Author: Li Xiaoyin
Source: Hard AI
In the overnight U.S. stock market, the semiconductor sector collectively surged, with the Philadelphia Semiconductor Index and the semiconductor industry ETF SOXX closing up more than 4.5% and about 4.3% respectively, both hitting historical highs. TSMC rose by 6.8%, reaching a new high with a market value of $840 billion. Boosted by this, A-share chip stocks continued to strengthen in the morning session, with Industrial Fulian hitting the daily limit.
On May 29th, NVIDIA founder and CEO Huang Renxun kicked off the COMPUTEX conference with a keynote speech, bringing together global tech giants including AMD, Intel, Arm, and MediaTek.
What are the main highlights of this year's Computex exhibition? What is the future trend of the AI chip industry? On June 5th, J.P. Morgan released the latest research report for detailed analysis.
<h2>1. Unprecedented Attention on Computex</h2>
J.P. Morgan stated that they have been attending Computex for about 18 years, and this may be the event with the highest global attendance. This highlights the importance of the Chinese Taiwan tech ecosystem and is a good sign for long-term development.
<h2>2. AMD AI Chip Iteration Speed Accelerates to &#34;One Update Per Year&#34;</h2>
AMD announced the annual iteration plan for the Instinct AI chip, with the roadmap extending to 2026. The MI325X (with up to 288GB of HBM3e memory) will be launched in the fourth quarter of 2024, and the MI350 series based on the CDNA 4 architecture supporting FP4/FP6 will be launched in 2025. The MI400 based on the new CDNA architecture design is planned for release in 2026.
AMD's iteration model seems to be similar to NVIDIA's roadmap, both focusing on increasing HBM density.
J.P. Morgan believes that with the accelerated adoption of N3 in the second half of the year and the widespread application of SoIC and CoWoS, HBM suppliers, TSMC, and the advanced packaging supply chain will benefit the most.
<h2>3. CoPilot+PC Makes Debut, Early Feedback Mixed, Qualcomm and AMD Leading Over Intel</h2>
Microsoft unveiled CoPilot+PC at the end of May at the annual Build global developer conference, featuring a new chip with a Neural Processing Unit (NPU) capable of over 400 trillion operations per second, or 40+ TOPS, and providing long-lasting battery life. At the Computex conference, most PC OEMs have already applied and showcased CoPilot+PC, starting at $1200
Initial feedback received by JP Morgan indicates that the AI performance of CoPilot+PC is unbalanced and not sufficient to drive a large-scale PC upgrade cycle, but application support may expand in the second half of 2024.
Among SoC suppliers, Qualcomm's Snapdragon X Elite CPU has captured the largest share in initial products, AMD's Ryzen AI 300 APU has also shown strong performance, while Intel's Lunar Lake CPU platform will only be available around the holiday season.
Given the rise in PC stocks over the past month (most PC OEM stocks have risen by 12-24% since the end of April), JP Morgan believes that there may be a pullback in the short term due to the lack of consensus on CoPilot+ application and concerns about subdued replacement cycle demand. Looking at the medium term, with the unit growth and potential ASP growth brought by the expiration of Windows 10 in 2025, the PC sector is expected to achieve healthy growth.
JP Morgan points out that TSMC may also emerge as a winner for CoPilot+PC, as all three CPU platforms supporting CoPilot+ are fully manufactured by TSMC (QCOM Snapdragon X Elite on N4, AMD Strix Point on N4/N6, Intel Lunar Lake on N3/N6).
<h2>4. Significant Breakthrough of ARM CPUs in the PC Sector</h2>
The attempt of Windows on ARM has been ongoing for many years, but this attempt seems to have the potential to achieve a meaningful market share breakthrough.
The CPU performance of Qualcomm Snapdragon X Elite has now matched Apple's M3 and surpassed its x86 counterparts, while NPU performance is also leading.
The CEO of Qualcomm stated that they plan to accelerate the iteration cycle to once a year, which is much faster than the typical two-year iteration cycle in the PC CPU market. It is reported that both Microsoft and Qualcomm have invested enough resources to ensure valuable application support.
This year, the supply chain forecast for ARM PCs remains low (expected to reach 2 million units in 2024), and the initial acceptance of Qualcomm PCs at the end of June will be a key observation point. JP Morgan expects more CPU suppliers to launch ARM CPUs in the next two years (NVIDIA may collaborate with MediaTek, and Samsung is also a possibility).
It is worth noting that the CEO of Arm also predicts that Arm's market share in the Windows market will rise to 50% within 5 years.
<h2>5. GB200 and Liquid Cooling Technology Everywhere, Signaling More GB200 Combinations</h2>
Almost every server supplier is showcasing the GB200 NVL72/36 rack solution. It appears that the GB200 configuration will become the mainstream specification in the future (currently estimated to be over 35%, but according to feedback from original equipment manufacturers, it may rise to over 50%), benefiting GB200 supply chain suppliers such as Foxconn, Pegatron, Wistron, and Inventec.
Liquid cooling solutions are also very common. Preliminary feedback received by Morgan Stanley indicates that market share may still be concentrated in key components, but system-level solutions (CDU, cooling systems) may face fierce competition in the coming years.
<h2>6. Intel and AMD to launch new generation server CPUs in the second half of the year</h2>
Intel and AMD have announced their next-generation server CPU products: Xeon 6 Sierra Forest (efficiency) / Granite Rapids (performance) from Intel and AMD Zen 5 (Turin), which offer higher performance compared to previous generations.
Intel's Sierra Forest, based on Intel 3 processor nodes, will have up to 288 cores, while AMD's Turin will have up to 192 cores/384 threads.
Morgan Stanley believes that Granite Rapids and Turin will be fully launched in the third or fourth quarter of 24, which may drive general server spending in the second half of 2024 and 2025.
Morgan Stanley expects AMD's share in the server CPU market to continue to rise from the current level of 33% m/s, which will benefit TSMC and the AMD ecosystem.
<h2>7. Deepening cooperation between MediaTek and NVIDIA</h2>
MediaTek did not announce any key new products in its keynote speech, which may be somewhat disappointing as some investors were expecting MediaTek to announce ARM products.
However, Morgan Stanley pointed out that the highlight of MediaTek's keynote speech was the emphasis on the development of Application-Specific Integrated Circuits (ASICs) for data centers and the strengthened collaboration with NVIDIA (which Morgan Stanley believes involves certain businesses in automotive, ARM computing, and enterprise/network ASICs).
The main points of this article are from the research report &#34;Asian Tech Computex takeaways-Part2&#34; released by Morgan Stanley analysts Gokul Hariharan, Albert Hung, William Yang, etc. on June 5th

INTC

AMD AI chip updates accelerate to "annual updates", CoPilot+PC early feedback mixed, GB200 configuration may become the mainstream specification in the future, Intel and AMD will launch a new generation of server CPUs in the second half of the year

AI chip stocks are all rising, what trends did ComputeX show?