Samsung Electronics 2nd quarter earnings 1/12
Revenue rose 1% year-on-year to ₩74.6 trillion (US$53.7 billion), analysts expected ₩76.33 trillion
Operating profit -55% to ₩4.7 trillion (₩6.27 T expected)
Chip division revenue -2% to ₩27.9 trillion
Chip operating profit -94% to ₩400 billion (₩2.73 T expected) $SSNLF $HXSCL $MU $TSLA $GOOGL #semiconductors #Samsung #HBM #DRAM
2/12 Samsung blamed a write-down of chip inventories, slower smartphone sales and foreign exchange rate losses for its 2nd quarter performance.
Its chip division suffered from the US ban on Nvidia H20 chip exports to China in the 2nd quarter. Samsung supplies the high-end HBM3 memory chips used with the H20. Samsung has been trying to catch up to rivals in HBM (high bandwidth memory) after missing the AI memory chip trend.
3/12 Samsung expects earnings to improve in the 2nd half of the year on steady AI momentum across all business areas, and better performance in its chip division.
“We’re thoroughly preparing for a turnaround in the 2nd half,” Samsung CFO Sooncheol Park said on the earnings call.
4/12 The US set South Korea’s tariff rate at 15%, which Samsung said “will ease uncertainties this year, but added Section 232 semiconductor tariffs, expected to be announced soon, “may significantly impact our business…” #tariff
5/12 Samsung said Tesla’s US$16.5 billion chip order means it will raise capital spending at its Taylor, Texas fab next year as it ramps up chip production. No change to 2025 capex. $TSLA #Texas #Samsung
6/12 Samsung sees DRAM prices rising throughout the 2nd half on strong momentum for a range of the chips, from legacy DDR4 to high end DDR5 server DRAM.
NAND flash memory chip prices will rise in the 3rd quarter on stronger demand for SSDs (solid state drives) for AI servers, Samsung said, noting its inventory has fallen for 2 straight quarters. #semiconductors #DRAM #NAND
7/12 Prices for high end HBM3E may drop as supply has grown faster than demand, which will cause the difference in price margins between HBM3E and conventional DRAM “to narrow sharply,” a Samsung official said.
Samsung is 3rd in the HBM market, which is dominated by SK Hynix, followed by Micron Technology. (High Bandwidth Memory)
8/12 Samsung’s HBM chip shipments rose 30% in the 2nd quarter from Q1, and the share of HBM3E increased to “a high 80% level” of Samsung’s overall HBM sales volume, and is expected to exceed a high 90% range in the 3rd quarter.
9/12 Samsung has shipped samples of next-generation HBM4 to some clients.
There was no word yet on whether Nvidia has approved Samsung’s HBM3E. $NVDA #Samsung
10/12 Samsung’s foundry continues to work on its 2nm production process, and already has one customer, its own Exynos 2600 smartphone chipset, which Samsung hopes will be inside the flagship smartphones of key customers in the 1st half 2026. $TSM $INTC #Samsung
11/12 Samsung Foundry has also made a strategic shift in how it rolls out new manufacturing technologies, adopting a dual track approach whereby it works with key clients to refine a new process before developing a standard offering for the broader market. The “single track roadmap is gone,” said Samsung VP Mijung Noh.
12/12 Elsewhere, Samsung said it expects to launch a Tri-Fold smartphone and XR headset (with Google) this year, and continues to work closely with Google on AI applications for its flagship Galaxy smartphones. $GOOGL #Samsung
TSMC Arizona has finished its 1st run of chips for Apple, AMD and Nvidia, media report, a milestone for the chip giant. Nvidia’s Blackwell chips made in Arizona are said to be shipped back to Taiwan for advanced packaging. 1/2 $TSM $AAPL $AMD $NVDA #Arizona #semiconductors #semiconductor ctee.com.tw/news/202506167…
2/2 Meanwhile, Apple’s A20 iPhone chips will be made using TSMC’s 2nm process and new WMCM (Wafer-Level Multi-Chip Module) advanced packaging done at its AP7 plant in Chiayi, south Taiwan. WMCM capacity at the plant is expected to be 50,000 wafers per month by end-2026. The A20 will be used in iPhone 18 Pro/Max, with 12GB of memory.
Advanced packaging visual explainer: chips are etched on silicon wafers. Finished wafers look like this:
1/6 Driven by surging demand for Nvidia Blackwell-based systems, Taiwan's top six AI server makers' monthly revenue is climbing at an average 39.4% year-on-year clip in 2025, up sharply from 10% last year. $NVDA $TSM #semiconductors
2/6 Check out the lift-off in 2025 versus prior years. $NVDA $TSM #AIservers #semiconductors
3/6 The “Big 6” reported record combined monthly revenue of NT$1.163 trillion (US$38.8 billion) in May 2025, a 40.1% year-on-year increase that topped last October's NT$1.162 trillion.
They are: Foxconn, Quanta, Wistron, Inventec, Wiwynn and Gigabyte. #Foxconn #Quanta #Wistron #Inventec #Wiwynn #Gigabyte
SemiAnalysis:
-China’s top AI server, the Huawei AI CloudMatrix 384, beats the best US one, the Nvidia GB200 NVL72
-TSMC manufactured the “vast majority of Ascend 910B and 910C” chips
-Samsung provided the HBM memory chips.
-China's semiconductor manufacturing is improving fast due to ease buying equipment
-The US needs to act fast to slow Huawei and China down. Thread 1/7 $NVDA $TSM #Samsung #Huawei #semiconductors #semiconductor #AIservers
2/7 SemiAnalysis: “Huawei and China now have AI system capabilities that can beat Nvidia’s…A full CloudMatrix system can now deliver 300 PFLOPs of dense BF16 compute, almost double that of the (Nvidia) GB200 NVL72” with a drawback, “it takes 3.9x the power of a GB200 NVL72.” Power deficiencies are not a limiting factor in China.
3/7 SemiAnalysis: “The CloudMatrix 384 consists of 384 Ascend 910C chips connected through an all-to-all topology. The tradeoff is simple: having five times as many Ascends more than offsets each GPU being only one-third the performance of an Nvidia Blackwell.”
Nvidia’s Aggressive Roadmap: Too Fast, Too Furious?
In May 2024, Nvidia unveiled a bold shift to a one-year release cycle for its AI data center chips, ditching the previous two-year pace. It’s a strategy designed to match the breakneck speed of the AI market and keep rivals scrambling. But has Nvidia bitten off more than it can chew? 1/7 $NVDA $TSM $HXSCL #SKhynix #Foxconn #semiconductors #semiconductor
2/7 A design flaw—“100% Nvidia’s fault,” as CEO Jensen Huang admitted—delayed the Blackwell rollout (B100 and B200) by three months. Now, with Blackwell systems finally hitting the market, they’re barreling toward a potential collision with Blackwell Ultra (B300) in the second half of 2025. This overlap could see the two vying for attention, resources, and customer budgets. It’s fair to ask: Did the accelerated design cycle play a role in that flaw slipping through?
3/7 Some Blackwell systems shipped late last year, but volume shipments of the GB200 kicked off in Q1 2025. Foxconn, the world’s biggest AI server maker, predicts its Q1 AI server revenue will soar 100% both year-on-year and quarter-over-quarter—a massive leap tied to Blackwell’s rollout.
Nvidia: Four hyperscalers bought 3.6 million Blackwell chips (B200) so far in 2025
-Blackwell Ultra (B300) and GB300 available from 2nd half 2025
Next gen chip:
-Vera Rubin available 2nd half 2026
-Vera Rubin Ultra in 2027
Next gen:
-Feynman scheduled for 2028
DGX personal AI computers with Blackwell Ultra chips inside
Quantum-X Photonics networking chips due out later 2025
Spectrum-X Photonics chips in 2026
Thread 1/ $NVDA $AMZN $GOOGL $META $MSFT $ORCL $DELL $HPQ $CSCO $SMCI #semiconductors #semiconductor reuters.com/technology/eve…
TSMC Chairman C.C. Wei said its US fabs are already fully booked through 2027, with new 2027 production lines also reserved, media report, adding despite expanded US investments, Taiwan won’t suffer as although 11 new production lines will be built in Taiwan this year, “…it is still not enough,” he said, adding there will be another 10+ more production lines built in Taiwan in the next few years, and even that won’t be enough. 1/5 $TSM $AAPL $NVDA $AVGO $AMD $MRVL $AMAT $ASML $LRCX $TOELY $KLAC #semiconductors #semiconductor money.udn.com/money/story/56…
2/5 TSMC’s Wei made the comments in a joint press conference with Taiwan President William Lai, and made use of the TV cameras by adding that there still isn’t enough land in Taiwan for fabs, to which President Lai immediately said the government will absolutely help provide more land, water and electricity.
3/5 TSMC’s chairman also dismissed concerns that a new US R&D center would undermine Taiwan, stating it will employ 1,000 researchers to optimize production and yields, reducing reliance on Taiwan. He emphasized that Taiwan’s 10,000-strong Hsinchu team will lead R&D for 2nm, 1.6nm, and 1.4nm technologies. (Note: Should be no surprise as all 3 nodes are already under development.)