読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
すべて技術製品投資株式市場その他

読書

2026-10-22168 投稿

更新

Alpha Exchange Interview: Harley Bassman on Long-Term Treasury Yields and the Trust Story

重大
2026-10-06 16:00 GMT+8

Alpha Exchange published a notable macro-rates interview on October 6, 2026, with guest Harley Bassman. Known as "The Convexity Maven," Bassman created the widely recognized MOVE Index during his tenure at Merrill Lynch and has held senior roles at PIMCO and Simplify Asset Management. This episode focuses on a central market contradiction: the persistent rise in long-term US Treasury yields.

According to the official episode description, Bassman offers a perspective distinct from mainstream narratives: he views the increase in long-term yields not merely as an inflation issue, but as a "trust story." With expanding US government fiscal deficits, growing interest expenses, and a massive supply of Treasuries coming to market, investors are demanding greater compensation to hold long-duration government debt. This supply-demand pressure is further exacerbated by substantial borrowing from AI hyperscalers and a shifting global buyer base.

Beyond the macro narrative, the content includes specific technical insights. Bassman revisits the origins of the MOVE Index, which he created in 1994 using data extending back to 1988. He explains what a MOVE reading above 100 actually signifies and compares current rate volatility levels with history, discussing why seemingly extreme levels can sometimes represent a return to an older market regime. For professional readers focused on fixed-income strategies and risk management, this episode offers a chance to examine current bond market structural changes through the lens of a seasoned practitioner.

ソース:alphaexchange.simplecast.com

リサーチ

CFA Institute Podcast Selection: Cheryll-Ann Wilson Discusses Valuation Opacity and Liquidity Mismatch Risks in Private Credit's Retail Shift

重大
2026-10-15 11:00 GMT+8

The flagship podcast of CFA Institute, 'Enterprising Investor', released a special interview on October 15, 2026, featuring Cheryll-Ann Wilson (PhD, CFA), Senior Research Affiliate at CFA Institute, hosted by Kevin Max. The core topic of this episode is the structural risks triggered by the penetration of private credit into retail investment portfolios after expanding to a $3.5 trillion asset scale.

According to the episode description, Wilson will focus on analyzing four key risk dimensions: opaque valuations, semiliquid fund structures, weaker loan covenants, and hidden leverage. These characteristics may expose investors to unexpected losses as private credit shifts from being dominated by institutional investors to broader retail participation.

For wealth advisers and portfolio managers, this episode offers a specific discussion framework: when private credit products enter the retail sector through digital platforms or nontraded business development companies (BDCs), how can investors identify the mismatch between underlying asset liquidity and fund redemption promises while pursuing higher yields? The program aims to help professional readers understand the tension between regulatory lag and market innovation, and why traditional due diligence standards need adjustment for retailized scenarios.

ソース:share.transistor.fm

リサーチ

Meritech Capital: With Identical Rule of 40 Scores, Growth-Focused Software Companies Command Valuations Up to 2.7 Times Higher Than Profitability-Focused Peers

重大
2026-10-13 09:33 GMT+8

In its Software Pulse report published on October 9, 2026, venture capital firm Meritech Capital analyzed the composition of the Rule of 40 across more than 100 public software companies, uncovering a result with direct implications for valuation: companies with similar Rule of 40 scores can see their market valuation multiples differ by up to 2.7 times depending on how those scores are achieved. Specifically, growth-driven companies (characterized by high revenue growth and low free cash flow margins) traded at a median multiple of 10.9x implied ARR, whereas profitability-driven companies (low growth, high free cash flow margins) traded at only 4.1x.

The Rule of 40 is the most frequently cited composite health metric in the software industry, defined as the sum of next-twelve-months (NTM) revenue growth and NTM free cash flow margin; a total score of 40 is generally considered healthy.

Meritech further quantified this disparity through a two-factor regression analysis. In a joint regression of NTM revenue growth and NTM free cash flow margin against ARR multiples, growth exhibited a correlation strength 4.5 times greater than that of profit margins—meaning a one percentage point increase in growth has the same impact on valuation as a 4.5 percentage point improvement in free cash flow margin. This aligns with the Rule of 40 composition findings: the market explicitly rewards growth over profitability when total scores are equal. Meritech also noted that public markets are concerned about the impact of AI on horizontal SaaS vendors; while most companies’ growth rates have not yet recovered, only a select few have accelerated growth through AI-enabled features.

ソース:meritech.substack.com

リサーチ

Alpha Exchange Interview: Wellington Fixed Income Manager Brij Khurana on AI Capex Impact on Bond Markets

トピック · AI基建支出前景重大
2026-10-01 16:00 GMT+8

The Alpha Exchange podcast released an interview with Brij Khurana, Fixed Income Portfolio Manager at Wellington Management, on October 1, 2026. According to the official show notes, the conversation focuses on the large-scale repricing underway in fixed income markets, exploring monetary policy, inflation dynamics, and opportunities across global bond markets.

The show notes mention that Khurana poses a provocative question: what if interest rates have become more effective at influencing financial markets than the real economy? The notes state he argues that the traditional relationship between rates and business investment has weakened considerably, suggesting ultra-low and negative real rates encouraged financialization—such as household leverage and corporate stock buybacks—rather than productive investment.

Additionally, the notes indicate Khurana's view that today's remaining price pressures are increasingly connected to wealth rather than wages. With significant household wealth tied to equities, asset-price appreciation may support spending in housing, healthcare, and travel even as labor-market demand cools.

For AI industry investors concerned about macro spillovers, the show notes specifically highlight a discussion on the US Federal Reserve (Fed) and the repricing of developed-market yield curves. Khurana is described as explaining how longer-dated yields incorporate debates around nominal growth, term premium, and the economic consequences of the AI investment boom. The notes explicitly list an examination of whether extraordinary capital spending and borrowing by hyperscalers are contributing to higher bond yields, changing the traditional stock-bond relationship, and potentially crowding out other borrowers.

This episode offers a relevant macro framework for readers seeking to understand the financing costs of AI infrastructure and their potential impact on global fixed income markets. Note that this recommendation is based on the publisher's listed topics and does not include unverified specific conclusions or predictions from the guest.

ソース:alphaexchange.simplecast.com

リサーチ

Hugging Face Releases Open-Source DNA Chat Model Carbon-X, Enabling Local Execution for Data Privacy

重大
2026-10-22 18:00 GMT+8

AI platform Hugging Face released Carbon-X, a new open-source biology model on Thursday that enables users to 'chat' with DNA sequences using natural language. The model is designed to run locally on hardware such as laptops. Developed in collaboration with researchers at Zhongguancun Academy in China and the University of Naples Federico II in Italy, Carbon-X aims to help researchers understand the effects of DNA changes on gene activity and health.

Lewis Tunstall, a senior research scientist at Hugging Face, stated that the tool unlocks a new capability in language models: the ability to converse with DNA. Unlike previous tools that rely on complex programming or numeric predictions (such as Google DeepMind's AlphaGenome), Carbon-X integrates DNA sequences with natural language. This allows even non-technical users to ask questions about whether a genetic mutation is benign or disease-causing, potentially aiding in the design of new drugs and genetic therapies.

To address privacy concerns regarding sensitive patient data, the team intentionally designed the model with 14 billion parameters so it can operate entirely on local hardware, ensuring no data is sent to Hugging Face or other cloud services. Currently, the model processes DNA segments of approximately 50,000 base pairs; users cannot upload full human genomes. Training data included sequences from mammals, vertebrates, invertebrates, plants, fungi, and protozoa, but excluded viruses to prevent misuse.

Despite the innovative interface, developers caution that Carbon-X has not yet been validated through actual wet lab experiments, meaning its predictive accuracy remains unproven. This release marks a shift from pure computational prediction to interactive interpretation tools, though clinical application still requires rigorous independent verification.

ソース:WIRED

リサーチ

SEMICON West 2026: AI Computing Bottlenecks Expand from Transistor Scaling to System-Level Integration

構造的
2026-10-22 15:05 GMT+8

When Patrick Vandenameele, CEO of imec, spoke about AI data centers at SEMICON West 2026, he posed a question that prompted semiconductor professionals to re-examine their work: These data centers are true engineering marvels, but they do not represent scaling in the semiconductor sense—they look more like multiplication.

This distinction hits at the core constraint of current AI computing growth. Transistor scaling continues, but the bottlenecks facing the industry have expanded: packaging, photonics, power delivery, and test are converging into a coupled optimization problem, where improving any one of them may push the bottleneck elsewhere. More compute increases pressure on memory bandwidth and power delivery; moving memory closer to compute units reduces data movement but concentrates heat; disaggregating large dies into smaller chips can improve yield and allow different functions to use different process nodes, but creates more interfaces that require connection, power, cooling, and test.

Lalitha Immaneni, Vice President of Semiconductor R&D at Intel Foundry, summarized this shift by stating that packaging is no longer just where semiconductor innovation ends, but where system innovation begins. Nitin Deshpande, Vice President of Advanced Packaging at Marvell, was more direct—packaging is becoming a system where compute, memory, interconnects, power, and cooling all converge.

This system-level coupling has specific implications for manufacturing economics. When certain silicon-HBM combinations are worth up to $30,000, any defect in a single layer of packaging wastes all the value invested in the surrounding components. The Known Good methodology described by Yin Chang, Executive Vice President of ASE, is a response to this: each layer is confirmed qualified before the most expensive components are attached. However, assembly itself creates new uncertainties—the interfaces created by bonding did not exist when individual components were inspected, and heat, pressure, and material interactions can alter the state of the assembly, with some defects only discoverable at the next process step.

For investors in the AI industry, this means that constraints on computing power expansion are no longer concentrated at a single process node or a specific packaging technology, but are distributed across the intersections of multiple engineering fields. Companies capable of solving cross-domain collaborative optimization—whether in design tools, packaging processes, or test infrastructure—will face a competitive landscape that is wider but harder than that of a single technological breakthrough. imec’s judgment of multiplication implies another layer of risk: if the complexity of system integration grows faster than its economic improvement, the expansion of AI infrastructure may face a ceiling defined by cost rather than density.

It should be noted that these judgments come from self-narratives by industry leaders at conferences. The fact that multiple executives pointed to the same issue indicates that the bottleneck is real, but it remains an industry perspective rather than independent measurement. Specific markers to watch in the coming months include: which engineering decisions agentic AI gains autonomous authority over without human approval; whether the pressure on inspection from heterogeneous integration translates into quantifiable yield improvements; and whether co-packaged optics can move from demonstrations to repeatable manufacturing.

ソース:semiengineering.com

リサーチ

Tesla Q3 Net Income Plunges 43% as AI R&D Costs Surge; Musk Admits Optimus Mass Production Delayed

トピック · 特斯拉Q3财报重大
2026-10-22 17:03 GMT+8

U.S. electric vehicle maker Tesla reported its third-quarter 2026 financial results, showing a roughly 43% year-over-year decline in GAAP net income attributable to common stockholders, falling to $787 million. While revenue remained stable at $28.05 billion, a sharp increase in operating expenses significantly compressed margins. Research and development costs surged to $2.76 billion from $1.63 billion in the same period last year, an increase of over 69%. Tesla attributed this rise to spending on "Super Intelligence" projects, other development initiatives, and higher stock-based compensation.

CEO Elon Musk used the analyst call to highlight the company's humanoid robot, Optimus. He claimed that early production has begun at the Fremont factory, with most equipment installed to support a nominal capacity of one million robots per year. However, Musk admitted that reaching the target rate of 20,000 units per week would "probably take at least a year, maybe a little longer." He cited changing hand designs, supply chain construction, and robot reliability as key bottlenecks. Tesla did not disclose current weekly production numbers, external customers, or unit costs for Optimus.

Operating margin fell from 5.8% to 2.6%. Non-operating factors distorted the bottom line: unrealized gains of $288 million from Bitcoin holdings were offset by $142 million in unrealized losses from SpaceX equity. Excluding these items and stock-based compensation, Tesla reported an adjusted profit of $1.70 billion, a 4% decline. Free cash flow halved to $3.3 billion over the past twelve months, while capital expenditures rose from $8.9 billion to $15.4 billion, reflecting heavy investment in autonomous driving and robotics without immediate returns.

ソース:Autos

リサーチ

CFA Institute Podcast 'Enterprising Investor': Financial Historian Mark J. Higgins Discusses Speculative Supply Chain Risks in Private Credit

重大
2026-10-01 11:00 GMT+8

CFA Institute's flagship podcast, 'Enterprising Investor', released a new episode on October 1, 2026, titled 'Mark J. Higgins: Are Speculative Supply Chains Fueling Private Credit?'. The guest is Mark J. Higgins, CFA, CFP®, a financial historian, investment adviser, and author of 'Investing in U.S. Financial History: Understanding the Past to Forecast the Future'. The host is Kevin Max, editor of Enterprising Investor.

According to the publisher's show notes, Higgins presents a core argument in this episode: financial bubbles may be driven less by irrational investors than by rational participants whose incentives increasingly align around keeping capital flowing. The discussion explores how these 'speculative supply chains' form, why risk becomes harder to identify as these chains mature, and whether patterns similar to those that helped fuel the Global Financial Crisis are now emerging in private equity and private credit markets.

For professional readers interested in alternative asset risks and financial history cycles, this episode offers a perspective on the current private credit boom through the lens of incentive structures. While no full transcript is provided, the guest's authority in financial history and CFA Institute's standing as a definitive program for the investment management industry make this discussion worth listening to for understanding potential systemic risk structures.

ソース:share.transistor.fm

リサーチ

Goldman Sachs Chief Credit Strategist Amanda Lynam on AI Capex Debt Financing

トピック · AI基建债务风险重大
2026-09-23 17:00 GMT+8

The Alpha Exchange podcast released a new episode on September 23, 2026, featuring Amanda Lynam, who currently serves as the Chief Credit Strategist in Global Investment Research at Goldman Sachs. This marks her return to the show nearly three years after their first conversation, set against the backdrop of an unprecedented capital expenditure cycle in the artificial intelligence industry.

According to the publisher's show notes, the core question addressed is how the corporate credit market will finance this wave of AI buildout and what the increasing supply of debt implies for investors. Lynam begins by assessing the broader credit landscape, noting that the market is supported by resilient economic growth, generally solid corporate fundamentals, and powerful demand from yield-oriented investors, including insurers, pensions, and foreign buyers. These forces have kept periods of spread widening relatively brief, even as issuance has accelerated.

The discussion then turns to the scale of the AI infrastructure buildout. The show notes indicate that Lynam walks through Goldman Sachs' estimates for hyperscaler capital expenditures and debt issuance, along with the financing needs of data centers, chips, and other businesses supporting the AI ecosystem. They discuss why companies are raising debt before an immediate funding gap appears, how much issuance the investment-grade market can absorb, and why private markets may take on a larger role as the cycle progresses.

Importantly, the notes state that Lynam sees little evidence so far that AI borrowing is crowding out other corporate issuers. Instead, investors appear increasingly attentive to their total exposure to the theme across equities, bonds, and private assets. This creates a renewed role for sectors such as banks, energy, healthcare, and food and beverage as sources of diversification. Finally, the episode explores risks that could challenge today's supportive credit backdrop, particularly an increase in rates volatility that weakens yield-based demand.

For readers interested in AI industry investment, this episode offers a perspective on the AI expansion's funding chain from the fixed income and credit market angle. While currently based only on the publisher's show notes, given Lynam's professional role overseeing credit strategy at Goldman Sachs, her views on AI debt financing scale and investor exposure management are worth listening to.

ソース:alphaexchange.simplecast.com

リサーチ

LG Electronics Researcher Lee Cheol-hee: AI-Defined Vehicles Cannot Let Models Control Everything

トピック · QNX物理AI安全クイックテイク
2026-10-22 15:22 GMT+8

At the BlackBerry QNX Technology Forum Korea 2026 held in Seoul, South Korea, LG Electronics researcher Lee Cheol-hee presented a specific assessment: vehicle software is transitioning from 'Software-Defined Vehicles' (SDVs) to 'AI-Defined Vehicles' (AIDVs). In the past, manufacturers pre-developed fixed function combinations such as sleep mode and comfort mode, pushing them to users via over-the-air (OTA) updates. Now, passengers can state their intent using natural language while driving, and the in-vehicle AI interprets this intent in real-time to combine the necessary functions.

Lee Cheol-hee provided an example: if a driver says, "The child behind me is asleep, don't wake him," the AI must independently determine whether to adjust lighting and temperature, lower the volume, and mute navigation prompts. This is not about specifying individual parameters but conveying a purpose.

The problem arises when the AI is responsible for both judgment and execution. To achieve the goal of "creating a quiet environment," the AI might inadvertently turn off safety warning sounds or emergency braking alerts. Lee Cheol-hee warned that AI models should not be allowed to control all vehicle actions because the vehicle interior is a "mixed-criticality" environment where safety-critical systems coexist with general convenience features.

His proposed solution involves having LG's Piccolo platform handle flexible upper-level judgments, while Canada-based BlackBerry's QNX real-time operating system handles deterministic lower-level execution. QNX is designed to ensure commands complete within a fixed time frame, guaranteeing that safety functions receive priority computing power even when convenience features heavily consume resources.

Lee Cheol-hee emphasized that this is not merely a matter of assigning priority numbers. Each workload requires fine-grained handling regarding how much resources can be reduced, which tasks can be delayed, and what backup measures are needed in case of failure. He explicitly stated that LG Electronics is currently exploring this linkage plan but has not yet announced its implementation.

ソース:zdnet.co.kr

リサーチ

AI Proxy Library LiteLLM Hit by Supply Chain Attack; Malicious v1.107.0 Steals Cloud and Model API Keys

トピック · LiteLLM供应链安全重大
2026-10-21 21:44 GMT+8

Security firm Wiz Research disclosed on October 21, 2026, that the popular Python LLM proxy library LiteLLM was targeted by a supply chain attack. Attackers published a malicious version, v1.107.0, which remained available on PyPI for approximately two hours before being quarantined. Simultaneously, related image tags on GHCR and Docker Hub (including latest, main-latest, etc.) were trojanized until they were removed around 13:53 UTC. The most recent safe version is v1.106.2.

According to analyses by Wiz and Datadog Security Labs, the malicious package extracts three files (cook.pth, cook.py, and sitecustomize.py) into the site-packages root directory upon installation. These files are designed to execute automatically whenever a Python interpreter starts, triggering even if LiteLLM is not imported. Specifically, cook.py contains a credential stealer named privcollect, which harvests sensitive data across 16 modules targeting browser credentials from Chrome/Firefox, AWS/GCP/Azure cloud credentials, SSH/GPG keys, and API keys for LLM providers like OpenAI and Anthropic stored in environment variables or config files. It also directly queries the LiteLLM Proxy's PostgreSQL database to extract upstream provider credentials.

The harvested data is encrypted with AES-256-CBC and exfiltrated via HTTP POST requests to the attacker-controlled lookalike domain www.litellm[.]dev/healthcheck ↗. The sitecustomize.py file includes an hourly backdoor mechanism checking a GitHub dead drop to download second-stage RAT payloads, although no active payload was observed in the dead drop at the time of analysis. Wiz noted that this dual-technique approach using both .pth files and sitecustomize.py appears to be the first documented case in the wild.

LiteLLM developer BerriAI stated that attackers used stolen GitHub CLI OAuth tokens from three engineers' accounts to trigger the legitimate build pipeline against a malicious fork repository (Berri-AI/litellm, noting the hyphen difference). Datadog added that this marks the second supply chain compromise of LiteLLM in 2026, following earlier incidents affecting versions 1.82.7 and 1.82.8 linked to TeamPCP activity, though no connection between the two compromises has been established yet.

Affected users should immediately manually remove the cook.pth, cook.py, and sitecustomize.py files from the site-packages directory, revert to v1.106.2, and rotate all potentially exposed cloud, AI, and browser credentials from a clean machine.

ソース:wiz.io

リサーチ

Meta's AI Assistant Muse Hits 10M US Downloads in 40 Days, Outpacing Competitors in Retention

トピック · Meta助手Muse重大
2026-10-21 15:21 GMT+8

According to reports compiled by Zhidongxi from CNBC and other US media, Meta's AI assistant Muse has seen rapid user growth since its launch on September 8, 2026. Data from market intelligence firm Sensor Tower shows that as of October 21, Muse had reached 10.9 million global downloads, with 10.2 million coming from the US market. This makes Muse one of the few mobile apps to exceed 10 million US downloads within 40 days of release, a milestone previously achieved by only 11 other apps, including Threads.

Compared to major competitors, Muse's early performance in the US market is significantly stronger. Sensor Tower noted that ChatGPT, Grok, and Claude recorded 3.7 million, 500,000, and 200,000 US downloads respectively during their first 40 days. Although ChatGPT still leads globally with approximately 15 million downloads over the same period, it is important to note that Muse is currently available only in the US and Canada, while ChatGPT covers markets like Japan and Brazil. Additionally, ChatGPT's US downloads dropped by 25% in the past 30 days compared to the previous cycle, highlighting a shift in momentum between the two in the US market.

Beyond downloads, Muse also demonstrates strong user stickiness. Data indicates that Muse's 7-day retention rate in the US is 30%, and its 14-day retention rate is 23%, both exceeding the average levels of competitors such as ChatGPT, Claude, and Grok (19% for 7-day and 16% for 14-day). On October 16, Muse's daily active users (DAU) reached 2.5 million for the first time, with a 15% week-over-week increase in DAU over the past seven days. Since its launch, Muse has consistently held the number one spot on the US App Store free charts.

Muse's rapid rise is directly linked to Meta's powerful promotional capabilities. Sensor Tower data reveals that Meta allocated 39% of its total ad spend to promoting Muse in October, up from 23% in September. This intense advertising push caused Muse to jump from rank 73 to rank 21 in US brand ad exposure rankings between September and October. Meta leveraged its social product ecosystem, including Facebook and Instagram, to drive traffic to Muse, concentrating resources to bring it to mass-market attention.

Wall Street remains optimistic about Muse's future revenue potential. Deutsche Bank analyst Benjamin Black projects that by 2030, Muse could contribute 8% of Meta's revenue, equivalent to approximately $36 billion; even under conservative forecasts, it could generate $2.4 billion in sales. Black argues that Meta possesses unique advantages in combining mobile ecosystems, seamless interfaces, and free features. However, whether the initial high retention stems from novelty and ad-driven acquisition or reflects long-term utility remains to be seen once promotional intensity subsides.

ソース:zhidx.com

リサーチ
次へ →
ログイン / 新規登録
読書リサーチレーダー投資フレームワーク