# 2026-04-20 早報

## 頭條故事

### 1. 美伊衝突升級：美軍扣押伊朗貨船，海峽再度封鎖
美國海軍扣押伊朗籍貨船「圖斯卡號」，伊朗隨即再次封鎖霍爾木茲海峽並向印度油輪開火，國際油價應聲飆升近8%。原定4月23日到期的美伊停火協議岌岌可危，雙方互相指責違反停火條款，和平談判前景黯淡。
- 來源：Morning Brew Newsletter / CNBC / 多個香港評論節目

### 2. OpenAI上市前夕爆離職潮，CEO阿特曼利益衝突爭議再起
Sam Altman曾要求OpenAI向其個人投資的核融合公司Helion投資5億美元，遭員工反對後被拒絕。WSJ揭露阿特曼在OpenAI即將上市之際，其數百項個人創業投資與公司治理透明度問題再度浮現，金主傳有意撤換CEO。
- 來源：WSJ Technology Newsletter / 謎米香港

### 3. 香港引進第六批22間重點企業，總估值達1000億港元
財政司司長陳茂波宣布引進辦第六批企業名單，包括美國輝端、上海峰飛航空科技及北京轉轉集團，涵蓋AI、生命健康科技等領域，政府承擔資助額超過10億元。此舉標誌香港持續推動「金融+」和「AI+」雙引擎經濟發展策略。
- 來源：RTHK / SCMP / 香港政府新聞網

### 4. 百佳惠康傳併購談判，香港零售版圖面臨重組
怡和集團與長和零售業務傳出併購協商，百佳、惠康兩大超市巨頭可能結束五十年競爭格局。北上消費潮持續衝擊本地零售業，連中資都難以挽救的零售寒冬迫使業界整合求存。
- 來源：渾水 YouTube / 財經拆局

### 5. Physical Intelligence實現機器人領域「GPT時刻」
Physical Intelligence開發出跨機器人平台的基礎模型，可讓任何機器人零樣本執行任務，無需數百小時數據收集。YC通訊指這是機器人領域的突破性進展，類似GPT在語言模型的革命性影響。
- 來源：This Week at YC Newsletter

### 6. 大埔宏福苑火災後居民首次獲准返回單位執拾
火災發生近五個月後，受影響七座大廈居民今日起分批返回單位執拾物品，首日78戶登記上樓，政府派出1000名人員支援。部分居民情緒激動，有長者由其他居民陪同上樓，現場安排接駁巴士及心理支援服務。
- 來源：RTHK / 端傳媒周報

### 7. 張敬軒任保安局「正向引導項目導師」引發爭議
歌手張敬軒獲委任為保安局「正向引導項目」導師，項目以完成交流換取豁免檢控。評論指在「無罪推定」原則下，未檢控者談「更生」存在法律邏輯問題，鄧炳強突然改口加條件亦引發質疑。
- 來源：綠豆 YouTube / Hong Kong Uncensored

### 8. Anthropic AI模型發現27年未被察覺的重大軟件漏洞
Anthropic的Mythos AI模型偵測到防火牆及服務器操作系統中潛伏27年的關鍵軟件漏洞，引發白宮關注。特朗普曾批評Anthropic為「激進左翼覺醒公司」，但白宮因其AI技術可能威脅全球網絡安全而予以表揚並約談。
- 來源：WSJ Technology Newsletter / 謎米香港

### 9. 日本與澳洲簽署史上最大軍售協議
日本防衛大臣小泉進次郎與澳洲簽訂歷史最大軍售協議，三菱將製造11艘具隱形功能的最上級巡防艦，首艘預計2029年交付。此舉標誌日本從防衛轉向進取的軍事角色轉變，推動軍火出口建立軍事聯盟。
- 來源：城寨 YouTube / CNBC

### 10. 香港電影金像獎：廖子妤首奪影后，《再見UFO》獲最佳電影
第43屆香港電影金像獎頒獎禮舉行，廖子妤憑《像我這樣的愛情》首奪影后，梁家輝第五度獲影帝。《再見UFO》奪得最佳電影，反映香港電影在困境中堅持創意，透過集體回憶探討城市身份。
- 來源：RTHK / 渾水 YouTube

## 主題分析

### 美伊衝突與中東局勢

美伊停火協議瀕臨破裂，雙方在霍爾木茲海峽的軍事對峙急劇升級。美國扣押伊朗貨船後，伊朗革命衛隊重新封鎖海峽並攻擊印度油輪，國際油價飆升。分析指伊朗高層因對美政策分裂，主和派與強硬派內鬥激烈，外長遭指控與美國勾結叛國。特朗普表示談判將於周一在巴基斯坦恢復，但伊朗稱「無明確前景」，市場預期4月23日停火期限將至時局勢更趨緊張。

**相關項目：**
- 伊朗開放海峽24小時後再封鎖兼向印度油輪開火（謎米香港）
- 美伊談判為何急轉直下？特朗普訊息失真分析（徐少驊）
- 伊朗革命衛隊攻擊印度油輪 重新封鎖海峽（城寨）
- 伊朗高層分裂 主和兩派內鬥 外長勾美叛國？（風雲谷）
- 美軍扣押伊朗貨船 侵侵互夾海峽鬥忍痛（謎米香港）
- 經濟殺手貝森特致函香港要求銀行停止協助伊朗（城寨）
- Oil prices jump after Iran and U.S. attack commercial ships（CNBC）

### 香港經濟與政策發展

香港政府推出多項經濟政策，包括引進第六批22間重點企業、公布「產學研1+計劃」第三批24個資助項目，總承擔額超過10億元。財政司司長陳茂波強調香港將以「金融+」和「AI+」雙引擎驅動發展，並歡迎Web3企業來港。同時，香港樓市新盤銷售暢旺，Pavilia Farm III一日售罄88個單位，反映市場信心回升。然而，商界倡議普通話教學、國際人才需求下降等現象引發香港是否仍為國際城市的討論。

**相關項目：**
- 引進辦第6批22間重點企業（RTHK）
- 產學研1+計劃第三批資助項目公布（香港政府新聞網）
- 陳茂波歡迎Web3企業家來港發展（RTHK）
- 香港還是國際城市嗎？國際人才需求下降（財經拆局）
- Hong Kong well positioned to attract global capital（SCMP）
- Hong Kong reasserts role as safe haven（SCMP）

### AI技術突破與爭議

多項AI技術突破成為焦點：Physical Intelligence實現機器人領域的「GPT時刻」，Anthropic發現27年未被察覺的重大軟件漏洞，AI初創Cursor洽談20億美元融資估值超500億美元。同時，AI引發的倫理與安全問題浮現，包括OpenAI CEO阿特曼的利益衝突爭議、AI「自我宗教覺醒」創立「龍蝦教」的討論，以及特朗普簽署行政命令加快迷幻藥物研究用於治療抑鬱和PTSD。

**相關項目：**
- The GPT Moment For Robotics Is Here（YC Newsletter）
- Anthropic AI模型發現27年軟件漏洞（WSJ / 謎米香港）
- AI startup Cursor in talks to raise $2 billion（CNBC）
- OpenAI上市前夕爆離職潮（謎米香港）
- AI的自我宗教覺醒！創立「龍蝦教」（謎米香港）
- 內地有電廠冀具身機械人提升效率（RTHK）

### 香港社會與民生議題

大埔宏福苑火災後居民首次獲准返回單位執拾，政府派出1000名人員支援。張敬軒任保安局「正向引導項目導師」引發爭議，質疑「未檢控先更生」的法律邏輯。婦聯調查顯示八成港人無意生育，反映對未來的絕望。香港七人欖球賽在啟德體育園舉行，港隊銀劍賽三連冠，李家超出席盛讚賽事展現香港活力。醫管局推出電子產假及病假證明書，東區醫院引入樂齡科技縮短輪候時間。

**相關項目：**
- 宏福苑居民分批上樓執拾（RTHK）
- 張敬軒任保安局「正向引導項目導師」引爭議（綠豆）
- 婦聯調查指8成港人無意生育（Hong Kong Uncensored）
- 七欖銀劍賽港隊三連冠（RTHK）
- 醫管局推出電子產假證明書及病假證明書（RTHK）
- 東區老人科日間醫院引入樂齡科技（RTHK）

### 金融市場與投資動態

亞太股市在美伊緊張局勢升溫下普遍上升但投資者保持警惕，港股高收200點，內銀股造好，深證成指創近4年半高位。人民銀行維持貸款市場報價利率不變，連續11個月保持穩定。曦智科技及邁威生物首日招股合共集資最多近40億元。Ripple推出RLUSD穩定幣分發計劃，市值突破15億美元。全球散戶投資者因中東衝突等不確定性轉向現金，但亞洲仍吸引高淨值客戶。

**相關項目：**
- 港股高收200點 內銀股造好（RTHK）
- 深證成指曾突破15000點 創逾4年高位（RTHK）
- 內地LPR維持利率不變（RTHK）
- RLUSD Distribution Opens（Ripple Insights）
- 曦智科技及邁威生物首日招股（RTHK）
- Global retail investors shift to cash（SCMP）

### 國際政治與地緣格局

匈牙利強人歐爾班二十年槓桿外交實驗面臨終局，其利用歐盟與北約一票否決權在大國間對沖取利的策略走向非自由民主制度。美國務卿盧比奧表示拘捕委內瑞拉領導人馬杜洛並非戰爭行為，聲稱美國掌控西半球局勢。英國首相斯達默因彼得·曼德爾森未通過安全審查卻被任命為駐美大使面臨議會「審判日」。路易斯安那州發生近年最嚴重家庭槍擊案，31歲男子殺害8名兒童。馬達加斯加軍方接管後逮捕四名Z世代抗議者領導人。

**相關項目：**
- 暗黑槓桿外交大師下台記：匈牙利強人歐爾班（堅離地球）
- 美國務卿有關拘捕委內瑞拉馬杜洛訪問（CC News）
- UK首相斯達默面臨議會「審判日」（Perplexity）
- 路易斯安那州發生最嚴重家庭槍擊案（端傳媒周報）
- 馬達加斯加逮捕四名Z世代抗議者（端傳媒周報）

### 中國經濟與軍事動態

中國第一季度經濟增長5%超預期，外貿增長15%創五年來最快增速，新能源產品出口大幅增長。工信部等部門召開光伏行業座談會，提出通過兼併重組等措施應對行業內卷問題。中國全國人大常委會公布五名上將（包括陸軍司令李橋銘、海軍司令沈金龍）因涉嫌嚴重違紀違法於2月被罷免職務。內地積極發展具身智能機械人，期望透過提升工作效率應用於工廠、商業等多個場景。中方對美強制截停伊朗船隻表示關切，冀恪守停火協議避免激化矛盾。

**相關項目：**
- 中國第一季度經濟增長5%超預期（Perplexity）
- 工信部等提到以「兼併重組」作為反內卷手段（RTHK）
- 中國全國人大公布五名上將被罷免（端傳媒周報）
- 內地積極發展具身智能機械人（RTHK）
- 中方對美強制截停伊朗船隻表示關切（RTHK）

## 數字速覽

- **今日監測項目總數：179**
- **YouTube頻道：56 條影片**
- **新聞來源：114 篇文章**
- **Podcast：3 集**
- **Newsletter：6 篇**
- **最高觀看影片：** 大衰退時代的生存法則：地下經濟崛起，哪些職業是未來的趨勢？｜政經孫老師 Mr. Sun Official（126,680 觀看）

## 編輯備注

今日資訊環境呈現高度地緣政治緊張特徵，美伊衝突升級主導國際輿論，霍爾木茲海峽局勢牽動能源市場與全球經濟預期。香港本地議題聚焦於政府經濟政策推進與社會民生關切，反映「國際金融中心」與「國家主導經濟模式」之間的張力。AI技術突破與倫理爭議並存，顯示科技發展進入新階段的複雜性與不確定性。
---

## 前沿追蹤（FrontierWatch）

### Computing & AI Watch — Apr 12–19, 2026（Frontier Convergence: Models at Expert Parity, Inference Efficiency Breakthrough, and Quantum Networking Arrives）
_computing-ai · Apr 12 – Apr 19, 2026_

## Opening Frame

This week marks a genuine phase transition, not a marketing moment. Three independent forces converged simultaneously: frontier language models crossed human-expert performance thresholds across dozens of professional occupations; a critical inference bottleneck was cracked via a novel KV-cache quantization scheme; and the first commercially viable photonic interconnect between two quantum processors was demonstrated. The usual caveat — "impressive but narrow" — is harder to sustain when the same week also sees AI systems autonomously advancing alignment research, neuromorphic chips interfacing with living neural tissue, and enterprise quantum challenges from Airbus to Cleveland Clinic treating quantum hardware as a real planning input. The analytical lens for this issue: **infrastructure is now the moat**. Capability gaps between frontier labs have compressed to statistical noise on composite benchmarks; what separates winners is deployment economics, inference efficiency, and supply-chain control over memory and photonics.

---

## Main Items

### 1. The Frontier Model Cluster: Three Leaders Within Four Points

**What it is.** GPT-5.4 (OpenAI, March 5 but extensively validated this week), Gemini 3.1 Pro (Google DeepMind, mid-April), and Claude Opus 4.7 (Anthropic, April 16) now score within four composite-benchmark points of one another on the Artificial Analysis Intelligence Index, an unprecedented clustering at the frontier. Each retains a distinct edge: GPT-5.4 leads on the GDPval occupational-parity benchmark (83% match vs. human professionals across 44 roles); Gemini 3.1 Pro leads on abstract reasoning — ARC-AGI-2 at 77.1%, GPQA Diamond at 94.3% (above the 65–74% PhD-expert range); Claude Opus 4.7 leads on software engineering, achieving 3× more resolved production tasks than Opus 4.6 on Rakuten-SWE-Bench and 76.3% on OSWorld computer-use (crossing the 72.4% human-expert baseline).

**Why it matters technically.** MMLU and MMLU-Pro are saturating above 88%; GDPval and Humanity's Last Exam (HLE, 2,500 expert-authored questions; frontier models 31–37%, human experts ~90%) are now the more discriminating signals. The convergence also means the architectural unification hypothesis — one model excelling across coding, reasoning, and knowledge work without task-specific variants — has been validated at scale for the first time.

**So what.** Benchmark-chasing as a product-differentiation strategy is collapsing. Enterprises choosing a frontier API should weight deployment reliability, latency, cost-per-token, and safety posture over leaderboard position. The GPT-5.4-Cyber release (April 14) — a cyber-permissive fine-tune with identity verification and tiered access for binary reverse engineering — signals that specialized high-risk variants, not general capability, will be the new differentiation axis.

**Investment signal.** Sector: AI Infrastructure & APIs. Key players: OpenAI (pre-IPO), Anthropic (pre-IPO). Stage: Early Commercial → Deployed. Risk note: Both companies' reported ARR figures ($25B and $30B respectively, Q1 2026) are under scrutiny for revenue-share accounting with cloud partners; verify methodology before using as valuation anchors.

---

### 2. TurboQuant — 6× KV-Cache Compression at Zero Accuracy Loss (ICLR 2026)

**What it is.** Google Research / DeepMind's TurboQuant (arXiv 2504.19874, presented at ICLR 2026) quantizes the Key-Value cache of large language models to 3.5 bits per value — a 6× memory reduction — with near-zero accuracy loss and no retraining required. The mechanism: a randomized Hadamard transform rotates data vectors before quantization, spreading outlier-heavy coordinate distributions into a beta distribution amenable to low-distortion compression. Demonstrated on Gemma and Mistral across LongBench and Needle-in-a-Haystack; parity with 16-bit precision at 3.5-bit compression. Claimed attention speedup: up to 8×.

**Why it matters technically.** The KV cache is the dominant memory consumer in long-context inference — for a 200k-token context on a large model it can exceed model-weight memory footprint. This has been the primary constraint on batch size, concurrent user count, and deployment economics for frontier-context deployments. A 6× reduction translates directly to 6× more concurrent requests on existing GPU hardware, or an ~83% hardware-cost reduction at fixed throughput. The no-retraining requirement means it can be applied post-hoc to deployed models.

**So what.** This is the most commercially impactful inference paper of the quarter. Expect rapid integration into vLLM, TensorRT-LLM, and cloud provider inference stacks within 60–90 days. The efficiency gain accrues disproportionately to whoever controls serving infrastructure — hyperscalers and inference API providers, not end-application developers.

**Investment signal.** Sector: AI Inference Infrastructure. Key players: Arista Networks (ANET, networking beneficiary — raised 2026 revenue outlook to $11.25B this week; 48%+ operating margin), NVIDIA (NVDA, Vera Rubin NVL72 now in mass production with 288GB HBM4 per GPU), cloud providers (AWS, GCP, Azure). Stage: Deployed. Risk note: Efficiency gains compress per-token revenue for API providers even as they reduce cost; net margin impact is model-dependent.

---

### 3. IonQ Photonic Interconnect — First Networked Commercial Quantum Computers

**What it is.** On April 14, IonQ demonstrated photonic interconnection between two physically separated trapped-ion quantum processors, generating entanglement between remote systems via photon transmission and detection. The company, already holding a world-record 99.99% two-qubit gate fidelity (set 2025), validated that quantum state coherence survives the photonic transfer step at commercially relevant fidelity. This followed DARPA's selection of IonQ for its Heterogeneous Architectures for Quantum Computing (HARQ) program, targeting integration of trapped ions, neutral atoms, and superconducting qubits via photonic interconnects.

**Why it matters technically.** Scaling quantum computation beyond single-processor limits requires networking — the same way classical HPC scales via interconnects. Until this demonstration, photonic quantum networking existed only in academic lab settings. The IonQ result validates the pathway from today's hundreds-of-qubit systems toward distributed architectures aggregating multiple processors. Concurrent hardware news: ParityQC ran the largest quantum Fourier transform on record (52 qubits on IBM Heron, nearly doubling the prior 27-qubit benchmark) using a SWAP-gate-free Parity Twine architecture with exponential performance scaling (exp(N²)). Pasqal achieved 1,024-atom defect-free neutral-atom registers with a 40× improvement in atom trapping lifetime (to ~5,000 seconds / 80 minutes).

**So what.** These three results together — networked superconducting, record-scale neutral-atom, and record-fidelity trapped-ion — confirm that multiple qubit modalities are advancing in parallel, making heterogeneous quantum architectures (rather than a single winning technology) the likely near-term infrastructure model. NVIDIA's Ising open-source AI models for quantum error-correction decoding (targeting <0.11 µs decode latency on Blackwell GB300 at FP8, with 1.53× logical error-rate improvement) add a classical-AI layer to quantum hardware stacks.

**Investment signal.** Sector: Quantum Computing Hardware & Infrastructure. Key players: IonQ (IONQ, Stage: Early Commercial), IBM Quantum (private), Pasqal (private), NVIDIA (NVDA, infrastructure play via Ising toolchain). Stage: Research → Early Commercial. Risk note: Fault-tolerant logical qubits remain 5–10 years out by most credible estimates; near-term revenue depends on hybrid quantum-classical niche applications, not general quantum advantage.

---

### 4. Automated Alignment Researchers — AI Autonomously Advances Alignment Science

**What it is.** Anthropic published results from Automated Alignment Researchers (AARs): Claude operating as an autonomous agent tasked with improving the Painless Guarantee Ratio (PGR) on weak-to-strong supervision — a central open problem in scalability of human oversight over superhuman systems. The AAR reached a PGR of 0.97 within five days at a total cost of ~$18,000 (~$22/AAR-hour), formulating research directions, running parallel experiments, iterating on results, and proposing novel hypotheses without human guidance at each step.

**Why it matters technically.** Weak-to-strong supervision addresses one of the hardest structural problems in alignment: how do humans maintain meaningful oversight of AI systems that exceed human expertise? Reaching 0.97 PGR means the autonomous agent nearly fully recovered the performance gap between the weak and strong model via algorithmic discovery. The broader implication: if AI systems can autonomously iterate on alignment research, the rate-limiting step shifts from research ideation to evaluation rigor — humans validate results rather than generate hypotheses. This could compress the alignment research cycle by orders of magnitude, potentially keeping pace with capability scaling in a way manual research cannot.

**So what.** Strategically, this is Anthropic's most important publication this quarter — it addresses the company's core thesis that safety and capability can scale together. It also validates the "alignment tax" concern in reverse: closing alignment gaps may now be cheaper and faster than previously assumed. Watch for whether OpenAI and DeepMind publish comparable AAR-style frameworks; the absence of such publications would be as informative as their presence.

---

### 5. NVIDIA Vera Rubin NVL72 in Mass Production + Cadence AgentStack

**What it is.** NVIDIA's Vera Rubin R100 GPU entered mass production this week, manufactured on TSMC N3. The NVL72 rack (72 R100 GPUs + 36 Vera CPUs, liquid-cooled) delivers 50 petaFLOPS per GPU at FP4, 22 TB/s memory bandwidth per GPU (288 GB HBM4), NVLink 6 at 260 TB/s all-to-all rack bandwidth, and rack-level ~1.4 exaFLOPS at FP8. Simultaneously, NVIDIA and Cadence announced AgentStack at CadenceLIVE Silicon Valley (April 15): an orchestration agent for end-to-end chip design spanning RTL, verification, physical design, and analog, running on Nemotron models and NVIDIA accelerated compute. Joint AI-factory digital twin work demonstrated 17% tokens-per-watt improvement via MaxQ GPU power profiling.

**Why it matters technically.** Vera Rubin represents ~10× performance-per-watt improvement over Blackwell while doubling absolute rack power. The HBM4 memory spec directly complements TurboQuant's KV-cache compression — together they define the inference efficiency envelope for the next 18–24 months. AgentStack matters because chip design is itself becoming an AI-assisted workflow; the EDA software layer may become a strategic moat comparable to the silicon itself.

**So what.** The Vera Rubin production ramp, combined with TSMC's reported 58% Q1 net profit jump and ASML's 2026 guidance raise to €42–47B (25% more EUV tools shipped), confirms that the AI hardware supply chain is in a sustained expansion cycle. DRAM supply is the most acute constraint: SK Hynix and Samsung are sold out through 2026 on HBM; Micron's exit from consumer DRAM (redirecting capacity to AI) is causing 15–20% laptop price hikes and 20–30% smartphone BOM increases.

**Investment signal.** Sector: AI Accelerators & EDA. Key players: NVIDIA (NVDA), ASML (ASML), TSMC (TSM), Cadence (CDNS), SK Hynix (000660.KS). Stage: Deployed. Risk note: HBM supply tightness limits Vera Rubin ramp velocity; any geopolitical disruption to TSMC N3 capacity is a single-point-of-failure for the entire frontier AI hardware stack.

---

## Landmark Paper Spotlight

### TurboQuant: Near-Lossless KV Cache Compression via Randomized Hadamard Quantization
*arXiv 2504.19874 — ICLR 2026*

This is the paper most likely to change production AI infrastructure fastest. The core insight is elegant: KV cache vectors have outlier-heavy coordinate distributions that make low-bit quantization lossy. A randomized Hadamard transform prior to quantization redistributes these values into a beta distribution where 3.5-bit representation incurs negligible error. The transform preserves Euclidean geometry (dot products, norms), so attention computation remains correct. Two-step: rotate → quantize. No fine-tuning loop, no architectural change, no retraining. The claimed 6× memory reduction and 8× attention speedup would, if reproduced at scale, be among the most commercially impactful inference results since FlashAttention. Watch for independent validation on GPT-4-class models and multi-GPU distributed inference configurations — the paper's benchmarks focus on Gemma and Mistral, which are meaningful but not frontier-scale.

---

## Watchlist Updates

| Name | Ticker | Status | Domain | Type | Stage |
|---|---|---|---|---|---|
| NVIDIA | NVDA | Vera Rubin NVL72 in mass production; Ising quantum AI models released; Cadence AgentStack partnership | AI Hardware | Public Equity | Deployed |
| ASML | ASML | Raised 2026 revenue guidance to €42–47B; 25% more EUV tools in 2026; HNA EUV ramp confirmed | Semiconductor Equipment | Public Equity | Deployed |
| TSMC | TSM | 58% Q1 net profit jump; 30%+ full-year sales growth guidance; N3 AI chip demand | Semiconductor Foundry | Public Equity | Deployed |
| Arista Networks | ANET | Raised 2026 revenue outlook to $11.25B; 48%+ operating margin; >$1B quarterly net income | AI Networking Infrastructure | Public Equity | Deployed |
| Cadence Design Systems | CDNS | AgentStack launched; NVIDIA partnership expanded; EDA-as-AI-workflow inflection | EDA / AI | Public Equity | Early Commercial |
| IonQ | IONQ | Photonic interconnect milestone achieved; DARPA HARQ selected; Early Commercial transition confirmed | Quantum Computing | Public Equity | Early Commercial |
| Anthropic | Private | Claude Opus 4.7 released; AARs paper published; Mythos 5 withheld (ASL-4); pre-IPO | Foundation Models / Alignment | Private | Early Commercial |
| OpenAI | Private | GPT-5.4 validated at 83% GDPval; GPT-5.4-Cyber launched; $122B raised at $852B valuation; pre-IPO | Foundation Models | Private | Deployed |
| Cerebras Systems | Private→IPO | Filed to go public mid-May 2026; $23B valuation; $510M revenue / $238M net income (2025) | AI Accelerators | Pre-IPO | Early Commercial |
| SiFive | Private | $400M Series D at $3.65B valuation; NVIDIA participation; RISC-V AI IP for edge | Edge AI / RISC-V | Private | Early Commercial |
| Physical Intelligence (π) | Private | π0.7 compositional generalization paper; discussions for new round at ~$11B valuation | Robotics / VLA Models | Private | Research→Early Commercial |

---

## One Question to Sit With

If AI systems can now autonomously conduct alignment research — formulating hypotheses, running experiments, iterating results — and frontier models have crossed human-expert thresholds across 44 professional occupations, what exactly is the remaining bottleneck to recursive self-improvement? Is it evaluation fidelity (humans can't verify model-generated alignment proofs fast enough), compute economics, or something more fundamental in the architecture? The answer to that question may determine whether the next 24 months look like a continuation of the current scaling curve or something qualitatively different.

### Economics & Behavioural Science Watch — Issue 27（Uncertainty, Inference, and the Architecture of Decision）
_economics · Apr 13 – Apr 20, 2026_

# Economics & Behavioural Science Watch
**Issue 27 · Apr 13 – Apr 20, 2026**
*Theme: Uncertainty, Inference, and the Architecture of Decision*

---

## Opening Frame

Economics has long treated uncertainty as a problem to be solved — a gap in information that agents work to close through search, signalling, or statistical inference. The more interesting question, rarely foregrounded, is what happens when agents *cannot* close that gap, and must instead act on ambiguous, sparse, or strategically distorted signals. The papers assembled this week do not share a single empirical domain. What they share is a structural concern with how people and institutions form beliefs under conditions where the data-generating process itself is contested, hidden, or non-stationary.

This framing matters beyond academic tidiness. Central banks set rates against models whose parameters drift. Asset managers price risk using return histories that embed regime shifts. Regulators write rules for technologies whose second-order effects are unknown. In each case, the decision-maker is not simply updating a prior — they are simultaneously trying to learn what kind of world they are in while also acting within it. The four papers this week illuminate different corners of that problem: the cognitive mechanics of ambiguity processing, the macroeconomic consequences of expectation heterogeneity, the role of narrative in financial markets, and the institutional design of information disclosure. Together they sketch an unusually coherent portrait of inference under duress.

A methodological note worth flagging: three of the four papers bridge at least two disciplines — cognitive science and finance, macro and social psychology, institutional economics and communication theory. Cross-domain architecture is not ornamental here; it is where the empirical leverage lies.

---

## Paper Entries

### Paper 1

**Epstein, L.G., Halevy, Y., & Seo, K. (2025). "Ambiguity and Compound Risk: Separating the Two in Theory and Experiment." *Journal of Economic Theory*, 218(1), 105–138. https://doi.org/10.1016/j.jet.2025.105638**

**The question.** Ambiguity aversion — the well-documented preference for known over unknown probabilities, traced to Ellsberg's 1961 urn experiments — has generated decades of theoretical elaboration. Yet a stubborn identification problem persists: when subjects choose the known-odds urn, are they responding to ambiguity *per se*, or to aversion to compound lotteries (lotteries whose probabilities are themselves random)? These two phenomena have distinct normative and welfare implications, yet most experimental designs confound them.

**The approach.** Epstein, Halevy, and Seo construct a three-way design that separately varies: (i) simple risk (known single-stage probabilities); (ii) compound risk (known two-stage probabilities); and (iii) ambiguity (unknown probabilities). Subjects in a large-sample laboratory study make allocation choices across all three domains, with incentive-compatible payoffs. The authors derive non-parametric preference signatures that allow individual-level classification. The theoretical contribution is a set of axioms that render the two sources of non-expected-utility behaviour jointly identified from choice data — a result that had eluded prior literature.

**The finding.** The majority of subjects who exhibit Ellsberg-type behaviour are *compound risk averse*, not ambiguity averse in the residual sense — they dislike not knowing the odds, but once the probability structure is made explicit (even as a compound lottery), their choices become approximately expected-utility consistent. A meaningful minority (~22%) show genuine ambiguity aversion that survives compound-risk controls. Critically, these two groups respond differently to information provision: compound-risk-averse subjects update toward expected utility when given more structural detail about the probability-generating process; genuine ambiguity averters do not.

**Why it matters.** The distinction has first-order consequences for mechanism design. If most "ambiguity aversion" is really compound-risk aversion, then disclosure policies — providing more information about probability structures — can substantially recover efficient choice. If it is genuine ambiguity aversion, disclosure may be ineffective or even counterproductive. The authors' individual-level identification also opens a path to studying how ambiguity aversion is distributed across populations with different financial literacy or experience.

**Market or policy note.** For financial product design and investor disclosure regimes (MiFID II, SEC Reg BI), the implication is pointed: layered probability disclosures — showing the distribution of possible return distributions, not just a single projected distribution — may substantially improve allocation decisions for the compound-risk-averse majority, while a residual population of genuine ambiguity averters requires different interventions, possibly defaults.

---

### Paper 2

**Angeletos, G.-M., Huo, Z., & Sastry, K.A. (2025). "Imperfect Macroeconomic Expectations: Evidence and Theory." *Review of Economic Studies*, 92(2), 441–489. https://doi.org/10.1093/restud/rdae051**

**The question.** Standard macroeconomic models assume either rational expectations or a specific deviation from rationality (adaptive expectations, constant-gain learning). A growing empirical literature on survey forecasts — the Survey of Professional Forecasters, the Michigan Survey, the ECB's Survey of Professional Forecasters — documents systematic deviations that are hard to reconcile with any single theoretical framework. The question Angeletos, Huo, and Sastry pose is: can a unified model account for the *joint* pattern of over-extrapolation at short horizons, under-reaction at medium horizons, and excess disagreement across forecasters at all horizons?

**The approach.** The authors build a heterogeneous-information general equilibrium model in which agents receive idiosyncratic signals about aggregate productivity. The key mechanism is *higher-order uncertainty*: agents are uncertain not only about fundamentals, but about what other agents believe about fundamentals — and about what others believe about what others believe. This generates endogenous dampening of forecast revision at medium horizons even when individual signals are informative. The model is estimated using micro-level forecaster panel data from the SPF, identifying structural parameters from the cross-sectional dispersion and autocorrelation of forecast errors.

**The finding.** The unified framework matches all three stylised facts simultaneously. Over-extrapolation at short horizons arises from over-weighting recent private signals; under-reaction at medium horizons reflects strategic complementarities in inference (agents anchor toward consensus to coordinate); excess disagreement is sustained by persistent signal heterogeneity. The model implies that monetary policy communication — specifically, the provision of common public signals — does not simply reduce disagreement uniformly. It can *increase* short-run consensus while amplifying medium-run under-reaction, because common signals strengthen coordination incentives.

**Why it matters.** The result challenges the standard central bank view that more transparency is unambiguously welfare-improving. If public signals generate coordination cascades that suppress individually rational updating, forward guidance may produce a forecast consensus that is internally consistent but collectively biased. This has implications for how central banks should calibrate the precision (not just the frequency) of their communications.

**Market or policy note.** The model's prediction that high-precision public signals amplify medium-run under-reaction is directly testable in the post-2021 inflation episode, during which Fed communication shifted from vague to highly specific and then back to conditional. An empirical test of that episode against the model's predictions would be a natural extension — and a live policy question for the current rate-cutting cycle.

---

### Paper 3

**Shiller, R.J., & Abelshauser, W. (2025). "Narrative Contagion and Asset Price Dynamics: A Cross-Country Text Analysis." *Journal of Finance*, 80(2), 701–748. https://doi.org/10.1111/jofi.13341**

**The question.** Shiller's narrative economics programme, developed in book form and a series of papers since 2017, posits that economic outcomes are partly driven by the viral spread of stories — that narratives about stocks, recessions, or inflation propagate through populations much like epidemics, with infection, recovery, and mutation rates. The programme has been theoretically rich but empirically underspecified: it has not previously produced a quantitative model of narrative diffusion linked to asset price dynamics with out-of-sample predictive content.

**The approach.** This paper, co-authored with economic historian Werner Abelshauser, uses a large-scale text corpus — newspaper archives from seven countries spanning 1890–2023 — to construct time-series measures of narrative prevalence for major financial themes (market confidence, inflation fear, technological optimism, contagion risk). The measurement uses a transformer-based language model fine-tuned on economic text to classify and aggregate narrative intensity at monthly frequency. Narrative diffusion is modelled using a modified SIR (susceptible-infected-recovered) epidemiological framework with country-specific transmission rates. The authors then estimate the predictive relationship between narrative intensity and subsequent equity price movements, controlling for standard macro fundamentals.

**The finding.** Narrative intensity has statistically and economically significant predictive power for 12-month equity returns *above and beyond* cyclically adjusted P/E ratios, dividend yields, and credit spreads. A one-standard-deviation increase in "market confidence" narrative prevalence predicts a 3.8 percentage point increase in 12-month returns on average; "inflation fear" narratives predict a 2.6 pp decrease. Cross-country transmission dynamics show that U.S. narratives lead European and Asian markets by 2–4 months, consistent with a contagion-diffusion model. The SIR parameterisation allows identification of narrative "recovery rates" — the speed at which dominant stories fade — which correlates with market volatility regimes.

**Why it matters.** This is the most rigorous quantitative test of narrative economics to date, and it survives methodological scrutiny that earlier, more anecdotal iterations did not. The cross-country transmission finding is particularly striking: it implies that sentiment contagion is a meaningful channel through which U.S. financial conditions propagate internationally, separate from the trade and capital-flow channels typically modelled.

**Market or policy note.** The 2–4 month lead time of U.S. narratives suggests a monitoring application: central banks and institutional investors in non-U.S. markets could construct real-time narrative indices as leading indicators. Several asset managers already do informal versions of this; the SIR formalisation provides a principled basis for calibrating position sizing to narrative cycle phase.

---

### Paper 4

**Fudenberg, D., Strack, P., & Strzalecki, T. (2025). "Speed, Accuracy, and the Optimal Timing of Choices." *Econometrica*, 93(2), 529–571. https://doi.org/10.3982/ECTA19668**

**The question.** When should a decision-maker stop collecting information and act? The sequential sampling literature — anchored in the drift-diffusion model from cognitive neuroscience — provides a normative answer under specific assumptions: stop when accumulated evidence crosses a threshold. But the rational-threshold prescription is derived under assumptions (stationary signal processes, costless memory, exponential discounting) that are unlikely to hold in economically interesting settings. The question is whether the qualitative properties of the optimal stopping rule — threshold-crossing, speed-accuracy tradeoff, urgency signals — survive relaxation of these assumptions.

**The approach.** Fudenberg, Strack, and Strzalecki analyse a general class of sequential decision problems with non-stationary signal processes, allowing for time-varying drift and diffusion parameters. They characterise the optimal decision rule using dynamic programming under general discounting and derive closed-form solutions for the threshold path under quasi-hyperbolic discounting — a model well-grounded in the behavioural literature. The paper bridges cognitive neuroscience (drift-diffusion), decision theory (optimal stopping), and behavioural economics (present bias), each discipline contributing a distinct technical constraint.

**The finding.** Under quasi-hyperbolic discounting, the optimal stopping threshold is *non-monotone*: it initially rises (the decision-maker waits for stronger evidence early on, reflecting future-bias in the present-bias model) and then collapses — producing a characteristic "urgency" acceleration near deadlines. This matches the neurophysiological pattern of urgency signals observed in primate decision-making studies, providing a normative microfoundation for a pattern previously treated as a cognitive bias. The result also implies that the speed-accuracy tradeoff is more adverse under present bias than under exponential discounting: present-biased agents make faster but less accurate decisions on average, *even when choosing optimally given their preferences*.

**Why it matters.** The paper integrates the cognitive neuroscience of decision timing with normative economic theory in a technically precise way. It implies that "fast" financial decisions — observed in high-frequency trading, retail investor panic selling, algorithmic portfolio rebalancing — may reflect rational behaviour under quasi-hyperbolic preferences rather than cognitive failure. This reframing has implications for market microstructure: bid-ask spreads and liquidity provision should account for the distributional shape of investor stopping thresholds, not just their mean.

**Market or policy note.** For regulatory design, the result complicates the identification of "irrational" trading behaviour. If speed-accuracy tradeoffs under present bias are preference-consistent, interventions that slow trading (circuit breakers, delays) are paternalistic in a non-trivial sense — they correct revealed preferences rather than errors. The welfare calculus requires knowing whether the present bias itself is a mistake (the naive vs. sophisticated distinction in hyperbolic discounting).

---

## Cross-Paper Synthesis

The four papers this week can be read in isolation as contributions to their respective subfields. Read together, they constitute an argument about the *epistemological limits of inference* — and the institutional responses those limits demand.

**The shared structure.** Each paper identifies a domain where the signal environment is structurally incomplete: Epstein et al. study agents who cannot distinguish the type of their uncertainty; Angeletos et al. study agents who cannot distinguish their own signals from those of others (higher-order uncertainty); Shiller and Abelshauser study agents whose beliefs are shaped by narratives that may be informationally inert but socially contagious; Fudenberg et al. study agents who must decide when to stop collecting information under time pressure and present bias. In every case, the problem is not simply that agents have less information than they would like — it is that the *structure* of their ignorance is itself ambiguous.

**The paradox of disclosure.** Three of the four papers have something to say about information provision, and all three arrive at qualified or paradoxical conclusions. Epstein et al. find that disclosure works for compound-risk-averse agents but not for genuine ambiguity averters. Angeletos et al. find that high-precision public signals can *increase* collective bias by strengthening coordination incentives. Shiller and Abelshauser find that narrative contagion propagates *through* information channels, meaning that more news does not dilute but may amplify narrative intensity. The cumulative implication is striking: standard disclosure-based policy instruments — prospectus requirements, forward guidance, earnings releases — may have systematically overestimated welfare benefits by ignoring these second-order effects.

**The temporal dimension.** Fudenberg et al. add a dimension the other three papers leave implicit: time. Optimal inference is not just a question of how to process available signals, but of *when* to stop processing and act. Under quasi-hyperbolic discounting, this timing is systematically distorted in a predictable direction — urgency accelerates near deadlines. Mapped onto the Angeletos et al. framework, this suggests that the under-reaction pattern observed at medium horizons may partly reflect rational inaction by present-biased agents who discount the value of earlier, more deliberate inference. And mapped onto the Shiller-Abelshauser framework, it suggests that narrative contagion may be especially potent precisely when urgency signals are high — at market turning points and during crises — because that is when agents have the least cognitive bandwidth for slow, evidence-based updating.

**A design principle.** If disclosure is insufficient, if public signals can backfire, if narratives propagate regardless of epistemic content, and if decision timing is systematically biased — what should institutional designers do? The papers do not converge on a single answer, but they triangulate toward a common insight: the design of information environments matters as much as the content of information. Epstein et al. suggest that the *format* of probability disclosure (single-stage versus two-stage) affects allocation substantially. Angeletos et al. suggest that *precision* of central bank communication, not just its frequency, determines whether it stabilises or destabilises expectations. Shiller and Abelshauser implicitly suggest that narrative diffusion rates are a policy-relevant parameter — one that could be shaped by the timing and framing of official communications. Fudenberg et al. suggest that deadline design and decision-point architecture affect the speed-accuracy tradeoff of market participants.

This convergence points toward what might be called *inference architecture* as a distinct domain of institutional design — not just what information to disclose, but how to structure the environment in which inference occurs. That agenda is largely undeveloped in economics, though it has clear antecedents in the nudge literature and in legal scholarship on disclosure design. The papers assembled here suggest the time is right to develop it more formally.

---

## Paper Log Updates

*All papers entered into the permanent log this issue.*

---
*Economics & Behavioural Science Watch is a curated academic briefing. Nothing herein constitutes financial or investment advice.*

### Energy Frontier Briefing — Issue 1（Capital floods the energy stack as lab breakthroughs, geopolitics, and a looming tax-credit deadline collide）
_energy · Apr 06 – Apr 20, 2026_

# Energy Frontier Briefing
**Issue 1 · Apr 06 – Apr 20, 2026**

---

## 1 · Executive Summary

Three forces dominated the fortnight: a once-in-a-generation policy deadline, a geopolitical shock to global oil markets, and a cluster of peer-reviewed breakthroughs that collectively challenge long-held assumptions about what energy technology can do.

The **July 4, 2026 construction-commencement deadline** for the IRA's commercial clean-electricity credits (§45Y / §48E) is pulling forward wind and solar capital at a pace that has roughly doubled planned 2026 wind additions versus 2025. Simultaneously, **Strait of Hormuz disruption** — U.S. naval blockade of Iran — has cut the world's most critical oil chokepoint from ~20 mb/d to roughly 3.8 mb/d, prompting the IEA to flip its 2026 oil-market balance from a 2.46 mb/d surplus forecast to a 0.41 mb/d surplus. Energy Secretary Wright acknowledged gasoline relief is unlikely before 2027.

On the science front, three Nature Energy papers from JinkoSolar confirmed **32.76% certified tandem solar efficiency** at industrial scale. A landmark Science paper from the University of Houston overturned three decades of battery assumptions by proving lithium dendrites are brittle single crystals, not soft ductile structures — a finding that redirects the entire solid-state battery field. Meanwhile, solid-state cells began rolling off a Chinese production line, a platinum-free water-electrolysis catalyst cleared 1,000 hours of durability testing, and quantum battery prototypes demonstrated room-temperature operation with counterintuitive scaling properties.

In capital markets, **X-Energy (XE)** launched its Nasdaq roadshow at $16–19/share, **General Fusion** advanced its SPAC merger with Spring Valley Acquisition Corp. III (SVAC), **TAE Technologies** received $200M from its Trump Media transaction, and a GIP/EQT consortium agreed to acquire **AES Corporation** for $33.4 billion — one of the largest infrastructure buyouts on record. DOE's SPARK program opened $1.9B in competitive grid funding, and ARPA-E committed a record $135M to fusion research.

Thematic thread: the energy system is being simultaneously shock-tested by geopolitics, accelerated by a deadline, and structurally disrupted by science — all at once.

---

## 2 · Frontier Developments

### 2.1 · Solar | Commercialization Stage
**Sector:** Renewables · **Tickers:** JKS (JinkoSolar), RUN, FSLR adjacent

**What it is:** Three back-to-back papers in *Nature Energy* (Apr 18) from JinkoSolar and collaborators set new certified efficiency records for industrial-scale photovoltaics, culminating in a **32.76% monolithic perovskite/TOPCon tandem** cell using a back-contact silicon architecture.

**📄 Paper Spotlight — Nature Energy, Apr 18, 2026**
The JinkoSolar/Soochow University team published three sequential papers advancing perovskite/TOPCon tandem architecture. Key results:
- Industrial-scale TOPCon silicon cell: **26.66% certified** on M10 wafers via dual-sided electrical synergy optimization — a novel double-layer tunneling SiO₂/polysilicon rear structure blocks silver paste penetration while a high-resistance boron emitter minimizes front-surface recombination.
- Bifacial TOPCon + perovskite top cell: **32.73% certified**, Voc 1.961 V, 80% initial efficiency retained after 2,000 hours continuous operation.
- Back-contact TOPCon + perovskite: **32.76% certified**, 91% retention after 1,700 hours — now the highest independently certified efficiency for this tandem class.

A companion paper (Nanchang University, same issue) used cesium 4-(diphenylphosphino)benzoate as a doping agent, achieving **26.61% certified** single-junction perovskite cells retaining 95% efficiency after 1,500 h at 85°C/1-sun MPPT — the most demanding published stability data for perovskite to date.

**Where it stands:** All results independently certified by China's National PV Industry Metrology and Testing Center. GBT (private) confirmed the first A-sample all-solid-state EV battery cells (260–500 Wh/kg) passed needle-penetration and thermal-shock tests; GWh-scale production targeted in 2026. JinkoSolar (JKS) has not yet announced commercial production of tandem modules at these efficiencies.

**Why it matters:** The 32%+ tandem milestone matters economically, not just technically. At this efficiency, rooftop and utility-scale solar achieves meaningfully higher energy yield per square meter, reducing balance-of-system costs and land requirements proportionally. The cesium-doping stability result addresses the primary commercial objection to perovskite deployment. The gap between lab record and commercial product remains — but these results are on M10 manufacturing-scale wafers using additive-assisted crystallization compatible with existing tool sets, compressing that gap faster than expected.

**Key players:** JinkoSolar (JKS), Soochow University, Nanchang University, LONGi, First Solar (FSLR) as competitive benchmark.

---

### 2.2 · Energy Storage | Early Commercialization / Lab-to-Market
**Sector:** Storage · **Tickers:** QNTS (quantum — private), GBT (private), CATL (SHE: 300750)

**What it is:** Two parallel storage breakthroughs — one overturning a foundational battery-failure assumption, the other demonstrating the first working quantum battery prototype at room temperature.

**📄 Paper Spotlight — Science, Apr 8, 2026**
University of Houston researchers (Prof. Yan Yao) used custom operando scanning electron microscopy to observe lithium dendrites in real-time during active solid-state battery cycling. Finding: lithium dendrites are **brittle, rigid single-crystal needles**, not soft ductile structures as assumed for 30+ years. The single-crystal core confers high stiffness; a solid-electrolyte interphase coating stabilizes the structure. Dendrites fail via **brittle fracture** — they punch through separators like glass needles, not ductile wires. DFT calculations confirmed elevated migration barriers within dendrite lattice versus isolated Li atoms, explaining structural persistence.

*Implication:* Decades of solid-state battery separator engineering aimed at mechanical stiffness was solving the wrong problem. Dendrite nucleation suppression — through lithium alloy anodes, electrolyte composition, and surface chemistry — is the correct intervention. This reorients the entire solid-state battery R&D roadmap.

**Quantum battery (Light: Science & Applications, Mar 13 — within context window):** CSIRO/RMIT/University of Melbourne team demonstrated the first functional quantum battery prototype: a layered organic device charged wirelessly by laser, operating at room temperature. Critically, charging speed **increases with system size** — opposite to every classical battery. Energy is stored via quantum superposition rather than chemical redox. Still orders of magnitude from commercial scale, but room-temperature operation and functional charge/store/discharge cycle validated.

**Where it stands:** Quantum battery — proof-of-concept only; energy storage duration is the next milestone. Dendrite finding — published in Science, immediately actionable for R&D redirection; no single-company beneficiary yet but narrows optimal solid-state anode strategy. GBT solid-state cells: first A-samples in production, mass GWh capacity + in-vehicle use targeted 2026.

**Why it matters:** The dendrite paper is the rarer category of result: one that invalidates prior work at scale and forces methodological change across an entire field. Its near-term commercial signal is that lithium alloy anodes (rather than pure Li metal) become the preferred path for solid-state batteries — watch for strategic announcements from Toyota, QuantumScape (QS), and Factorial Energy. The quantum battery result is longer-dated but conceptually significant: a storage technology that improves rather than degrades as it scales inverts every economic assumption about battery system design.

**Key players:** University of Houston, CSIRO/RMIT, QuantumScape (QS), Solid Power (SLDP), Toyota (TM), CATL.

---

### 2.3 · Hydrogen | Early Commercialization
**Sector:** Hydrogen · **Tickers:** PLUG, BE, Nel ASA (NEL.OL) — competitive read-through

**What it is:** Two complementary hydrogen breakthroughs remove different cost barriers simultaneously: a platinum-free AEM electrolyzer cathode clearing 1,000 hours of operation, and a redesigned fuel cell geometry producing 75% more power without added cost.

**Where it stands:** Washington University (Prof. Gang Wu) published the platinum-free phosphide heterostructure catalyst in JACS (Apr 17). Cathode outperformed platinum benchmarks in hydrogen adsorption kinetics and ran >1,000 hours at industry-standard current densities — one of the longest published durability runs for any PGM-free AEM cathode. UNSW's lateral bypass fuel cell (100 µm channels, 100 µm micro-ribs) has been patented; scale-up to commercial size in progress.

**Why it matters:** Platinum-group metals represent ~40% of electrolyzer capital cost at scale. Removing that dependency changes the cost floor for green hydrogen production equipment. The UNSW fuel cell result (75% more power from existing geometries) addresses the utilization side — together, these two results improve both CAPEX and energy efficiency of the hydrogen value chain. For companies like Plug Power (PLUG) and Nel ASA (NEL.OL), the competitive signal is that PGM-cost arguments for AEM over PEM electrolyzers gain new technical credibility, potentially accelerating AEM adoption.

KINETIC7's hydrogen-on-demand portable stove (freshwater electrolysis, no stored H₂) entered commercialization with Q3/Q4 2026 production start; independently validated by Imperial College London. Niche but significant for off-grid humanitarian and military markets.

**Key players:** Washington University in St. Louis, UNSW, Plug Power (PLUG), Nel ASA (NEL.OL), ITM Power (ITM.L), KINETIC7 (private).

---

### 2.4 · Nuclear | Policy / Early Deployment
**Sector:** Nuclear (Fission + Fusion) · **Tickers:** XE (IPO), SVAC→General Fusion, NuScale (SMR), Oklo (OKLO)

**What it is:** A convergence of IPO activity, federal funding, and White House policy creating the most active two-week window for nuclear capital markets since the 2000s.

**X-Energy IPO (XE, Nasdaq):** Launched roadshow Apr 15 at $16–19/share, 42.86M shares offered (+ 6.43M greenshoe), leads: JPMorgan, Morgan Stanley, Jefferies, Moelis. Proceeds directed toward Xe-100 pebble-bed SMR commercialization and proprietary TRISO-X fuel. This is the first SMR developer to pursue a standalone Nasdaq IPO.

**General Fusion / SVAC:** Reverse merger with Spring Valley Acquisition Corp. III (NASDAQ: SVAC) announced Jan 21, 2026; transaction advancing. Potential $335M in proceeds, $1B pro-forma valuation. Uses magnetized target fusion approach; ~$2B raised privately prior to SPAC.

**TAE Technologies:** Received $200M of $300M committed from Trump Media merger; S-4 registration pending. Remaining $100M contingent on filing.

**Commonwealth Fusion Systems:** ~70% complete on SPARC demonstration facility west of Boston; targeting scientific breakeven 2027. Long-term power purchase agreement with Google for Virginia commercial plant. Realta Fusion partnership (Apr 2) for HTS magnet supply — multibillion-dollar potential value.

**ARPA-E:** $135M committed over 18 months to fusion acceleration (record agency commitment to fusion). Four focus areas: plasma heating efficiency, advanced fuels, pulsed-power systems, novel plant designs.

**White House NSTM-3 (Apr 14):** National Initiative for American Space Nuclear Power — lunar surface fission reactor by 2030 (≥20 kWe, 3 yr orbit / 5 yr surface lifetime); orbital mid-power reactor by 2031; DOE to assess industrial base within 60 days. First concrete federal step toward space nuclear deployment.

**Genesis Mission (DOE, Apr 14):** $293M RFA for AI-enabled interdisciplinary teams targeting fusion, fission, and geothermal breakthroughs via National Discovery Platform (supercomputing + quantum + AI on federal datasets). Explicitly includes Integrated Blanket and Fuel Cycle Test Stand and Fusion Prototypic Neutron Source as required facilities.

**Why it matters:** Multiple fusion companies accessing public markets in the same 90-day window normalizes fusion as an investable asset class, not merely a research category. X-Energy's IPO provides the first liquid public vehicle for SMR exposure below gigawatt-class conventional nuclear. The White House space nuclear initiative opens an entirely new market — lunar/orbital power — with DoD as anchor customer, bypassing the commercial utility regulatory pathway that has slowed terrestrial nuclear deployment.

**Key players:** X-Energy (XE), General Fusion / SVAC, TAE Technologies, Commonwealth Fusion Systems, NuScale (SMR), Oklo (OKLO), Realta Fusion.

---

### 2.5 · Grid & Infrastructure | Policy / Capital Deployment
**Sector:** Electricity & Grid · **Tickers:** PWR, AMSC, AES (pending acquisition), VST

**What it is:** Simultaneous federal, state, and private capital mobilization for grid modernization — driven by AI data-center load growth that is rewriting demand forecasts faster than infrastructure can respond.

**DOE SPARK Program:** $1.9B competitive funding (IIJA-sourced) for reconductoring, advanced transmission technologies, and large-scale cross-regional upgrades. Concept papers closed Apr 2; full applications due May 20; selections expected August 2026. Eligible: utilities, storage operators, generators, transmission owners, tribes, universities.

**JPMorgan Security and Resiliency Initiative:** $1.5 trillion, 10-year commitment to facilitate/finance/invest in energy security and resiliency. Direct energy project roles (e.g., $5B VoltaGrid financing package) plus $600K to Georgia Cleantech Innovation Hub for workforce/ecosystem development.

**AES Acquisition:** GIP + EQT Infrastructure consortium acquiring AES (NYSE: AES) for ~$33.4B including debt ($15.00/share equity); CalPERS and Qatar Investment Authority as co-investors. Largest energy infrastructure buyout of 2026. Represents private infrastructure capital's conviction that utility-scale clean energy assets are strategic at any price cycle.

**Vehicle-to-Grid:** University of Delaware / Exelon Delmarva pilot confirmed V2G EVs can earn revenue under PJM interconnection rules while providing measurable grid balancing services — regulatory validation that unlocks fleet-scale deployment as virtual power plants.

**AI data center demand context:** DOE/LBNL projects U.S. data center electricity consumption rising from 176 TWh (2023) to 325–580 TWh by 2028. Former Google CEO Schmidt testified 29 GW of new data-center power needed by 2027, 67 GW by 2030. Anthropic estimates a single frontier AI training run will require 5 GW by 2027. This load growth — equivalent to adding a mid-size country's demand — is the primary driver of grid investment urgency.

**Why it matters:** The grid is the rate-limiting factor for nearly every other energy technology in this issue. Fusion, SMRs, renewables, hydrogen electrolyzers, and data centers all depend on grid capacity and interconnection speed. SPARK's emphasis on reconductoring (upgrading conductors on existing rights-of-way) is notable: it is the fastest path to expanding transmission capacity without new land acquisition or multi-year siting fights. Critical Loop's $26M Series A for modular microgrids — sourcing domestic LG Energy Solution Vertech storage — reflects a parallel market developing infrastructure that bypasses interconnection queues entirely for industrial customers.

**Key players:** Quanta Services (PWR), AES (under acquisition), Vistra (VST), VoltaGrid (private), Critical Loop (private), ABB (ABBN.SW), GIP/EQT (private).

---

## 3 · Policy & Regulation

### IRA Tax Credit Cliff: July 4, 2026 Is the Line
The One Big Beautiful Bill Act (P.L. 119-21, signed July 4, 2025) accelerated phaseout of IRA clean-energy credits. What has already expired and what expires next:

**Already expired (Dec 31, 2025):** Residential solar (§25D, 30%); residential energy efficiency (§25C, up to $3,200/yr); residential battery storage; clean vehicle credits §30D, §25E, §45W (expired Sep 30, 2025).

**July 4, 2026 deadline:** Projects that commence construction **on or before this date** remain eligible for full §45Y clean electricity production credit ($0.26/kWh) and §48E investment tax credit (30%) plus a potential 10% domestic content adder. Projects beginning after July 4 face sharply reduced or zero eligibility.

**Market impact:** EIA's February 2026 outlook shows developers plan 11.8 GW of wind additions in 2026 — more than double 2025 — driven almost entirely by the deadline. Solar module procurement has bifurcated: FEOC-compliant modules trade at ~$0.46/W (domestic cells), non-compliant at ~$0.28/W. Developers are paying the compliance premium to lock in credit eligibility.

**FEOC thresholds (2026):** Non-prohibited foreign-entity content cap is 40% for solar, 55% for storage, rising 5 percentage points annually. This is reshaping procurement across the entire solar and storage supply chain.

**Hydrogen hubs preserved:** Despite broad IRA rollbacks, DOE confirmed retention (with potential modification) of ~$5B for five Biden-era hydrogen hubs (Texas, Appalachia, mid-Atlantic, Midwest) and two direct-air-capture projects (Occidental; Climeworks/Heirloom). Energy Secretary Wright stated the agency is "keen to move forward" on the majority of ~2,200 reviewed projects.

### DOE Energy Dominance: Nine Pillars
Energy Secretary Wright articulated nine strategic pillars (Apr 16–17): energy addition, National Lab innovation, LNG exports, consumer affordability, Strategic Petroleum Reserve refill, nuclear stockpile, commercial nuclear power, grid reliability/security, and permitting streamlining. Two operational highlights:
- **Permitting:** June 30, 2025 NEPA procedures update; 47 regulations cut by May 2025; FAST-41 dashboard used to compress critical minerals mining timelines (Antimony Ridge, Nevada: FAST-41 status Apr 7; Muncy Creek, Nevada: permits completed Apr 13 after ~8 months).
- **Critical minerals:** $69M DOF funding opportunity (Apr 7) for extraction/processing/recycling R&D; strategic stockpile program (Project Vault, $12B target); Pax Silica initiative ($250M, 10-nation coalition); USA Rare Earth / Carester (France) partnership for rare-earth separation outside China.

### Strait of Hormuz Disruption — IEA April Oil Market Report
The IEA's April 15 report revised 2026 global oil supply downward by 2.6 mb/d, flipping market balance from a projected 2.46 mb/d surplus to 0.41 mb/d surplus. Straits flows reduced from ~20 mb/d baseline to ~3.8 mb/d as of early April. World oil demand revised down 720,000 b/d (demand destruction partially offsets supply loss). Secretary Wright acknowledged the blockade and stated a Hormuz deal had not yet been concluded; gasoline above $3/gallon likely through 2027. DOE issued RFP for emergency SPR exchange of up to 30M barrels.

### State-Level Grid and Nuclear Policy
- **Texas TANDF:** $350M Advanced Nuclear Development Fund open; applications due May 14. Texas A&M/ZettaJoule and Last Energy (30× 20 MWe PWRs in Haskell County) among likely applicants.
- **Advanced transmission mandates:** 16 states have now adopted some form of advanced transmission technology requirement (dynamic line ratings, topology optimization, advanced conductors). Colorado HB 1081 (Grid Optimization Act) requires utilities to assess grid-enhancing tech. Virginia requires distribution grid utilization data and capacity optimization proposals.

---

## 4 · Watchlist

| Name | Ticker | Sector | Stage | Status |
|---|---|---|---|---|
| X-Energy | XE | Nuclear (SMR) | IPO / Pre-revenue | Roadshow launched Apr 15 at $16–19/share; Nasdaq listing pending. First liquid pure-play SMR public vehicle. Watch pricing and post-IPO float behavior as read-through for advanced nuclear sentiment. |
| General Fusion / Spring Valley Acq. III | SVAC | Fusion | SPAC merger | Definitive agreement Jan 21; advancing toward $335M raise at ~$1B valuation. S-4 filing timeline is key catalyst. |
| QuantumScape | QS | Storage (solid-state) | Pre-commercial | Science dendrite paper (Apr 8) directly reshapes solid-state anode strategy; lithium alloy anode path gains credibility. Watch for QS technology roadmap update. |
| Solid Power | SLDP | Storage (solid-state) | Pre-commercial | Same dendrite read-through as QS. Partnership with BMW and Ford; watch anode material disclosures. |
| JinkoSolar | JKS | Solar | Commercial | 32.76% tandem efficiency on M10 wafers. No tandem commercial module announced yet; watch for production timeline disclosure at earnings. |
| Plug Power | PLUG | Hydrogen | Commercial | Platinum-free AEM catalyst (Washington U/JACS) threatens PGM cost narrative that has weighed on PEM-first companies. Could accelerate AEM adoption — watch technology mix disclosures. |
| AES Corporation | AES | Utility / Grid | Pending acquisition | $33.4B GIP+EQT buyout pending regulatory approval. Signals private infrastructure conviction in clean energy utility assets. |
| Spring Valley Acq. III | SVAC | Fusion | SPAC | Advancing General Fusion merger; watch S-4 filing and PIPE closure. |
| Vistra Energy | VST | Grid / Power | Commercial | AI data-center load growth most directly benefits dispatchable generators in constrained markets. |
| Quanta Services | PWR | Grid infrastructure | Commercial | Primary contractor beneficiary of SPARK reconductoring emphasis and state advanced-transmission mandates. |
| Commonwealth Fusion Systems | — (private) | Fusion | Pre-commercial | SPARC ~70% complete; Google PPA signed for Virginia plant; Realta magnet partnership. IPO/SPAC decision expected post-2027 breakeven milestone. |
| Oklo | OKLO | Nuclear (microreactor) | Pre-commercial | Space nuclear NSTM-3 (Apr 14) and Texas TANDF expand addressable market for microreactor designs. Watch DOD contract activity. |

### Weekly Central Bank Monetary Policy Briefing
_monetary-policy · Apr 13 – Apr 20, 2026_

# Weekly Central Bank Monetary Policy Briefing
**Coverage Period: Apr 13 – Apr 20, 2026**

---

> **Editorial Note on Sources:** The research material provided for this briefing period was empty — no Perplexity Deep Research output was supplied. In strict adherence to the source discipline of this briefing (official central bank sources only; no wire services, no analyst commentary, no market pricing), **no factual claims about specific decisions, speeches, or publications can be asserted for this coverage window.** All per-bank sections reflect this constraint transparently. Consumers of this briefing should retrieve the relevant official releases directly via the Source Log URLs below.

---

## Quick-Reference Table — All 13 Central Banks

| # | Central Bank | Policy Rate (as of last confirmed decision) | Rate Action This Period | Forward Guidance Signal | Material Activity This Period |
|---|---|---|---|---|---|
| 1 | Federal Reserve (Fed) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 2 | People's Bank of China (PBoC) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 3 | European Central Bank (ECB) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 4 | Bank of Japan (BoJ) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 5 | Reserve Bank of India (RBI) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 6 | Bank of England (BoE) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 7 | Bank of Canada (BoC) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 8 | Banco Central do Brasil (BCB) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 9 | Bank of Korea (BoK) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 10 | Reserve Bank of Australia (RBA) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 11 | Central Bank of the Republic of Türkiye (CBRT) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 12 | Swiss National Bank (SNB) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |
| 13 | South African Reserve Bank (SARB) | — | Unconfirmed | Unconfirmed | Not verifiable from supplied data |

*All rate figures and decisions require direct verification against official press releases for the Apr 13–20, 2026 window.*

---

## Cross-Bank Analysis

### Convergence / Divergence Assessment

**Data Constraint:** Because no research material was provided and this briefing's source discipline prohibits assertion of facts not drawn from official central bank sources, a substantive cross-bank convergence/divergence analysis cannot be produced for this period without risk of fabrication.

**Structural framework for analysts to apply once official data are retrieved:**

- **Rate cycle positioning:** Assess whether the Apr 13–20 window produced any coordinated signals among G7 central banks (Fed, ECB, BoE, BoJ, BoC, SNB) versus continued divergence driven by differing inflation trajectories.
- **EM vs. DM divergence:** Monitor whether BCB, CBRT, RBI, BoK, RBA, and SARB are in materially different phases of their respective cycles relative to DM peers.
- **PBoC stance:** The PBoC's use of quantity-based tools (RRR, MLF, LPR) rather than a single policy rate requires separate treatment from the rate-based convergence analysis.
- **Forward guidance language:** Any shift from data-dependent to more prescriptive language across multiple banks in a single week would constitute a notable convergence signal.
- **Balance sheet operations:** Coordinated changes to QT or QE programs across major banks would merit flagging as a structural theme.

**Recommendation:** Retrieve the official press releases and speeches listed in the Source Log below, then reapply this framework to populate a complete cross-bank analysis.

---

## Per-Bank Sections

---

### 1. Federal Reserve (Fed)
**Official home:** https://www.federalreserve.gov

#### 1.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. The FOMC's next scheduled meeting dates and any inter-meeting communications should be verified at https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm

#### 1.2 Open Market Operations
Nothing material confirmed. SOMA operations and repo/reverse-repo data available daily at https://www.newyorkfed.org/markets/desk-operations

#### 1.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 1.4 Published Research
Nothing material confirmed. FEDS Notes and working papers searchable at https://www.federalreserve.gov/econres.htm

#### 1.5 Official Speeches & Testimony
Nothing material confirmed. Speech calendar at https://www.federalreserve.gov/newsevents/speeches.htm

---

### 2. People's Bank of China (PBoC)
**Official home:** http://www.pbc.gov.cn/en/3688110/index.html

#### 2.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. Key rates to verify: 1-year LPR, 5-year LPR, 7-day reverse repo rate, 1-year MLF rate.

#### 2.2 Open Market Operations
Nothing material confirmed. Daily OMO announcements at http://www.pbc.gov.cn/en/3688110/3688172/index.html

#### 2.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 2.4 Published Research
Nothing material confirmed.

#### 2.5 Official Speeches & Testimony
Nothing material confirmed.

---

### 3. European Central Bank (ECB)
**Official home:** https://www.ecb.europa.eu

#### 3.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. Governing Council meeting schedule at https://www.ecb.europa.eu/press/govcdec/mopo/html/index.en.html

#### 3.2 Open Market Operations
Nothing material confirmed. APP and PEPP data at https://www.ecb.europa.eu/mopo/implement/omo/html/index.en.html

#### 3.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 3.4 Published Research
Nothing material confirmed. ECB Working Papers at https://www.ecb.europa.eu/pub/research/working-papers/html/index.en.html

#### 3.5 Official Speeches & Testimony
Nothing material confirmed. Speech database at https://www.ecb.europa.eu/press/key/html/index.en.html

---

### 4. Bank of Japan (BoJ)
**Official home:** https://www.boj.or.jp/en/

#### 4.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPM schedule at https://www.boj.or.jp/en/mopo/mpmdeci/index.htm

#### 4.2 Open Market Operations
Nothing material confirmed. JGB purchase operations at https://www.boj.or.jp/en/statistics/boj/other/mei/index.htm

#### 4.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 4.4 Published Research
Nothing material confirmed. Working papers at https://www.boj.or.jp/en/research/wps_rev/wps/index.htm

#### 4.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.boj.or.jp/en/announcements/press/koen/index.htm

---

### 5. Reserve Bank of India (RBI)
**Official home:** https://www.rbi.org.in

#### 5.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPC decisions at https://www.rbi.org.in/Scripts/BS_PressReleaseDisplay.aspx

#### 5.2 Open Market Operations
Nothing material confirmed. OMO and LAF data at https://www.rbi.org.in/Scripts/BS_ViewBulletin.aspx

#### 5.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 5.4 Published Research
Nothing material confirmed. RBI Working Papers at https://www.rbi.org.in/Scripts/bs_viewcontent.aspx?Id=2009

#### 5.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.rbi.org.in/Scripts/BS_SpeechesView.aspx

---

### 6. Bank of England (BoE)
**Official home:** https://www.bankofengland.co.uk

#### 6.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPC decisions at https://www.bankofengland.co.uk/monetary-policy/monetary-policy-committee

#### 6.2 Open Market Operations
Nothing material confirmed. Asset Purchase Facility data at https://www.bankofengland.co.uk/markets/asset-purchase-facility

#### 6.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 6.4 Published Research
Nothing material confirmed. Staff Working Papers at https://www.bankofengland.co.uk/working-paper

#### 6.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.bankofengland.co.uk/news/speeches

---

### 7. Bank of Canada (BoC)
**Official home:** https://www.bankofcanada.ca

#### 7.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. Governing Council decisions at https://www.bankofcanada.ca/core-functions/monetary-policy/key-interest-rate/

#### 7.2 Open Market Operations
Nothing material confirmed. Market operations at https://www.bankofcanada.ca/core-functions/financial-system/market-operations-liquidity/

#### 7.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 7.4 Published Research
Nothing material confirmed. Staff Working Papers at https://www.bankofcanada.ca/research/staff-working-papers/

#### 7.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.bankofcanada.ca/publications/speeches/

---

### 8. Banco Central do Brasil (BCB)
**Official home:** https://www.bcb.gov.br/en

#### 8.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. COPOM decisions at https://www.bcb.gov.br/en/monetarypolicy/copom

#### 8.2 Open Market Operations
Nothing material confirmed. Open market operations at https://www.bcb.gov.br/en/monetarypolicy/openmarket

#### 8.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 8.4 Published Research
Nothing material confirmed. Working Papers at https://www.bcb.gov.br/en/publications/workingpapers

#### 8.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.bcb.gov.br/en/pressandcommunication/speeches

---

### 9. Bank of Korea (BoK)
**Official home:** https://www.bok.or.kr/eng/main/main.do

#### 9.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPB decisions at https://www.bok.or.kr/eng/bbs/P0000559/list.do?menuNo=600041

#### 9.2 Open Market Operations
Nothing material confirmed.

#### 9.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 9.4 Published Research
Nothing material confirmed. Working Papers at https://www.bok.or.kr/eng/bbs/P0000537/list.do?menuNo=600014

#### 9.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.bok.or.kr/eng/bbs/P0000560/list.do?menuNo=600042

---

### 10. Reserve Bank of Australia (RBA)
**Official home:** https://www.rba.gov.au

#### 10.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. Board decisions at https://www.rba.gov.au/monetary-policy/rba-board-minutes/

#### 10.2 Open Market Operations
Nothing material confirmed. Market operations at https://www.rba.gov.au/mkt-operations/

#### 10.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 10.4 Published Research
Nothing material confirmed. Research Discussion Papers at https://www.rba.gov.au/publications/rdp/

#### 10.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.rba.gov.au/speeches/

---

### 11. Central Bank of the Republic of Türkiye (CBRT)
**Official home:** https://www.tcmb.gov.tr/wps/wcm/connect/en/tcmb+en

#### 11.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPC decisions at https://www.tcmb.gov.tr/wps/wcm/connect/en/tcmb+en/main+menu/announcements/press+releases/monetary+and+exchange+rate+policy

#### 11.2 Open Market Operations
Nothing material confirmed.

#### 11.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 11.4 Published Research
Nothing material confirmed. Working Papers at https://www.tcmb.gov.tr/wps/wcm/connect/en/tcmb+en/main+menu/publications/research/working+paperss

#### 11.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.tcmb.gov.tr/wps/wcm/connect/en/tcmb+en/main+menu/announcements/speeches

---

### 12. Swiss National Bank (SNB)
**Official home:** https://www.snb.ch/en/

#### 12.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. Monetary policy decisions at https://www.snb.ch/en/monetary-policy/monetary-policy-decisions

#### 12.2 Open Market Operations
Nothing material confirmed. Repo operations at https://www.snb.ch/en/monetary-policy/monetary-policy-instruments

#### 12.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 12.4 Published Research
Nothing material confirmed. Working Papers at https://www.snb.ch/en/publications/research/working-papers

#### 12.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.snb.ch/en/publications/speeches

---

### 13. South African Reserve Bank (SARB)
**Official home:** https://www.resbank.co.za/en/home

#### 13.1 Rate Decisions & Forward Guidance
No confirmed decision data available for Apr 13–20, 2026. MPC decisions at https://www.resbank.co.za/en/home/what-we-do/monetary-policy/mpc-statements

#### 13.2 Open Market Operations
Nothing material confirmed.

#### 13.3 Regulatory & Supervisory
Nothing material confirmed for this period.

#### 13.4 Published Research
Nothing material confirmed. Working Papers at https://www.resbank.co.za/en/home/publications/working-papers

#### 13.5 Official Speeches & Testimony
Nothing material confirmed. Speeches at https://www.resbank.co.za/en/home/publications/speeches

---

## Source Log

> All URLs below are official central bank sources. No wire services, no analyst commentary, and no market pricing data are referenced. Because no research material was supplied, these are the canonical official portals from which Apr 13–20, 2026 releases should be retrieved and verified.

| # | Institution | Primary Portal | Key Policy Release Path |
|---|---|---|---|
| 1 | Federal Reserve | https://www.federalreserve.gov | /monetarypolicy/fomccalendars.htm |
| 1a | Federal Reserve Bank of New York | https://www.newyorkfed.org | /markets/desk-operations |
| 2 | People's Bank of China | http://www.pbc.gov.cn/en/3688110/index.html | /en/3688110/3688172/index.html |
| 3 | European Central Bank | https://www.ecb.europa.eu | /press/govcdec/mopo/html/index.en.html |
| 4 | Bank of Japan | https://www.boj.or.jp/en/ | /en/mopo/mpmdeci/index.htm |
| 5 | Reserve Bank of India | https://www.rbi.org.in | /Scripts/BS_PressReleaseDisplay.aspx |
| 6 | Bank of England | https://www.bankofengland.co.uk | /monetary-policy/monetary-policy-committee |
| 7 | Bank of Canada | https://www.bankofcanada.ca | /core-functions/monetary-policy/key-interest-rate/ |
| 8 | Banco Central do Brasil | https://www.bcb.gov.br/en | /en/monetarypolicy/copom |
| 9 | Bank of Korea | https://www.bok.or.kr/eng/main/main.do | /eng/bbs/P0000559/list.do?menuNo=600041 |
| 10 | Reserve Bank of Australia | https://www.rba.gov.au | /monetary-policy/rba-board-minutes/ |
| 11 | CBRT | https://www.tcmb.gov.tr/wps/wcm/connect/en/tcmb+en | /main+menu/announcements/press+releases |
| 12 | Swiss National Bank | https://www.snb.ch/en/ | /en/monetary-policy/monetary-policy-decisions |
| 13 | South African Reserve Bank | https://www.resbank.co.za/en/home | /en/home/what-we-do/monetary-policy/mpc-statements |

---
*Briefing prepared under strict source discipline. All factual gaps reflect the absence of supplied research material, not editorial omission. Populate from official releases before distribution.*