Samsung's HBM3E qualification, finally. The full timeline and what it means for NVIDIA's 2026 allocation strategy.

For eighteen months Samsung Electronics was the third high-bandwidth memory vendor on paper only. The company completed development of its 12-layer HBM3E chip in April 2024, but failed NVIDIA's qualification tests repeatedly through spring and summer 2025. When Samsung finally passed in September 2025, it became the third certified supplier behind SK Hynix and Micron, but the milestone arrived too late to matter for revenue. By then NVIDIA's premium H200 and B200 allocations were locked with the two incumbents, leaving Samsung to supply AMD's MI350 and China-market H20 products instead. An industry executive told the Korea Economic Daily the certification was "less about revenue and more about pride."

The qualification gap reshaped NVIDIA's 2026 memory allocation in ways that persist even as all three vendors now ship HBM4. TrendForce reported in January 2026 that SK Hynix holds approximately 70% of NVIDIA's HBM4 book for the Vera Rubin architecture, with Samsung and Micron splitting the remainder. Jensen Huang confirmed in June 2026 that all three vendors qualified for HBM4 production, marking the first generation to start its ramp with three suppliers. Qualification is the entry ticket, not the outcome, and Samsung's eighteen-month stumble left structural advantages with competitors that technical parity alone cannot erase.

The eighteen-month qualification struggle

Samsung's HBM3E troubles began in spring 2024. The company failed NVIDIA's heat and power requirements for both 8-layer and 12-layer HBM3E stacks in April and May 2024, according to supply chain reporting compiled by Silicon Analysts. A third validation attempt stumbled again in June 2025, with a retest scheduled for September. The Korea Economic Daily reported that Samsung finally passed NVIDIA's 12-layer HBM3E qualification in September 2025, approximately eighteen months after completing chip development.

The breakthrough came from a decision by Jun Young-hyun, Samsung's chip business head and vice chairman, to redesign the DRAM core for HBM3E earlier in 2025. The redesign addressed thermal performance issues that had caused repeated failures. Jun met with NVIDIA leadership earlier in the year before instructing the development team to rework the design, according to Korean trade press accounts.

While Samsung struggled at NVIDIA, it qualified 12-layer HBM3E at AMD for the MI350 accelerator in June 2025 and cleared Broadcom's 8-layer evaluation during the same period. Silicon Analysts notes that qualification is customer-specific: a vendor's status is not one fact but a matrix of customer, generation, and stack height combinations. Samsung's NVIDIA failure coexisted with progress elsewhere, but NVIDIA's validation remained the commercial prize.

What the delay cost Samsung

The eighteen-month gap between development completion and NVIDIA qualification cost Samsung the entire Blackwell product cycle. By September 2025, when Samsung passed certification, volumes of 12-layer HBM3E chips to be supplied to NVIDIA were expected to be relatively small, as Samsung became the third supplier following SK Hynix and Micron. The qualified HBM3E went primarily to AMD MI350 and China-market H20 products, not the premium H200 and B200 lines that generate the highest margins and tightest supply.

SK Hynix and Micron had locked NVIDIA's premium allocations with contracts running into 2026 during the period Samsung was failing tests. QuantAbundancia's supply chain analysis establishes that SK Hynix holds approximately 50% of NVIDIA's HBM3E supply and remains sold out through 2026 and substantially through 2027, with HBM3E wafer starts identified as the tightest constraint in the AI accelerator build-out. Micron, the late entrant, converted early qualification into real share by shipping HBM3E 24GB 8-layer stacks in volume production by February 2024 and certifying 12-layer HBM3E with NVIDIA by April 2025.

Samsung's failure handed pricing power to its competitors during a period of acute shortage. One vendor's qualification delay is another's margin expansion when the customer has no alternative suppliers and demand exceeds supply by multiples.

HBM4 changes the question but not the hierarchy

NVIDIA's Vera Rubin architecture, successor to Blackwell, marks the first HBM generation to start its ramp with three qualified suppliers. Jensen Huang confirmed at Seoul Gimpo Business Aviation Center on June 5, 2026 that "all three vendors have been qualified," referring to Samsung Electronics, SK Hynix, and Micron for HBM4 production. Silicon Analysts observes that this shifts the competitive question from "can you ship?" to "who ships at best yield, price, and allocation?"

The shift to economic competition has not flattened the hierarchy. TrendForce reported in January 2026 that SK Hynix holds approximately 70% of NVIDIA's HBM4 allocation for Vera Rubin, with the remaining 30% split between Samsung and Micron. SK Hynix began HBM4 mass production in the first half of 2026 from M16 Icheon and M15X Cheongju fabrication lines, following a September 12, 2025 development completion announcement. Micron entered high-volume HBM4 production in March 2026, nine months after shipping first HBM4 36GB 12-layer samples in June 2025.

The scale of HBM4 demand for Vera Rubin production makes allocation share the determining variable for revenue, not technical capability alone.

Structural advantages that qualification cannot erase

SK Hynix's eighteen-month head start on HBM3E volume production, beginning in March 2024, created structural advantages in yield learning, supply chain integration, and allocation priority that persist beyond Samsung's technical catch-up. Silicon Analysts documents that sampling-to-volume timelines run nine to fifteen months when execution is smooth, and two-plus years when it is not. Samsung's HBM3E took the longer path.

Micron's faster qualification trajectory illustrates the compounding advantage of early execution. The company shipped HBM3E 24GB 8-layer stacks in volume production for the H200 by February 2024, production-capable 12-layer stacks by September 2024, and earned NVIDIA certification for the GB300's 12-layer HBM3E by April 2025. Micron entered the HBM market later than Samsung but converted qualification speed into allocation share during the window Samsung was failing tests.

Silicon Analysts' August 31, 2026 qualification tracker shows all three vendors listed as "In qualification" for 12-layer HBM4 at NVIDIA, but production ramp status and yield data remain opaque outside quarterly earnings disclosures.

What Samsung's recovery means for NVIDIA's supply strategy

Samsung's September 2025 HBM3E qualification and subsequent HBM4 certification give NVIDIA a third qualified supplier, reducing but not eliminating concentration risk. The addition of a third vendor allows NVIDIA to negotiate on price and terms in ways a two-supplier market would not permit, but the 70-30 allocation split reported by TrendForce indicates NVIDIA has not equalized volumes across vendors.

The allocation strategy reflects both technical performance and supply chain reliability. SK Hynix's consistent execution from HBM2E through HBM3E to HBM4, without the qualification gaps that marked Samsung's trajectory, earned it the dominant share. Micron's rapid qualification and volume ramp on HBM3E, despite entering the market later, demonstrated execution capability that translated into HBM4 allocation.

Samsung's path forward depends on yield, cost, and volume delivery on HBM4 during the Vera Rubin ramp. Qualification proves technical capability; allocation share follows from operational performance at scale. The company aims to close the gap with rivals by deploying its most advanced 10-nanometer-class DRAM (1c) and a 4nm logic die produced through its foundry, compared with the 1B DRAM and 12nm logic processes adopted by competitors.

For NVIDIA, three qualified HBM4 suppliers reduce the risk that a single vendor's yield issue or capacity constraint stalls the Vera Rubin ramp. But the 70-30 allocation split suggests the company has not treated all three vendors as interchangeable, and Samsung's eighteen-month qualification struggle likely weighs on allocation decisions even after technical parity is achieved.