Jensen's "all three passed" line restored the triopoly but didn't break the price line
On Friday afternoon at Gimpo, Jensen Huang stepped off the plane and into a press scrum that had been waiting seven months for him to say one specific thing. He said it. "All three Korean memory makers passed HBM qual. They're in production." For a market that had spent eighteen months modeling HBM as a SK Hynix-Micron duopoly with Samsung on the outside looking in, that single sentence — delivered between the tarmac and the samgyeopsal-and-soju dinner he hosted that night for the chairmen of SK, Hyundai, LG, and Naver — was supposed to be the news.
It wasn't. The news is what didn't happen next. Allocation queues didn't shorten, contract prices didn't soften, and capacity guidance didn't pull back. Samsung walked back through the HBM4 door, and the door behind it stayed wedged open by demand that had already booked through 2029.
The Door Reopens
The qualification matters less for what it adds — Samsung's incremental HBM4 volume in 2026 is still constrained — than for what it ratifies: a three-supplier topology for a market every analyst now models as structurally short. Susquehanna sparked memory-price-surge speculation on Tuesday. By Wednesday Korea Investment & Securities had lifted its second-half KOSPI top to 11,000, Goldman Sachs put 12,000 in print, and brokers were openly arguing memory had stopped being cyclical. By Thursday Korean media was reporting HBM contract prices set to rise again. DDR5 16Gb spot closed Friday at $43.4 — still elevated, still climbing.
The reason the qualification news didn't break the price line is that it never threatened to. Three doors open into the same room. The room is not bigger.
The Capacity Answer Doesn't Help Either
SK Hynix told its key suppliers this week that it intends to roughly double DRAM wafer input from today's ~550K wpm (including ~200K from Wuxi) to ~1M wpm by 2030-31, pulling Yongin fab equipment move-in forward to February 2027. The internal forecast underlying that plan: memory undersupply through 2030. Read that again — the largest single-vendor DRAM capacity expansion ever announced is sized to a shortage the planner believes will persist for five years.
On the packaging side, TSMC is racing CoWoS from ~70K wpm in 2025 to 130-140K wpm by end-2026 while tightening a "closed ecosystem" around next-gen panel-level CoPoS. Taiwan's substrate trio — Unimicron, Nan Ya PCB, Kinsus — has order visibility through 2027. Broadcom this week formalized its plan to lock HBM through 2029 to support its custom AI ASIC roadmap. Micron is publicly pursuing multi-year HBM supply deals to smooth the cycle. The supply chain is converting spot exposure into multi-year commitments faster than capacity can come online.
This is what Samsung walked into. The third door opens onto the same allocation queue.
The Macro Tape Just Confirms the Microeconomics
Korea posted an April current account surplus of the second-largest on record, driven by a 171% YoY jump in semiconductor exports. The cumulative January-April trade surplus reached $102.6 billion. The Bank of Korea print landed the same week the OECD raised Korea's 2026 GDP forecast from 1.7% to 2.6% — the largest single-economy revision among G20 members, cited explicitly to chip demand upside.
Aggregate Korean listed-firm net profit is now projected to surpass Japan's for the first time in eighteen years, with AI semiconductors named as the decisive driver. The numbers are no longer being framed as a sectoral upcycle. They are being framed as the rotation of Asia's profit pool.
Wednesday's wobble — Broadcom's peak commentary triggering a 6% KOSPI plunge and a sell-side sidecar — was the only thing that argued otherwise this week. It was bought. Samsung Electronics and SK Hynix were the cleanup names by Friday's close. The trade that gets faded now is the trade that says this ends soon.
What Reset, and What Didn't
The Seoul visit reset two things. First, Samsung is back in the HBM4 supply pool — that is real, and it shifts the customer's allocation math from binary to trinary. Second, NVIDIA-SK Hynix-TSMC have publicly described themselves as a one-stop tri-party pipeline for custom HBM. Bilateral became trilateral; trilateral now includes Samsung as a redundant lane.
What didn't reset is the queue. Broadcom 2029. Taiwan substrates 2027. Yongin pulled forward to 2027 to feed a 2030 shortage. Samsung walking through the door doesn't shorten anyone's wait — it gives the customer a third name to put on the same purchase order.
The cleanest read of the week is this: the market is no longer pricing supplier risk. It is pricing capacity risk. Three suppliers passing qualification is good news; it is not enough news. The asymmetry that matters now is between booked demand and committed wafers, and the cleanest expression of that asymmetry is that the Korean fab footprint — Yongin pulled forward to 2027, EUV delivery time cut from 34 days to 9 days by sovereign-side regulatory compression — is being treated as a national balance-sheet item by both Seoul and Washington.
Position into qualification news that doesn't cut prices. Position into capacity expansion announcements sized to multi-year shortage. The triopoly returned this week. The queue didn't notice.
Key Sources: - Jensen Huang in Seoul: All 3 Korean memory makers passed HBM qual, now in production (TheElec, 2026-06-05) - SK Hynix to Double DRAM Wafer Capacity to ~1M wpm by 2030-31; Yongin Move-in Pulled Forward to Feb 2027 (TheElec, 2026-06-04) - TSMC tightens CoPoS supply chain; CoWoS capacity to hit 130-140k wpm by end-2026 (TheElec, 2026-06-04) - Korea posts 2nd-largest current account surplus on record in April; chip exports surge 171% (Google News, 2026-06-05) - Broadcom Plans to Secure HBM Supply Through 2029 (Google News, 2026-06-04) - plus 55 more
If this analysis was helpful · ☕ Support Us · ✈️ Telegram