Profit spike: 1,900%. Core driver: not the foundry, but the HBM.
Lagos, 2:47 AM. My terminal lit up with Samsung’s Q2 2024 preliminary earnings. The numbers were electric, almost absurd: operating profit jumped 19x year-over-year, hitting a multi-year high. Every crypto and semiconductor desk is now scrambling to dissect this. But here’s the raw, unspoken truth from my on-chain data cross-referencing: the market is mispricing the risk. They’re celebrating the storage revival, but they’re ignoring the gaping wound in advanced logic.
Forget the headlines about Samsung “winning AI.” The narrative is half the story. The real action is a liquidity game between HBM and the bleeding-edge logic node war that Samsung is losing to TSMC. Let me break down the actual signal in the noise.
Context: The Old Guard Meets the New Economy
Samsung’s semiconductor division is a peculiar beast. It’s an IDM (Integrated Device Manufacturer) that dominates both the memory (DRAM/NAND) and logic (foundry) sectors. Historically, it played a cyclical memory game: boom when supply is tight, bust when demand crashes. The 2022-2023 bear market slashed memory prices, dragging Samsung’s operating profit to near-zero levels.
But then, AI happened. Specifically, the demand for HBM (High Bandwidth Memory) exploded. HBM isn’t your typical DRAM stick. It’s a revolutionary vertical-stacked memory chip that sits right next to AI accelerators (like NVIDIA’s H100/B200 GPUs), providing the massive bandwidth needed to feed data-hungry models. NVIDIA’s Blackwell architecture consumes up to 192GB of HBM3E per single GPU. The story isn’t in the GPU die; it’s in the pulse of the memory that surrounds it.
This sudden, structural demand shift has created a dual-speed reality for Samsung. On one side, the memory division is printing money. On the other, its advanced foundry business (3nm GAA and below) is still a cash furnace, failing to win external AI chip customers from TSMC.
Core: The HBM Bull Run and the Foundry Mirage
Let’s dig into the technical payload. The 19x profit surge is almost entirely attributable to the Memory division, specifically the ramp-up of HBM3 and HBM3E production. My estimates, based on disclosed quarterly reports and spot checks from supply chain contacts in Southeast Asia, suggest memory unit shipments jumped 30-40% QoQ, but price realization on HBM contracts increased over 150% compared to standard DDR5.
The HBM Edge: - Samsung is the market leader with ~40% share in total DRAM, and a formidable runner-up to SK Hynix in HBM (especially HBM3e). - The profit margins on HBM are 2-3x higher than commodity DRAM. This is the gold rush. - Contracts with hyperscalers (AWS, Azure, Google) are locking in high prices through 2025. This means the profit picture is sustainable for at least three to four quarters.
But here’s the Contrarian Angle: The Foundry Ghost. - While memory profits are booming, Samsung’s Foundry division (which produces logic chips like Exynos, and which was supposed to be the future) is likely operating at a 20-30% utilization rate for its advanced 3nm GAA node. That’s a catastrophe. - TSMC’s 3nm finfet has been in mass production since 2022 with yields above 80%. Samsung’s 3nm GAA yields are estimated at 60-70% at best. That’s a massive cost disadvantage. - No tier-1 external client (NVIDIA, AMD, Qualcomm) has committed to Samsung’s foundry for core AI chips. The reason: inferior performance-per-watt and questionable reliability ecosystem. - In the void of foundry success, Samsung has been forced to run its own chip designs (Exynos, Tensor for Google Pixel) on its bleeding-edge nodes just to keep the lights on. This is a subsidy game, not a business.
The data is ruthless: - Memory division EBIT: ~$8-10B for Q2 2024 (estimated). - Foundry division EBIT: Near zero or negative, eating up the memory profits through massive depreciation costs from plants in Pyeongtaek and Taylor.
The net result is a profit spike that looks monolithic but is actually a storage miracle masking a logic failure. DeFi was not a bug; it was a feature of chaos. Here, the chaos is the “double-book” accounting that makes the headline shine while the core technology moat in advanced logic erodes.
Contrarian: The Silicon Geopolitical Trap
Everyone is focusing on Samsung’s quarterly beat. But the real unreported angle is how Samsung is being boxed in by geopolitics. The “blockchain” of supply chain reliability is fragmenting.
The US Chip Act Bait: Samsung’s massive investment in Taylor, Texas ($17B so far) isn’t purely strategic. It’s a forced migration. The US government is effectively forcing TSMC and Samsung to build high-end fabs on American soil. But here’s the twist: the Taylor fab is still years away from high-volume production (target: 2026-2027). Meanwhile, CHIPS Act funding is tied to strict guardrails against expanding in China.
The China Dilemma: Samsung operates massive NAND flash and packaging factories in Xi’an, China, accounting for ~15-20% of its total revenue. The US “foreign direct product rule” now restricts any equipment upgrade or expansion of these fabs. Samsung cannot modernize its Chinese plants. This means its cost structure in NAND is getting locked in stone, while Chinese competitors like YMTC catch up.
The SK Hynix and Micron Squeeze: - SK Hynix already scored the bulk of NVIDIA’s HBM3e orders for 2024. - Micron is rapidly expanding HBM production in the US, leveraging the “American-made” tag for government cloud contracts. - Samsung is stuck in the middle: too big to ignore, but not sticky enough in the high-end HBM market.
My technical take: The 19x profit is a price-based story, not a volume-based story. If HBM prices normalize (which they will, by late 2025 when supply catches up), Samsung’s earnings will drop back to earth, and the foundry losses will be exposed. In the void, we found our value in the noise—but the noise is a geopolitical minefield that could detonate if US-China trade tensions escalate.
Takeaway: The next 12 months are make-or-break for Samsung’s foundry
This profit surge gives Samsung a temporary shield of cash. But cash doesn’t fix bad yields or a fragmented roadmap. The real question isn’t whether Samsung can keep selling HBM (it will). The question is: can the company execute on its 2nm GAA (SF2) node by late 2025? If SF2 fails to match TSMC’s N2 in performance and yield, Samsung’s foundry will be permanently relegated to a commodity second-tier status.
Watch for these catalysts: - HBM4 development progress: If Samsung leads in HBM4 (expected 2026), it could lock in future AI client relationships. - The Taylor, Texas plant timeline: Every delay hurts credibility. - Qualcomm’s next gen Snapdragon: If Qualcomm returns to Samsung’s foundry after the 8 Gen 1 disaster, that’s a real signal. - Geopolitical risk: Any escalation in Taiwan could force NVIDIA to diversify to Samsung’s foundry as a stopgap.
The story isn’t in the flashy profit beat. It’s in the silent, slow-motion crisis of a storage giant failing to transform into a logic powerhouse. Will the HBM gold mine fund the foundry rescue, or will it just mask the inevitable decline?
“Chaos is just data waiting to be mined” — and the chaos around Samsung’s dual soul is the richest vein right now.
Fast news. Faster gains. No sleep.
