〔Overnight Brief〕AI Orders Stay Full as SpaceX Plans $24bn Texas Buildout, Yet Nvidia’s >4% Drop Opens a Leadership Test — Anthropic Presses IPO

0. Overnight Arc

The overnight arc was a two-speed AI trade: deployment announcements continued to expand, while Nvidia’s governance shock returned leadership, valuation, and the cost of the AI boom to the center of investor debate ^{[1][2][3][4][5]}

The tension is whether new model, chip, and compute cycles can keep attracting capital fast enough to offset the reputational and policy risk surrounding the sector’s leading supplier ^{[6][7][8][9][10]}

• **[NEW] Hardware demand:** AI-related PCB orders were full at several listed companies, some delivery schedules were tight, Kexiang announced capacity expansion, and the Wind Circuit Board Index rose more than 4% intraday ^[1]
• **[ESCALATED] Leadership risk:** Nvidia fell more than 4% to a near seven-month closing low after CEO Jensen Huang apologized to employees and the board opened an independent investigation ^[2][3]
• **[NEW] Catalysts:** Anthropic released Fable 5.2 and Mythos 5.2 before a reported Nasdaq IPO roadshow, while CoreWeave gained 5.56% after signing a multi-year Wayve agreement ^[6][11][12]

1. Infrastructure Orders and Capacity

• **[NEW] PCB demand:** AI-related PCB orders were full at several listed companies, some deliveries were tight, and Kexiang announced expansion of capacity for AI-compute PCBs ^[1]
• **[NEW] Space compute:** According to a Starbase municipal hearing notice, SpaceX plans to build, own, and operate a gas-fired power plant and supercomputing facility in Starbase, Texas, with initial investment expected at $24 billion ^[2]
• **[NEW] Terafab:** Musk’s Terafab chip-manufacturing project is in talks with Japanese suppliers, but the packet provides no project scale, financing, customer, or timetable ^[13]
• **[NEW] Custom silicon:** A market-wire item says Google unveiled two new TPU chips and East China’s first domestic thousand-GPU TPU cluster was completed, but it provides no chip specifications and does not identify the investee said to supply the cluster ^[8][14]
• **[NEW] Space cloud:** Zhongke Xingyun Space Intelligence Technology in Hefei began operations with RMB 1.5 billion in registered capital and 16 founding entities, aiming to move satellite-service scheduling into space and lower user access barriers ^[15]
• **[NEW] Robotics:** Qiyuan Robotics signed and unveiled a production base and national headquarters project in Hangzhou’s Fuyang district, with a three-party operating entity and headquarters operations center ^[16]
• **[ONGOING] Policy support:** Shanghai Party Secretary Chen Jining visited AI growth companies in Xuhui’s Model Speed Space and called for stronger ecosystem, infrastructure, resource allocation, and service support ^[17]

2. Model, Platform and IPO Catalysts

• **[NEW] IPO path:** The Wall Street Journal said Anthropic was preparing to push ahead with an IPO despite market jitters, IFR reported a Nasdaq roadshow for Tuesday morning, and Anthropic released Fable 5.2 and Mythos 5.2 ahead of the IPO ^[18][11][12]
• **[NEW] Rollout and price:** Anthropic opened Claude Fable 5.2 to all paid users across Claude Platform and major cloud platforms, charging $8 per million input tokens and $40 per million output tokens versus $10 and $50 previously ^[7]
• **[NEW] Model claims:** Anthropic says Fable and Mythos share the same underlying model and Fable adds strict anti-abuse safeguards; it also says the 5.2 versions outperform Opus 5.5 on nearly every evaluated capability and produce output 30% faster than their 5.1 predecessors ^[19][20]
• **[NEW] CoreWeave-Wayve:** Wayve will use Nvidia Blackwell GPUs on CoreWeave’s cloud to train, evaluate, and simulate autonomous-driving foundation models under a multi-year agreement, and CoreWeave closed up 5.56% ^[6]
• **[NEW] Distribution:** OpenAI put ChatGPT Enterprise on AWS Marketplace, while Alibaba Cloud raised monthly token-plan quotas by about 74-78% without changing the monthly fees for its Lite, Essential, or Standard plans ^[21][14]

3. Nvidia Governance and Valuation Shock

• **[ESCALATED] Price break:** Nvidia fell more than 4% to close at a near seven-month low, and secondary reporting said roughly $180 billion of market capitalization was erased in overnight trading ^[2][9]
• **[ESCALATED] Governance:** On November 8, Jensen Huang apologized to employees, admitted an extramarital affair, and Nvidia’s board opened an independent investigation; a company spokesperson said he would remain CEO and his duties would not be interrupted ^[3]
• **[ESCALATED] Control math:** Nvidia has no dual-class structure, and as of March 23, 2026, Huang controlled 870.6 million common shares equal to 3.58% of the company, while the Jensen Huang and Lori Huang Living Trust held 528.5 million shares equal to 60.7% of the shares he controlled ^[3]
• **[ONGOING] Fleet Intelligence link:** Secondary commentary says a Southeast Asian Nvidia Cloud Partner requirement to adopt Fleet Intelligence was reported before Huang’s apology and frames the sequence as hidden AI-infrastructure risk, but the packet does not establish a causal link ^[9]
• **[ONGOING] Evidence quality:** The supplied New York Times, USA Today, and CNBC entries are headline-only on the affair, alleged intelligence targeting, and scrutiny of Huang’s Trump ties, while Bloomberg supplies only a broad warning that the AI situation is more fraught ^{[22][23][24][5]}
• **[ONGOING] Investor framing:** MarketWatch said investors now face fresh questions over leadership, valuations, and the cost of the AI boom ^[4]

4. Positioning, Falsifiers and Source Discipline

• **[ESCALATED] Positioning shock:** After Bernie Sanders and other Democratic lawmakers proposed a ban on artificial superintelligence, the Philadelphia Semiconductor Index fell 12.25%, Nvidia fell more than 9.2%, and the Nasdaq fell more than 5.5% ^[10]
• **[ONGOING] Broker read:** A brokerage report said the bill’s landing probability was low, but identified the more important development as US investors beginning to price an AI-investment slowdown ^[10]
• **[NEW] Cross-market cushion:** Broad-based ETFs did not post large net inflows last Friday, A shares showed resilience, a RMB 36 billion active-ETF pool began building positions, margin-financing positions were trimmed for a second straight day, and retail panic appeared to ease ^[10]
• **[ONGOING] Trend caveat:** The same report said A shares were unlikely to gain a sustained trend until the US market stabilized, while expectations of tighter overseas restrictions strengthened the domestic-AI investment case ^[10]
• **[NEW] Falsifiers:** Three tests can resolve the split: demand and valuation at Anthropic’s Nasdaq roadshow, the outcome of Nvidia’s independent review and any change to Huang’s role or control, and US stabilization after the semiconductor selloff ^{[18][11][3][4][10]}
• **[NEW] Demand confirmation:** The infrastructure thesis needs evidence beyond announcements, including sustained PCB deliveries and concrete commercial schedules for SpaceX compute, Terafab, domestic TPU capacity, and space-cloud services ^{[1][2][13][8][15]}
• **[ONGOING] Thin watchlist:** A research-feed headline claims a Musk-linked space-compute deployment timetable and scaled humanoid-robot output for an unnamed global customer, but supplies no customer, volume, revenue, or timetable ^[25]
• **[ONGOING] Source discipline:** Treat the Terafab talks, domestic TPU-cluster report, and Fleet Intelligence sequence as single-feed watch items rather than confirmed operating trends ^[13][8][9]