The Week the Macro Map Was Redrawn
Revisiting the June 14 Brief — Warsh Era Confirmed, Hormuz MOU Signed, MRVL Inclusion Live, China AI Acceleration Deepens
The June 14 brief identified six forward risk vectors for the week ahead. In the span of five trading sessions, two structural regime changes materialized simultaneously: the Warsh era formally opened with a hawkish communication overhaul, and a US-Iran memorandum of understanding collapsed the energy floor thesis anchoring global inflation models since February. Meanwhile, China’s AI industry chain continued its systematic advance across policy, model capability, physical infrastructure, and primary capital markets — developments that carry direct read-through implications for the AI semiconductor and supply chain positioning this publication tracks. This edition closes the execution loop, integrates the China intelligence layer, and establishes the forward playbook for the week of June 21–27.
A structured check of how the June 14 forward risk vectors matched realized market outcomes over the past seven days.
Granular analysis of active market themes that moved this week, with updated positioning across Western macro, AI semiconductors, Asian supply chain, and China’s AI industry chain.
The Dual Shock: Warsh Hawkishness Meets Hormuz Deflation
June 17 delivered a rare simultaneous macro regime shift across two orthogonal variables. Chairman Warsh opened his tenure with a fundamentally altered communication architecture. On the same afternoon, the US and Iran signed a memorandum of understanding designed to bring the conflict to a formal end within 60 days. The two events structurally contradict each other: a hawkish Fed tightening bias met a deflationary energy shock in the same news cycle. Furthermore, the July 14 CPI print will force that contradiction into the open — making it the dominant market risk event for the summer.
The collision between Warsh hawkishness and Hormuz deflation is not a contradiction to avoid — it is the central market tension to position around for the next six weeks. If the MOU holds through its 60-day window, energy inputs will pull headline CPI toward or below 4.0% at the July 14 print. As a result, the Fed’s hawkish Dot Plot becomes immediately inconsistent with incoming data, and Warsh cannot afford to appear reactive in his inaugural policy regime. That credibility test is the dominant volatility source for the summer. Maintain structural energy hedges until MOU durability is confirmed through at least two full weeks of uninterrupted Hormuz tanker traffic, then rotate into growth assets to capture the deflationary rebound in semiconductors and cloud infrastructure.
FOMC: The Warsh Communication Reset
The policy rate was held unchanged at 3.50%–3.75% — the universally expected outcome. However, the substance was entirely in the architecture overhaul. The FOMC statement ran approximately 130 words, down from over 300 words in recent meetings, stripping all forward guidance. Warsh stated explicitly that forward guidance was “not well suited for the current policy conjuncture.” The Dot Plot was retained but delivered a hawkish signal: nine of eighteen voting members now project a rate hike before year-end 2026. The Fed’s revised PCE inflation projection was lifted sharply to 3.6% from 2.7% in March, while GDP growth was trimmed to 2.2%. FOMC
The Warsh reset is structurally bearish for rate-sensitive and highly leveraged technology. Without forward guidance, every data release between now and September becomes a high-volatility binary event. By contrast, cash-rich companies funding AI expansions directly from operating cash flow — leading semiconductor platforms and hyperscale cloud providers — are structurally insulated from this regime. Speculative tech and overextended industrials, therefore, bear the most exposure. The September FOMC is now the fulcrum event for all second-half positioning.
Hormuz MOU: The Energy Floor Has a Ceiling
The US-Iran memorandum of understanding signed June 17 — mediated by Pakistan — extends the ceasefire framework for 60 days with a stated objective of formal conflict resolution. Brent declined approximately 8.5% across the week to near $80 per barrel as tanker traffic recovered, including the first Saudi-owned vessels to transit since the conflict began. Additionally, the roughly 100-million-barrel inventory overhang accumulated at sea will suppress spot prices for several weeks as it unwinds. Subsequent Swiss diplomatic talks were, however, cancelled — injecting meaningful uncertainty about MOU durability going forward. BRENT ~$80
The $90-plus structural Brent floor has been replaced by a $75–$85 range contingent on MOU durability. The accord remains fragile — nuclear negotiations were explicitly deferred from the June 17 MOU text, and Swiss follow-on talks were already cancelled. Consequently, do not fully unwind energy hedges until August 16 has passed without incident. Use this window, however, to selectively rotate capital from upstream energy into semiconductor and AI infrastructure names that benefit most from sustained energy deflation at the data-center cost level.
SPCX: Post-ATH Consolidation and the Lockup Clock
After a first-day close of $160.95 on June 12, SPCX reached an all-time high of $225.64 on June 16 before retreating to approximately $185 by June 19 — an 18% drawdown in three sessions, consistent with momentum exhaustion against a 4% public float structure.
The $175 level is the first meaningful technical support. Sustained distribution below that threshold flags early insider liquidity extraction ahead of the December 2026 lockup expiry. No audited financials arrive until November. The three fundamental monitoring anchors remain: Starlink net subscriber additions, xAI capex run rates, and Starship launch frequency. These are the structural inputs behind the valuation premium — and, furthermore, the metrics that will eventually replace narrative as the primary price driver.
Asian Hardware Proxies: Momentum Phase Complete, Earnings Phase Opens
The proxy capital flow thesis that drove triple-digit rallies across the SPCX Asian supply chain during the June 8–12 window has fully played out. As a result, the analytical question shifts from whether these names rally on SpaceX news to whether underlying revenue streams justify their re-ratings at current levels.
WNC and Universal Microwave hold verified long-term Starlink purchase orders — these re-ratings are defensible and represent legitimate hold positions on pullbacks. However, Chin-Poon faces a SpaceX-driven production migration to Thailand, making Q3 revenue updates the decisive catalyst. Filtronic’s pending RNS carries asymmetric event risk: a contract announcement is a meaningful re-rating event, while absence of news is incrementally bearish for a stock trading primarily on aerospace optionality. Monitor weekly.
AVGO & MRVL: The Semiconductor Pair Trade Update
Broadcom (AVGO)
Trading at approximately $411 as of June 20, recovering from the post-Q3-guidance drift to $382. JPMorgan issued an “aggressive buyer” call at current levels, citing structural dominance in advanced packaging design and new custom AI accelerator wins across Google, Anthropic, and OpenAI. The September 2 earnings date is the next hard catalyst. Additionally, AVGO’s operating cash flow profile insulates it from the Warsh rate-hike regime — making it the preferred structural hold in the AI semiconductor pair. AVGO
Marvell Technology (MRVL)
Entering S&P 500 inclusion June 22, with the front-running largely complete. MRVL surged over 12% by June 18 to approximately $324 after Jensen Huang framed it as a candidate trillion-dollar chipmaker. B. Riley raised its target to $345 from $240. However, the caution signal is clear: approximately 85 times trailing earnings, a consensus analyst price target below the current quote, and the outgoing CFO disclosed a roughly $60 million share sale near $290. The inclusion trade is substantially spent. Post-inclusion consolidation toward $290–$305 is therefore the re-entry window. MRVL
AVGO is the higher-conviction structural hold between the two. Its multi-hyperscaler customer base, strong cash generation, and intact FY27 AI run-rate thesis are the key differentiators. MRVL carries higher optionality but also higher risk — the stretched multiple and CFO distribution near $290 are meaningful caution signals. Consequently, new MRVL exposure is best built on a post-inclusion pullback. Use the September FOMC as the next formal re-evaluation point for both positions.
A structured review of policy, model capability, physical infrastructure, and capital market developments across China’s AI industry chain — and their direct read-through implications for US market positioning in AI semiconductors, supply chain, and adjacent sectors.
China Policy: State-Directed Compute Acceleration
Beijing’s central government and six ministries co-issued a formal action plan this week targeting national computing power infrastructure at scale. The framework mandates computing capacity exceeding 300 EFLOPS by 2025, with intelligent computing required to represent at least 35% of total output. Furthermore, the plan prioritizes national hub node architecture under the integrated computing network and explicitly mandates liquid cooling as a preferred energy-efficiency technology — a direct policy tailwind for domestic liquid cooling manufacturers currently winning major procurement bids.
Concurrently, the NDRC issued draft guidance formally listing AI chip development, intelligent computing center construction, and AI-assisted drug research as encouraged industries. At the municipal level, Shanghai distributed over 100 million RMB in computing vouchers targeting large-model startups and research institutions, while Shenzhen committed to constructing more than three thousand-card-level intelligent computing centers across priority districts with domestic chip procurement subsidies attached. The CAC additionally registered over 300 new deep synthesis algorithms this month, reflecting a normalized “file as you ship” regulatory posture that is materially accelerating AIGC application deployment.
The coordinated central-and-local policy architecture confirms that China’s compute buildout is a national infrastructure mandate — not discretionary spending — with multi-year funding certainty behind it. For US investors, the read-through is structural: every GPU cluster Beijing deploys domestically through this program is a cluster that will not be sourced from NVDA. The magnitude of this substitution is precisely why the unverified HBM3 export restriction rumor detailed below carries such strategic weight — it determines whether China must accelerate its domestic memory supply chain faster than planned, and whether Western HBM producers consequently gain structural pricing leverage in the process.
China Physical AI: Chips, Networking, and the HBM Substitution Race
The most significant hardware development this week is an unverified market report that Huawei’s new-generation Ascend AI chip has completed internal tape-out on a domestic 7nm process, benchmarking the NVIDIA H100 in performance. Separately confirmed, Moore Threads (unlisted) announced mass production and initial shipment of its MTT S4000 full-function GPU, completing adaptation with multiple server manufacturers and securing thousand-card cluster orders. On the packaging front, Tongfu Microelectronics (listed: 002156.SZ) reported at its earnings briefing that Chiplet and 2.5D/3D advanced packaging projects with leading domestic AI chip companies are on track for large-scale mass production in the second half of 2026. JCET Group (listed: 600584.SH) is co-participating in comparable advanced packaging development programs.
Foreign media citing sources familiar with the matter report that the US Department of Commerce is evaluating new export restrictions on HBM memory to China, potentially covering HBM3 and above. If confirmed, this is a direct demand accelerant for SK Hynix and Micron in the Western supply chain — and simultaneously forces acceleration of China’s domestic HBM substitution program, catalyzing attention on Yoke Technology (listed: 688232.SH), which has disclosed active HBM precursor material supply relationships with both international and domestic memory producers.
On the networking layer, China Mobile (listed: 0941.HK), together with listed partners, jointly validated a full-stack domestic AI computing power cluster based on the RoCEv2 protocol — achieving lossless low-latency transmission at the thousand-card scale and establishing a functioning domestic substitute for InfiniBand. On thermal management, Sugon (listed: 002279.SZ) won a liquid cooling procurement contract exceeding 500 million RMB from a major domestic telecom operator, while a listed cooling specialist introduced a new CDU series achieving a PUE below 1.05 in adapted domestic GPU server configurations.
China’s physical AI infrastructure program is executing across every node of the stack simultaneously — compute, packaging, memory materials, networking, and thermal management. The RoCEv2 validation at thousand-card scale is, furthermore, structurally significant: NVDA’s networking business has been a key component of its AI data-center margin expansion story, and a credible domestic Chinese alternative therefore begins to erode that moat’s long-term addressable market. The two most actionable signals from this week’s China physical AI data: first, monitor for official DoC confirmation of HBM3 export restrictions as a catalyst for MU and SK Hynix; second, treat every confirmed Chinese substitution milestone as incremental evidence of the bifurcated supply chain thesis — accelerating the timeline for TSMC’s advanced packaging capacity to be absorbed entirely by non-China hyperscaler demand.
China Virtual AI: Model Competition and Pricing Pressure
China’s large model ecosystem registered notable capability and pricing advances this week. Baidu (listed: 9888.HK / BIDU) released ERNIE Bot 4.0 Turbo, cutting API input pricing by 50% while adding long-video understanding capabilities. A Tsinghua-founded multimodal private firm upgraded its context window fivefold to 5 million characters at no additional cost and launched an integrated agent platform spanning office, academic, and legal workflows. Additionally, a leading listed technology group’s AI division announced its next-generation large model, targeting a GPT-4 Turbo performance benchmark at its upcoming release.
In generative content, Kuaishou (listed: 1024.HK) opened its Kling video generation model to full public beta — supporting 1080p output at up to two minutes with image-to-video capabilities — positioning it as a direct Sora competitor. A private AIGC firm released a 3D generation model capable of producing textured assets from a single image in approximately 30 seconds, with direct application to game and film production pipelines. On the infrastructure layer, Baidu’s PaddlePaddle 3.0 beta launched with full adaptation to domestic AI chips, while a listed vector database specialist released a major update supporting hundred-billion-level retrieval with deep GPU indexing optimization.
Baidu’s 50% API price cut is the most consequential competitive signal in this dataset for Western AI platforms. Chinese model providers are engaged in systematic commoditization of inference pricing — a dynamic that will compress margins across the global AI API market over the next 12–18 months. As a result, the investment implication for US positioning is a structural tilt away from generalist AI API plays and toward the infrastructure layer underneath them (NVDA, AVGO, MRVL) and application-layer companies with genuine proprietary data moats. The infrastructure layer benefits regardless of which model wins; however, the API layer faces escalating Chinese pricing pressure that is structural rather than temporary.
China Embodied AI: Factory Deployment and Capital Concentration
China’s humanoid robot sector recorded two milestones this week that move it from prototype to deployment validation. A leading listed robotics-adjacent group’s humanoid platform completed practical training in a battery electric vehicle assembly plant — achieving a success rate exceeding 90% across complex tasks including chassis screw tightening and interior panel installation, with self-developed core joint motors and dexterous hands. Furthermore, two well-funded private embodied AI firms closed significant financing rounds this week, with total capital raised across the pair approaching 1.7 billion RMB, backed by a combination of strategic investors, major venture funds, and state-linked capital.
The 90%-plus factory success rate in a live automotive assembly environment is the most credible humanoid deployment benchmark yet recorded in China. Combined with the scale of private capital concentration into embodied AI this week, the signal is unambiguous: Chinese AI investors have high conviction that physical AI deployment in manufacturing is an imminent commercial event — not a 2028 projection. For US portfolio positioning, the indirect beneficiary of the global humanoid buildout — regardless of geography — is the semiconductor layer providing inference compute inside these systems. Additionally, the inference chip efficiency investment pattern (sparse computing, algorithm-defined architectures) mirrors exactly the architectural direction that makes AI deployment cost-competitive at factory scale. Watch this space as a leading indicator for the next wave of AI semiconductor demand.
The decisive data transitions and event catalysts required to navigate the week ahead across Western macro, AI semiconductors, and China intelligence inputs.
MRVL S&P 500 Inclusion: Passive benchmark buying goes live. The trade is largely front-run. Watch for a “sell the news” consolidation. Post-inclusion dip toward the $290–$305 band is the structural re-entry window, not the Sunday open. MRVL
Iran MOU Durability Watch: Formal deadline approximately August 16. Track weekly tanker traffic volumes through the Strait of Hormuz. Any breakdown reignites the full energy risk premium. Do not fully unwind energy hedges until mid-August confirmation. XOM XLE
US DoC HBM3 Export Restriction: Monitor for official confirmation. If confirmed, this is a direct demand catalyst for Western HBM producers and simultaneously accelerates China’s domestic substitution timeline. The single most important unverified item in the current intelligence stack. MU
SPCX Distribution Signals: Monitor order books for block trades above 5 million shares. Sustained selling below $175 flags early insider distribution ahead of the December 2026 lockup expiry window. SPCX
Fed Speaker Calibration: In a forward-guidance-free regime, every FOMC official comment is high-signal. Watch for any language hinting at timing or pace of potential hikes. The September FOMC remains the fulcrum event for all H2 positioning.
June CPI — The Regime Test: The most consequential macro release in H2 2026. If Hormuz normalization sustains, a print below 4.0% directly contradicts the Warsh hawkish Dot Plot. That contradiction is where summer volatility concentrates. Track weekly shipping data as the leading indicator.
Asian Supply Chain Revenue Disclosure: WNC, UMT, and Compeq Q3 earnings are the first opportunity to verify whether SpaceX aerospace revenue is showing up as explicit line-item expansion or pure narrative premium — the gateway to the next supply chain trade cycle.
Huawei Ascend Confirmation: Any official announcement or verified third-party benchmark confirming H100-equivalence in the new-generation Ascend chip would materially reprice NVDA’s China TAM assumptions across institutional models. Track closely. NVDA