has warned that existing multi-year agreements may offer only limited protection if memory prices collapse sharply after new supply enters the market.
So Samsung’s challenge is not simply securing orders.
It must expand capacity carefully enough that it does not recreate the oversupply conditions that devastated industry profitability in previous cycles.
Micron’s Results Suggest AI Demand Is Still Extremely Strong
The bullish evidence is also difficult to ignore.
U.S. memory rival Micron recently issued a quarterly revenue forecast of $61.5 billion, well above analyst expectations of roughly $57 billion.
Its latest quarterly revenue more than quadrupled to $54.23 billion.
Micron also said most of its 2027 HBM output is already committed.
That is powerful evidence that AI infrastructure spending remains extremely strong.
Amazon, Alphabet, Microsoft, Meta and other hyperscalers continue pouring enormous amounts of capital into data centers.
That demand supports Samsung even if the pace of memory-price inflation moderates.
AMD Is Also Looking for More Samsung Memory
Another encouraging sign came this week from AMD.
CEO Lisa Su said the company plans to significantly expand chip supply in 2027 and is planning its manufacturing requirements three to five years ahead.
She is expected to meet suppliers including Samsung and SK Hynix as AMD attempts to secure enough memory for future AI systems.
That shows Samsung is not dependent on only one AI ecosystem.
Nvidia may dominate accelerator shipments today.
But AMD, custom AI-chip designers and hyperscalers are all becoming increasingly large memory customers.
A broader customer base would reduce Samsung’s dependence on any single chip company.
Samsung’s Foundry Business Is Also Improving
Memory is the current profit engine.
But Samsung wants a larger role in actually manufacturing advanced logic chips too.
That business has struggled for years against TSMC.
In July, Samsung announced a major partnership with Broadcom expected to involve more than $200 billion of cooperation through 2030 across advanced memory, contract manufacturing and packaging.
Samsung will use advanced sub-2-nanometer manufacturing for future Broadcom products.
The company has also won a major Tesla chip-manufacturing agreement.
Those wins could gradually improve utilization at Samsung’s expensive foundry facilities.
If memory remains profitable while foundry losses shrink, the earnings story becomes much more durable.
But TSMC Remains the Foundry Benchmark
Samsung still has a long way to go.
TSMC reported a 77% jump in second-quarter net profit to a record T$706.6 billion, powered by demand from customers including Nvidia and Apple.
TSMC remains the dominant advanced contract manufacturer.
Samsung therefore has to compete on several fronts simultaneously:
against SK Hynix in HBM,
against Micron in memory,
against TSMC in foundry manufacturing,
and against Chinese manufacturers in lower-end DRAM and NAND.
That competitive complexity is one reason Samsung deserves a different valuation framework from a pure-play AI company.
Chinese Memory Makers Are Becoming a Bigger Threat
China is another risk.
Chinese semiconductor manufacturers remain concentrated primarily in lower-end memory products.
But they are improving.
Summit Insights analyst Kinngai Chan says more original-equipment and original-design manufacturers are beginning to adopt Chinese DRAM and NAND products.
That may not immediately threaten Samsung’s most advanced HBM products.
But increased Chinese supply could eventually pressure conventional memory pricing.
Samsung therefore has a strong incentive to move its product mix upward toward HBM and premium server memory before competition intensifies at the lower end.
AI’s Power Shortage Could Become an Unexpected Risk
There is also a problem outside the semiconductor factories.
Electricity.
Morgan Stanley warned this week that shortages of power for U.S. data centers could delay some AI deployments.
Nvidia and Broadcom may be relatively insulated because customers prioritize their key processors.
Secondary suppliers—including memory and optical-chip companies—could be more exposed if data-center projects are delayed.
That is a subtle but important risk for Samsung.
AI demand can be enormous on paper.
But memory chips only generate revenue when data centers are actually built and powered.
A shortage of electricity could slow deployments even if semiconductor supply remains tight.
Samsung’s Stock Shows Investors Are Already Nervous
Samsung shares have more than doubled since the beginning of 2026.
But they are also approximately 25% below their June record high.
That combination tells the story perfectly.
Investors remain dramatically more optimistic than they were at the beginning of the year.
But the easy confidence has disappeared.
The stock market is no longer rewarding Samsung merely for reporting record profit.
It wants evidence that those profits will remain extraordinarily high.
That creates an unusual situation.
A company can announce a nine-fold earnings increase and still disappoint investors.
Expectations Have Become the Real Enemy
Samsung’s biggest challenge may therefore be expectations.
When profit was weak, almost any recovery looked impressive.
Now analysts expect records.
Customers expect enormous supply.
Investors expect AI demand to remain powerful.
And Samsung’s share price reflects much of that optimism already.
That means each quarter must answer increasingly difficult questions:
Are HBM shipments still accelerating?
Are conventional memory prices still rising?
Are margins holding?
Are long-term contracts protecting pricing?
Is foundry profitability improving?
And are hyperscalers still spending aggressively?
One weak answer can become enough to trigger a selloff.
The AI Boom Is Still Real — But the Market Wants a Longer Story
There is little evidence today that AI memory demand has collapsed.
Quite the opposite.
Micron’s backlog is enormous.
SK Hynix says the industry could remain undersupplied beyond 2030.
AMD is trying to secure additional memory years in advance.
Samsung expects HBM to consume an increasing share of DRAM manufacturing capacity.
And the world’s largest technology companies continue spending extraordinary sums on AI infrastructure.
But investors have learned something during 2026.
Explosive earnings alone do not prove explosive earnings are permanent.
Samsung may soon announce around 106 trillion won in quarterly operating profit, an amount that would have seemed almost unimaginable only a few years ago.
Yet its stock remains well below its June high.
That disconnect explains the entire Bloomberg story.
Samsung has already proved that the AI boom can create record profits.
Now it has to prove that those profits are not simply the peak of another memory cycle arriving just before supply finally catches up.
