404K Tech Evening Brief 2026-06-11 — AI Cloud Orders, Storage Price Hikes, Equipment Revisions
Oracle has pushed AI cloud orders, financing and capex debates to the foreground; the storage chain is spreading from HBM4 to DRAM, NAND and server memory; equipment, advanced packaging, PCB/CCL, power and high-speed interconnect continue to absorb spillover demand from AI infrastructure.
404K | 2026-06-11
Pre-Market Takeaways
The pre-market theme in U.S. tech is not disappearing AI demand, but rather geopolitical oil prices, financing dilution, cloud capex and memory supply constraints pressuring valuations at the same time. Materials show the S&P 500 down 1.6% and the Nasdaq down 2.0%, while Super Micro Computer, AMD, Micron, TSMC and NVIDIA all pulled back across the semiconductor chain, suggesting positions and balance-sheet quality are being repriced.
The hard data for AI infrastructure remains strong. Oracle’s 4Q26 cloud infrastructure revenue grew 93% y/y, and RPO reached $638 billion, up 363% y/y. On the storage side, Vera Rubin is scheduled to ship in 2H26, HBM4 has entered volume production, CSPs have locked available 2027 LTA capacity and have begun discussing 2028 supply. Near-term risks are oil prices, financing and high capex; the longer-term bottlenecks remain memory, power, advanced packaging, PCB/CCL, optical interconnect and equipment.
AI/Semiconductor Full Value Chain
Compute: GPU/ASIC/CPU
计算：GPU/ASIC/CPU
NVIDIA fell 3.7%, but supply-chain information indicates Vera Rubin remains scheduled for 2H26 shipments, HBM4 has entered volume production, and SK hynix, Samsung Electronics and Micron have all received HBM4 certification and begun production. The tension is that demand visibility is very strong, but high memory prices are leading some AI server customers to reduce configurations from 128GB to 64GB or 96GB.
“CSP leaders have already booked all available 2027 LTA capacity and are now moving to secure 2028 supply”
AMD fell 4.9%, but CPU signals strengthened: Chipstrat argues that agentic AI is moving the CPU:GPU ratio from about 1:4 toward nearly 1:1. AMD Venice Dense has 256 cores/512 threads, while MI455X connectivity bandwidth is said to be about 1.8TB/s. BofA’s work also upgraded Intel by two notches, on the logic that the agentic CPU TAM is expanding.
Intel was upgraded two notches by BofA to Buy, with the price target raised from $96 to $135, citing easing constraints in advanced-process wafers/packaging and expansion of the agentic CPU market. Chipstrat notes that Intel is strong in executors and standard host sockets, but currently lacks a coherent host; Coral Rapids gains an entry point through NVLink cooperation in 2027.
Amazon Graviton5 is now generally available. AWS says it offers 30%-40% better price-performance than comparable instances, 98% of the top 1,000 EC2 customers use Graviton, and Meta has committed to using tens of millions of cores for agentic AI. The signal here is not AWS’s single-quarter number, but that cloud vendors’ self-developed CPUs are starting to absorb agentic invocation-side workloads.
Arm/Qualcomm/Marvell Technology/Broadcom retreated with the semiconductor chain. Materials show declines of ARM -5.4%, QCOM -6.9%, Marvell Technology -5.4% and AVGO -5.1%. This type of pullback mainly reflects high-beta deleveraging and valuation pressure; if agentic CPU demand proves out, IP, host CPUs, network ASICs and high-speed interconnect chips will still be repriced separately.
“The CPU:GPU ratio has moved from about 1:4 toward nearly 1:1; the coherent host is the only protected socket with the highest ASP”
HBM/DRAM/NAND/eSSD
HBM/DRAM/NAND/eSSD
Micron had its Wolfe price target raised from $550 to $1,250, with Outperform maintained, on the view that AI-driven supply-demand gaps persist at least through CY27 and that storage pricing and earnings expectations have risen materially. Morgan Stanley’s storage report says storage stocks trade at about 5x 2027E P/E; if LTAs cover more than 70% of supply, DRAM stocks can rerate to 8-10x.
“DRAM is the main bottleneck to the AI build-out”
SK hynix benefits from HBM4 and long-term capacity locks. Materials show SK hynix purchased KRW 44.2 billion of TC bonders from Hanmi, interpreted as HBM4 line cadence entering 2026. Nikkei also reports the company plans to lift wafer capacity to around 3x by about 2034 and double it within the next five years.
Samsung Electronics was rated Overweight and Top Pick on ordinary shares by Morgan Stanley, with a W381,000 price target and bull/bear cases of W430,000/W190,000. The report argues Samsung Electronics and SK hynix currently trade at about 5.2x and 4.7x forward P/E, still not reflecting rerating room after LTAs stabilize earnings. Key risks are supply catching up after 2028 and HBM competition pressuring margins.
SanDisk/Western Digital chain benefits in the same direction as NAND pricing. The storage report estimates 3Q26 NAND contract prices may rise another roughly 30% q/q, mainly driven by eSSD. Multiple supply-chain checks also confirm that NAND and server memory are being crowded out by AI demand, increasing capital pressure on ordinary OEMs and module makers.
DRAM/NAND industry price anchors are highly concentrated: 3Q26 DRAM prices still have more than 20%-30% upside, and HBM 2027 price negotiations may imply at least 50%-100% y/y increases. At the beginning of 2Q26, DRAM inventory was about 2-3 weeks and NAND about 4-5 weeks. AI-related products’ share of supply allocation is expected to rise from more than 50% this year to nearly 70% next year.
“Lower DRAM prices reduce the cost of running AI inference”
Foundry/Advanced Packaging/Equipment Materials
TSMC supply-chain updates are concentrated on advanced nodes and CoPoS. The Tainan city government disclosed that it has signed a 50-year superficies agreement with TSMC covering 22 land parcels in the Southern Taiwan Science Park special district. Industry posts say CoPoS is expected to enter volume production in 2H28, aiming to improve the economics of ultra-large packages above 9.5x reticle size, with NVIDIA’s Feynman AI chip a possible early adopter candidate.
Applied Materials/Lam Research/KLA saw UBS raise global WFE forecasts for 2026/2027/2028 to about $147.0 billion/$198.0 billion/$247.5 billion, up roughly 27%/35%/25% y/y. On price targets, Applied Materials was raised from $515 to $570, Lam Research from $310 to $375, and KLA from $1,770 to $2,180.
“2028 WFE On Track For ~$250B”
ASML/ASMI/TEL/NAURA benefit from what UBS calls the “early stage of a supercycle.” The report argues Samsung Electronics P4 and SK hynix M15X are ramping, P5/Y1 are expected to ramp in 2027, and Micron ID1, Tongluo and Singapore 10B also add incremental demand. ASML capacity is sufficient to support more than $46 billion of systems revenue in 2027 and more than 100 EUV shipments per year in 2027/2028.
Hanmi/TC Bonder is a hard upstream signal for HBM4. SK hynix’s KRW 44.2 billion purchase is interpreted as preparation for HBM4 volume production, not just an ordinary equipment order. The investment implication is that the HBM profit pool is spilling beyond DRAM makers into bonding equipment, advanced packaging, CoWoS and test materials.
High-stack NAND materials chain is also seeing incremental demand: SK hynix plans year-end volume production of 375-layer 3D NAND, with Cheongju M15 switching from 176/238/321 layers. Some word-line metals are moving from tungsten to molybdenum, with Air Liquide, Entegris and Merck mentioned as molybdenum material suppliers. Samsung Electronics’ molybdenum purchases rose from about 4 tons last year to about 10 tons this year, with the roadmap pointing to 80 tons in 2030.
PCB/Substrate/Connectors/High-Speed Interconnect
Shengyi Technology/WUS Printed Circuit/Victory Giant Technology/Shennan Circuits are entering the tight AI PCB and CCL supply chain. Materials say PPE resin is a key material for high-frequency low-loss CCL, with SABIC said to account for 70% of global supply. After the Jubail shutdown, epoxy resin lead times extended from 3 weeks to 15 weeks; PCB prices rose by up to 40% m/m in April, and South Korean CCL import prices reached $20,728/ton in March, up 74.5% y/y.
“This is not a broad PCB industry benefit; companies without material self-supply and non-AI server products will struggle to pass through costs”
Kingboard/EMC/Doosan Electronics are listed as beneficiaries closer to materials and vertical integration. The logic is that AI server PCB pricing elasticity comes from material shortages and low-loss requirements, not ordinary PCB beta. If SABIC restarts production or war risk eases, near-term price hikes may reverse, so investors need to focus more on AI server exposure, material self-supply and customer certification.
Amphenol/Samtec/TE Connectivity are placed by SemiVision in the CPC and high-speed connector ecosystem. Materials say CPC requires high-precision connector and cable technologies, while experience in data-center DAC cables and high-speed connectors has transfer value. The investment implication is that rack-level interconnect is expanding from copper cables and connectors to fiber and CPO, rather than following a single replacement path.
Marvell Technology fell 5.4% with the market, but materials note that Photonic Fabric offers up to 2x energy efficiency versus copper interconnect and up to 2x compute within the same power envelope. A delayed CPO timing does not mean the optical chain has failed; lasers, substrates and fiber remain shared layers, with the difference lying in timing of mass production and customer adoption cadence.
Optical Communications/Optical Chain
Applied Optoelectronics/Lumentum/Coherent benefit from continued momentum in optical interconnect discussions. Materials say AAOI management believes hyperscalers are still buying available optical transceiver capacity, with a target of more than $450 million of monthly capacity by mid-2027, corresponding to about $6 billion of annualized revenue in 2H27. On the indium phosphide supply chain, AXT and Sumitomo are said to account for about 80% of the global substrate market.
Nokia fell 3.3% in the same trading window, but AI infrastructure buildout still supports demand for optical networks and IP networks. There was no new price-target change that day, so it is best treated as a watch item: if AI factories move from GPU competition to bandwidth competition, network equipment and optical transmission will follow data-center interconnect upgrades, not only traditional carrier capex.
Ouster reported Q1 revenue of $49 million, up 49% y/y, with product sales growing for the 13th consecutive quarter. REV8 is a native color lidar based on its in-house L4 silicon, doubling detection distance and resolution. It is not a data-center optical module, but represents the embodied-intelligence perception layer, validating lidar reliability demand in robotics and warehousing scenarios.
“Embodied AI is scaling faster than its perception layer”
Thermal/Liquid Cooling/Power/Servers
Hon Hai Precision is the core system name in Taiwan’s AI Factory supply chain. SemiVision writes that Hon
Hai’s 2026Q1 revenue was NT$2.12 trillion, up 29% y/y and down 19% q/q, with gross margin of 6.18%, operating margin of 3.57% and operating profit up 63% y/y. AI server revenue already accounts for more than 50% of total revenue, and 2026 capex is expected to grow more than 30% y/y.
Wiwynn/Auras/Delta Electronics absorb rack-scale systems, liquid cooling and 800V power. Auras showed a 2MW row-level CDU supporting 12 AI racks and a 300kW in-rack CDU. Wiwynn Helios has 18 compute nodes, 6 switch trays and 240kW per rack. Delta Electronics plans small-batch shipments of 800VDC to NVIDIA in 3Q26, with +/-400VDC planned for volume production in the same period.
“compute per watt as revenue”
Super Micro Computer fell 28%, as the market viewed about $7 billion of equity and equity-linked financing as a dilution-risk example for the AI server chain. Wolfe initiated coverage at Peerperform with a fair value range of $26-31. While this is not a price-target change, it shows that orders and backlog do not automatically translate into shareholder returns; capital structure matters as well.
Applied Digital/CoreWeave reflect the AI data-center financing chain. APLD issued $1.59 billion of high-yield bonds with a 7% yield, and demand was about 5x the issue size. Bernstein maintained Underperform on CoreWeave with a $67 price target; the focus is not performance but insider selling and Jane Street lockups, suggesting the neocloud business model still needs validation.
Power infrastructure became one of the largest pre-market consensus themes. Gartner expects global data-center electricity consumption of 565TWh in 2026, up 26% y/y, with AI servers accounting for 31%, and more than 1,200TWh in 2030. Discussions around OpenAI’s 10GW Ohio data center involve a 20-year lease, an initial 800MW phase and potential construction costs above $500 billion; SB Energy plans $33.3 billion of 9.2GW gas-fired power generation.
Internet/Platforms
Oracle is today’s strongest platform signal. Morgan Stanley maintained Equal-weight with a $207 price target, estimating 4Q26 OCI revenue of about $5.8 billion, up 93% y/y, and RPO of $638 billion, up $85 billion q/q and 363% y/y. But FY27 revenue of $90 billion and EPS of $8.05 did not provide enough room for upward revisions, and the market needs to validate GPUaaS economics.
“Oracle has more AI demand than it can currently serve”
Oracle also had Citi maintain Buy with a $330 price target, emphasizing that 4 customers signed more than $8 billion of contracts and about $50 billion of RPO comes from non-CPU/GPU business, with available capacity sold out for the next 6 months. The split is that Morgan Stanley is more concerned about gross margin, lease liabilities and the credibility of FY30 targets, while Citi puts more weight on prepaid AI contracts and the BYOC model reducing cash pressure.
“RPO was $638B, up 363% y/y”
Oracle earnings details: revenue was $19.18 billion, above the $19.10 billion estimate; EPS was $2.11, above the $1.96 estimate; cloud revenue grew 47% y/y, IaaS revenue was $18.1 billion, and FY26 capex was $55.7 billion. The real debate is whether FY27 total capex of $90-95 billion, cash spending of about $70 billion and financing of about $40 billion can translate into sustained EPS upgrades.
Google remains a strong platform for AI application testing and search/cloud monetization. The related report has a Buy rating and a $445 price target, implying about 22% upside in the base case. 2026E net revenue is $413.503 billion, up 21% y/y, with operating margin around 40%. But this report only treats it as background for platform AI monetization and cloud competition, and does not include the target price in the changes table because there was no price-target change.
Amazon appears across multiple materials as context for AI cloud and self-developed CPUs: Graviton5 shows cloud vendors continue migrating toward in-house CPUs, while another view expects the company to reach $1 trillion in revenue in 2028, but without AWS breakdowns or price-target changes. For evening-brief readers, the focus is AWS CPU price-performance and demand for tens of millions of agentic AI cores, not a single trading recommendation.
Meta appears in Graviton5 customer discussions and AI application discussions, committing to use tens of millions of cores for agentic AI. The materials do not disclose Meta financial or price-target actions for the day, but it reinforces one signal: hyperscale platforms’ AI workloads consume not only GPUs, but also large amounts of general-purpose CPU, networking and memory.
Software/SaaS
Zscaler was reiterated Outperform by Wolfe, but its price target was cut to $150. The analyst believes market confidence in Zscaler’s role in AI security has strengthened, driven by management’s positive tone and customer feedback at the user conference. The target cut goes into the table at the end, but the body takeaway is that the AI security narrative has support while valuation assumptions are still being compressed.
Rubrik had its Baird price target raised from $100 to $110, with Outperform maintained, citing accelerated ramp in adjacent businesses and more strategic partner engagement supporting a sustainable growth algorithm. Together, Rubrik and Zscaler show that the software theme is not all SaaS rising together; security, data protection and AI workflows are more likely to earn valuation premiums.
Cloudflare is used as evidence for agentic CPU demand: materials say that if 100 million knowledge workers each run one agent, with about 10 agents per CPU, the U.S. alone would need 10 million CPUs; if each person runs 10 agents, it would need 100 million CPUs. This framework explains why software agents can feed back into demand for CPUs, servers and cloud infrastructure.
“AI agents may create a huge CPU/server infrastructure problem”
Anthropic CEO’s article links AI policy with infrastructure: if scaling continues for another 1-2 years, national-level AI could appear inside data centers; models above compute thresholds should undergo four risk tests covering cyber, biological weapons, loss of control and automated R&D. The investment implication is that regulation may raise entry barriers for top-tier models and compute platforms.
OpenAI/Oracle cooperation is expanding, making OpenAI frontier models and Codex more accessible to Oracle Cloud Infrastructure customers. For Oracle, this confirms cloud demand; for the software ecosystem, it shows model distribution is moving into a multi-cloud phase. For investors, the key is whether RPO can steadily convert into revenue, not only the headline.
AI application testing discussions point to one judgment: the application layer is still searching for commercialization paths, while the infrastructure layer has already been priced first by RPO, capex, CPU/memory and power. Software investors need to move from “whether it has AI” to “whether it can convert AI into budget.”
Consumer Electronics/Smart Vehicles
Apple still lacks direct financial increment after WWDC. Discussions from Stratechery and Ben Bajarin lean more toward product strategy: Apple AI and compute constraints have drawn attention to the pace of on-device agent deployment, while RTX Spark-like local AI PCs prove from the other side that edge compute is moving higher. Materials do not disclose Apple price-target changes, so the body only discusses product and architecture direction.
Apple supply chain is connected with TSMC CoPoS and M5 Pro architecture clues. Industry posts say Apple’s MacBook M5 Pro chip reflects vertical integration across semiconductors, software and services; however, there are no order, shipment or gross-margin numbers, so it cannot be extrapolated to specific suppliers and should only be used as background for edge AI design capability.
RTX Spark/AI PC is an edge-agent example in SemiVision’s note: 20-core CPU, 6,144 CUDA cores, 200 TOPS, up to 1 PFLOPS of local AI performance, TSMC 3nm, support for 128GB unified memory, and first 14-16 inch, roughly 14mm laptops expected in fall 2026. It shows AI PC competition is moving from lightweight NPUs toward local models and unified memory.
Gigabyte Technology quietly displayed the Z990 motherboard at Computex, revealing high-speed local storage expansion for the Nova Lake-S platform: 6 NVMe slots, including 3 PCIe 5.0 M.2 bays. This is not an earnings forecast, but it represents high-end motherboards preparing for denser local high-speed storage and next-generation CPU platforms.
Ouster/embodied intelligence can sit at the boundary between consumer electronics and smart hardware: Q1 revenue was $49 million, up 49% y/y, and REV8 native color lidar doubled range and resolution. Its core tracking point is not a single automotive scenario, but whether lidar becomes a must-have in warehousing, robotics and harsh-light environments.
Tesla information for the day was mainly concentrated on FSD rollout in Denmark and debate over SpaceX valuation. The high-signal parts of the smart-vehicle chain today remain edge compute, sensing hardware, memory costs and TSMC advanced nodes, not whole-vehicle sales.
Other Developments
Energy and oil prices are the pre-market risk source. DeItaone and Kobeissi materials show U.S. crude futures up $2 in the morning, with U.S. oil prices rising above $92/barrel. Multiple reports say the Strait of Hormuz closure order covers tankers and merchant ships. Tech stock pressure is not from a single fundamental change, but from high-valuation duration assets encountering oil-price and geopolitical risks.
Leveraged ETFs/trading structure amplified semiconductor volatility. Materials say nominal turnover in U.S.-listed leveraged and inverse ETFs reached $90 billion on Tuesday, up more than 3x over the past 12 months, while the 3x short semiconductor ETF SOXS traded more than 1.3 billion shares in one day. This background explains why semiconductors and AI infrastructure can see sharp drawdowns even when fundamentals remain strong.
Hyperscaler capex continues to be revised up. ZeroHedge cited Barclays as saying hyperscale data-center operator spending will exceed $1.1 trillion by 2028, about $240 billion above Wall Street expectations, with capex estimates about 26% higher. This supports demand for upstream supply chains, but for cloud vendors themselves it is pressure on cash flow and debt financing.
Goldman Sachs U.S. strategy noted that AI infrastructure stocks have risen 34% year-to-date, while non-AI infrastructure constituents are up only 2%, broadly consistent with earnings-estimate revisions. Meanwhile, AI infrastructure has risen 40% since 2Q, with median forward P/E rising to 26x. The conclusion is that the AI chain is not lacking fundamentals, but optical networks, semiconductors, grids, power supplies and liquid cooling have already been priced ahead.
Grid and data-center tension continues to intensify. ZeroHedge cited Goldman Sachs warning that the Texas grid is tightening as data centers grow rapidly. Gartner forecasts global data-center electricity consumption of 565TWh in 2026 and more than 1,200TWh in 2030. AI investment is no longer just GPU procurement, but a combined race for land, power, financing and grid-connection capacity.
India manufacturing and AI services exports discussions point to one direction: AI will redistribute national competitive advantages across services, manufacturing and compute infrastructure. For the tech evening brief, this is only long-term background and does not change today’s company ranking.
Investment-Bank Target Price Changes in the Past 12 Hours
DirectionInstitutionDateNameTarget Price ChangeRating ActionRaisedUBS2026-06-11Applied Materials$515 → $570Prefers equipment stocksRaisedUBS2026-06-11Lam Research$310 → $375Prefers equipment stocksRaisedUBS2026-06-11KLA$1,770 → $2,180Relatively smaller valuation upsideRaisedBofA2026-06-11Intel$96 → $135Upgraded two notches to BuyRaisedWolfe2026-06-11Micron$550 → $1,250Maintains OutperformRaisedBaird2026-06-11Rubrik$100 → $110Maintains OutperformCutWolfe2026-06-11ZscalerCut to $150 (prior value undisclosed)Maintains Outperform