South Korea released customs data on 21 October showing semiconductor exports of $27.5 billion in the first twenty days of October — a 222.5% year-on-year increase. The release sent Samsung Electronics up 5.56% and SK Hynix up 7.11% in Seoul trading on 21 October. Samsung separately reported Q3 2026 preliminary earnings showing operating profit of approximately KRW 107.4 trillion (roughly $80.2 billion), a 782.5% increase from Q3 2025, on revenues of approximately KRW 195 trillion — up 127% year-on-year. The driver is high-bandwidth memory (HBM), the specialised memory architecture required by Nvidia’s GPU clusters and by data-centre AI infrastructure broadly. SK Hynix is Nvidia’s primary HBM supplier. Samsung is the second. Both companies are benefiting from AI infrastructure buildout funded by Meta ($130–145 billion 2026 capex), Microsoft, Google, and other hyperscalers. On Thursday, those hyperscaler stocks fell: Nasdaq contracts slipped 0.9%, the S&P 500 dropped 0.5%, and the Kospi — often used as a proxy for the AI trade — fell 1.8%, as investors awaited the verdict from Meta and Tesla Q3 earnings.
1. The Number: What 222.5% Means
South Korea’s customs service published export data on 21 October covering the first twenty days of October 2026. Semiconductor exports during that window totalled $27.5 billion — a 222.5% increase from the same period in 2025. [Established — Invezz, “Samsung, SK Hynix stocks soar 7%: Korea’s chip comeback is riding on a 222% surprise,” 21 October 2026, citing South Korea customs data.]
A 222% year-on-year increase in semiconductor exports is not a rounding-error phenomenon. It is an order-of-magnitude shift in demand. For context: the semiconductor industry as a whole does not grow at 222% annually under normal conditions. The figure reflects two simultaneous dynamics: the recovery from the 2022–2024 memory cycle downturn, which saw Samsung and SK Hynix operating at heavy losses on excess DRAM inventory, and the structural demand uplift from AI infrastructure spending, which requires a class of memory — high-bandwidth memory — that did not exist at commercial scale three years ago.
HBM is not conventional DRAM. It is a three-dimensional stacked memory architecture in which multiple DRAM dies are bonded directly onto a logic chip, dramatically increasing bandwidth while reducing energy consumption. Nvidia’s H100 and successor AI accelerators require HBM; so do AMD’s MI300X and essentially every AI training cluster operating at scale. The global HBM market is an effective duopoly: SK Hynix holds approximately 50% of global HBM capacity, Samsung holds most of the remainder, with Micron a distant third. [Assessed with high confidence — widely reported market structure, corroborated by BabyPips analysis, “Why Does Meta’s AI Spending Move Chip Stocks?,” October 2026.]
2. Samsung’s Record Quarter
Samsung Electronics issued preliminary Q3 2026 results showing operating profit of approximately KRW 107.4 trillion — roughly $80.2 billion at current exchange rates — representing a 782.5% year-on-year increase. Revenue reached approximately KRW 195 trillion, up 127% year-on-year. [Established — KuCoin, “Samsung Q3 2026 Earnings: Operating Profit Surges 783% as AI Chip Demand Fuels Memory Boom,” October 2026; Yahoo Finance, “Samsung Q3 2026 earnings: $80 billion record profit forecast,” October 2026.]
The 783% operating profit increase is not explained by modest demand growth. It reflects the interaction of three compounding factors: higher conventional memory prices as the 2024 inventory overhang cleared; volume and price expansion in HBM as AI infrastructure spending increased through 2026; and the absence of the restructuring charges that depressed Samsung’s 2025 results. Samsung is simultaneously recovering from a down-cycle and surfing an AI-driven up-cycle. The combined Samsung and SK Hynix combined third-quarter operating profit has been estimated at approximately KRW 190 trillion ($142 billion), which would make the Korean memory duopoly collectively one of the most profitable industrial entities on the planet in a single quarter. [Established — Bloomberg, “Stock Market Today, Dow, S&P Live Updates for October 22,” 22 October 2026, citing Korea semiconductor data.]
3. The AI Capex Flywheel
The arithmetic connecting US hyperscaler capex to Korean chip profits runs directly and is increasingly well-documented. Meta’s 2026 capital expenditure forecast stands at $130–145 billion, the upper end of a revised range disclosed with its Q2 results. Microsoft, Alphabet, and Amazon are each running comparable AI infrastructure programmes. Each dollar of AI compute buildout requires a proportional quantity of HBM, DRAM, and NAND flash — and those components flow primarily from Samsung and SK Hynix’s fabs in Korea’s Gyeonggi Province and SK Hynix’s Icheon campus. [Established — BabyPips, “Why Does Meta’s AI Spending Move Chip Stocks?,” October 2026; ABC News, “Live: ASX opens lower as Wall Street surges, Meta raises spending forecast to fund AI expansion,” 22 October 2026.]
The relationship is not merely financial; it is a supply-chain dependency. Nvidia cannot manufacture its AI accelerators without SK Hynix’s HBM3E stacks. Google’s TPU v5 cluster is similarly constrained by memory supply. When AI capex rises, HBM demand rises with it inside a roughly six-to-nine-month lag for fab ramp-up. Korea’s 222% export surge in October 2026 is the ledger entry for AI infrastructure commitments made in the first half of the year.
4. The Platform-Hardware Divergence
The Thursday market session told the opposite story. Contracts for the Nasdaq 100 Index fell 0.9% and those for the S&P 500 dropped 0.5% on 22 October, as megacap technology stocks fell ahead of earnings from Meta and Tesla. The Kospi Index — widely used as a barometer for the AI trade because of Samsung and SK Hynix’s weighting — fell 1.8%, giving back a portion of Wednesday’s surge. [Established — Bloomberg, 22 October 2026; ABC News (Australia), 22 October 2026.]
The divergence between Tuesday’s Korea hardware surge and Thursday’s platform-equity retreat is not a contradiction. It reflects a structural asymmetry in how value accumulates in the AI economy at this stage of the cycle. Samsung and SK Hynix have produced, shipped, and been paid for their Q3 chips. The revenue is realised. Meta, Microsoft, and Alphabet have spent the capital. Whether those expenditures generate returns that justify the investment — and when — is what earnings season is being asked to verify. The hardware layer books its gains when the fabs run. The platform layer books its gains when the AI applications monetise. Those are different time horizons, and the market is currently pricing uncertainty about the second one while the certainty of the first is already recorded.
5. What Meta and Tesla Earnings Determine
Meta reports Q3 2026 results this week. Investors are asking two questions that bear directly on the Korea chip thesis: Did Meta’s AI investments improve advertising yield and audience engagement sufficiently to justify $130–145 billion in annual capex? And does Meta’s forward guidance sustain or revise the capex commitment that underpins HBM demand into 2027? [Assessed with high confidence — standard megacap AI earnings framework, consistent with analyst commentary in Bloomberg and ABC News, 22 October 2026.]
Tesla’s earnings matter for a different reason: Tesla delivered 486,532 vehicles in Q3 2026, beating Wall Street expectations of approximately 461,100 units. [Established — Bloomberg, 22 October 2026.] Tesla’s AI narrative rests on its autonomous driving programme and the Dojo supercomputer. If Dojo’s results in Q3 are disappointing, the market will apply a discount to Tesla’s AI multiple that compounds downward pressure on AI-adjacent equities, including the Kospi.
Prediction: Samsung and SK Hynix will maintain combined HBM market revenue growth above 60% year-on-year through Q1 2027, regardless of Meta and Tesla Q3 2026 earnings outcomes this week. The primary driver is forward-committed AI infrastructure spending from hyperscalers, which takes six to nine months to flow through to memory demand. The 2026 capex commitments already made by Meta, Microsoft, and Alphabet will sustain Korean HBM order books into mid-2027.
Confidence: Assessed with moderate confidence. The failure mode is a coordinated hyperscaler capex reduction in response to investor pressure following disappointing Q3 AI monetisation metrics — the scenario the market is currently pricing as tail risk. If Meta’s Q3 earnings show strong revenue and stable forward capex guidance, confidence revises to high.
Resolution: Samsung and SK Hynix Q1 2027 preliminary earnings; Korea customs semiconductor export data for Q1 2027.
Bottom line: The 222.5% export surge and Samsung’s $80 billion quarter confirm that AI infrastructure spending is real, large, and flowing directly to the memory-layer suppliers in Korea. The platform layer — Meta, Tesla, and the other hyperscalers — is still being asked to demonstrate monetisation at the scale the capex requires. The hardware makers have already answered their question. The platform makers face earnings season with theirs still open.