贝盈策略 Intel Computex 2026 Keynote — First Take

无锡当当网信息技术有限公司成立于2010年8月，法定代表人为李俭，注册资本1.5亿人民币，经营范围包括计算机软件及硬件研发、技术服务及技术培训等，由北京当当网信息技术有限公司、北京当当科文电子商务有限公司共同持股。

What's happened/new

Intel's keynote (CEO Lip-Bu Tan) was a disciplined "back-to-engineering, execution" pitch built around one message: 18A is now ramping across every compute category. Net-net for investors, a proof-point keynote, not a product-shock one. The debate stays where it was: 18A yield/readiness as Intel Foundry's external-customer credential, delivered ~40km from TSMC's HQ.

Key takeaways

1. 18A in volume, end-to-end. Panther Lake (Core Ultra Series 3 — first 18A client SoC; 300+ design wins) for laptops, now extended to handhelds via Arc G3 / G3 Extreme; Nova Lake desktop previewed for 2H; Clearwater Forest (Xeon 6+, 288 E-cores — first 18A server CPU) for data center. One node, every category — the read-through to foundry customers is the readiness signal itself.

2. Agentic AI = CPU demand (the DC thesis). Intel argued agentic workloads (orchestration, tool use, file I/O, agent-spawning) are CPU-hungry and the CPU:GPU ratio is converging; a live pipeline demo split GPU vs CPU stages on Xeon 6+ (288 cores/socket, 576 dual-socket), citing up to 150k agents. Read: Intel staking a CPU-side claim on the AI buildout it missed on GPU — credibility vs AMD EPYC / Arm is the open question.

3. AI PC made concrete (Perplexity). Aravind Srinivas joined to demo hybrid agentic AI — small models local on Core Ultra 3, heavy work to cloud, model-agnostic (up to 20 models), privacy-preserving. Notably a finance/PE use case (a confidential LBO model kept on-device). A tangible refresh-cycle hook for Panther Lake.

4. Edge / Physical AI push. 170+ edge designs, 4,000+ ecosystem partners (robotics, retail); Physical AI framed as a ~$25T-by-2050 TAM. Foundational DC power seen rising 80→100GW by 2030.

System-level. A new open-standard "reference blueprints" effort with partners (Xeon 6-based) for agentic performance/density.

Valuation & recommendation

Our read: execution looks on-track and the 18A narrative is finally cohesive — but nothing here disturbs the GPU-led AI compute hierarchy. Intel's two levers are CPU and foundry, both still "show-me."

Source: Intel Computex keynote, 2 Jun 2026 (attended live). Intel is not a rated name for us, keynote color, not a recommendation.

本讲明分析师：

王爱琳

SFC HK执业文凭编号：BUP648

张晓飞

SFC HK执业文凭编号：BND704

崔冰睿

SFC HK执业文凭编号：BUD052

免责声明

本公众号触及的实质仅供参考，并不组成对所述证券的保举、邀约或邀约邀请。同期亦不组成任何合约的建立，不组成营业任何证券或交往的提议，亦不组成投资提议、财务、法律、税务或其他认识。本公众号关于因此信息和实质八成据此进行投资而凯旋或盘曲引起的任何亏本不承担任何牵扯亦不作出任何保证。

本公众号中所发布的信息、不雅点以及数据有可能因发布后的阛阓走势或其他不成预念念之成分而影响其有用性。文中的认识瞻望和揣摸反应了咱们戒指制作本文之日历的评估，本公众号并不承担更新信息、不雅点以及数据的牵扯，而同期在对关连信息进行任何更新时亦不作另行见告。

本公众号过头关联机构可能会关于上述信息、不雅点的宗旨公司所刊行的证券进行交往，还可能为这些公司提供其他做事。本公众号过头关联机构可能购买、合手有或出售通常或关连居品。

版权及关连权力归海通海外及关连实质提供方整个。未经海通海外书面许可，任何东说念主不得以任何表情或材干修改、复制、分发或转载官方微信中之任何实质。

《磋议讲明》免责声明

辉煌优配 ]article_adlist-->

本公众号不是海通海外磋议讲明的发布平台。本公众号触及的磋议讲明信息由海通海外证券有限公司于香港极度行政区发布，越大配资服务实质从于发布之时从以为可靠的渠说念赢得的信息或基于海通海外已发布的磋议讲明编制而成。磋议讲明信息中的不雅点及评级的发布取决于讲明作家所明察的多样阛阓成分及公司内在成分，以及海通海外对其之盈利瞻望。狡计价钱是基于一系列的假定和前设条目，投资者需在准确了解关连磋议讲明中的一皆信息的基础下，方可对磋议讲明所抒发的不雅点酿成比拟全面的清爽。本公众号仅为对关连磋议讲明部天职容的援用八成复述。受公法时刻或其它客不雅条目所限，无法同期提供不雅点、评级的依据或盈利瞻望及狡计价钱背后的假定等关连信息。因此其实质可能无法齐全或准确抒发关连磋议讲明的不雅点或认识。海通海外并不保证本公众号触及的磋议讲明所包含的信息、不雅点以及数据的准确性及齐全性。如欲了解齐全不雅点，应参见海通海外磋议网站(www.equities.htisec.com)所载齐全讲明，或与海通海外售售东说念主员有计划。

投资触及风险，海通海外并莫得洽商投资者的具体投资狡计贝盈策略，财务状态等的特定需求，任何东说念主不得依赖磋议讲明包含的信息、不雅点以及数据的准确性及齐全性作出投资决定。