LesenForschungRadarAnlageframework
Anmelden / Registrieren
AlleTechnikProdukteInvestierenBörseSonstiges

Lesen

2026-10-2453 Beiträge

Aktualisieren

Goldman Sachs Trading Head Rich Privorotsky on AI Investment: How Rising Capital Costs and Spending Glut Lower Risk Appetite

Thema · AI资本支出与回报争议Wesentlich
2026-10-23 21:28 GMT+8

Rich Privorotsky, Head of European Equities Trading within Goldman Sachs FICC and Equities, discussed the market's scrutiny of the crowded AI trade in an episode of The Markets podcast. He noted that investors are reassessing technology infrastructure spending plans while factoring in rising capital costs.

Privorotsky highlighted that third-quarter earnings from mega-cap tech companies will be critical. Investors are not only watching hyperscaler investment plans but also looking for evidence that this spending is translating into revenue growth. He further described a "seismic shift" from a savings glut to a spending glut, which, along with high energy prices, is lowering investor risk appetite.

Recorded on October 22, 2026, this resource selection aims to provide readers with an entry point to understand institutional perspectives on shifting AI investment logic. Specific arguments and data require listening to the full audio.

Quellen:youtube.com

Forschung

Former Energy Options Market Maker & Compute Exchange CEO Carmen Li: Why Buying GPUs Should Work Like Trading Stocks

Thema · GPU算力交易市场Kurzfassung
2026-10-23 23:00 GMT+8

This recommended resource is a video interview published by AIE CODE featuring Carmen Li, CEO of Compute Exchange and Silicon Data. Li, a former energy options market maker, discusses in the program why purchasing GPU compute capacity should adopt an exchange model similar to stock trading, rather than traditional point-to-point sales methods.

According to the episode description, Li details the mechanics of Compute Exchange: it is a neutral marketplace that owns no GPUs, verifies both providers and buyers, and handles reserved and forward contracts for GPUs and tokens, while also operating a busy market for used servers.

After a platform demo, Li turns to Silicon Data's numbers. As stated by her, forward curves for H100 and B200 have shifted up, the market expects prices to keep rising, and used H100s and B200s are holding their value well above standard depreciation.

This resource is suitable for professional readers interested in understanding AI infrastructure market structures, compute trading mechanisms, and GPU asset valuation logic. Although a full transcript is not provided, the publisher's chapter list and summary clearly define the scope of discussion, including the advantages of a neutral exchange, supply/demand verification processes, and lessons learned from energy markets.

Quellen:youtube.com

Forschung

US Treasury Proposes Expanding Outbound Investment Review to AI, Supercomputing, and Biotech

Thema · 美国对外投资执法Wesentlich
Verifiziert 2026-10-24 03:13 GMT+8

The U.S. Treasury Department proposed a rule on October 22, 2026, to strengthen oversight of American investments in sensitive technologies within "countries of concern." This proposal implements the Comprehensive Outbound Investment National Security Act passed by Congress in December 2025, upgrading the previous executive-order-based program into a statutory framework with broader scope.

Under the proposed rule, restricted sectors expand from semiconductors and quantum computing to include artificial intelligence (referred to by the Treasury using Trump's favored term "superintelligence" or SI), supercomputing, biotechnology, and hypersonic systems. The geographic scope covers China, Russia, North Korea, Cuba, and Iran. This shift signals Washington's intent to leverage capital controls to hinder adversaries' military modernization capabilities.

The rule sets a clear threshold: it applies only to investments exceeding $50,000 and explicitly excludes many financial investments, such as stock purchases. This means private equity, venture capital, and direct investments involving tangible assets or technology cooperation will be the primary targets. Jason Chipman, a former Justice Department lawyer and partner at WilmerHale, noted that this will require companies to establish more substantial compliance programs to manage risk.

The rule is currently open for public comment and has not yet become final law. Ryan Majerus, a former U.S. trade official, described it as just the first step toward a more enforcement-minded regime. Meanwhile, Chinese Foreign Ministry spokesperson Mao Ning responded that overstretching the concept of national security and taking discriminatory measures serve no one's interests, stating that China will closely monitor developments and take necessary measures to safeguard its interests.

Quellen:U.S. Department of the Treasury

Forschung

Amazon CEO Jassy: AI Demand Is Barbell-Shaped; Inference Costs Determine Whether the Middle Segment Opens Up

Wesentlich
2026-10-23 23:40 GMT+8

In an October 23, 2026 episode of the a16z podcast, Amazon CEO Andy Jassy described a demand structure he has repeatedly observed: AI demand is distributed in a barbell shape. One end of the barbell features AI labs and a few runaway applications (such as ChatGPT and Claude Code) that consume unprecedented amounts of compute power; the other end features numerous enterprises using AI for cost avoidance and productivity gains, such as automating customer service, fraud detection, and business process optimization. The middle segment of the barbell, which involves integrating existing enterprise production workloads with inference, represents the largest absolute market opportunity in his view, but the vast majority have not yet truly begun.

This observation itself is not new—many cloud providers have said similar things. Jassy’s contribution lies in linking a specific cost mechanism to this structure: why is the middle segment stuck? He believes inference costs are still an order of magnitude more expensive than the level required for widespread adoption. His reasoning is that if enterprises need to rewrite existing CPU-based applications to use AI accelerators, and the cost gap between the two solutions is large, the rewrite must yield very high incremental returns to be worthwhile; however, if inference costs come down, this threshold drops significantly, making almost all applications better, smarter, and more capable with AI accelerators and inference.

He then embeds Amazon’s chip strategy into this logic. Annapurna Labs was not originally acquired for AI chips—the core motivation for the 2015 acquisition was to offload virtualization, networking, security, and storage overhead from CPUs to specialized network interface cards, allowing servers to dedicate as much compute power as possible to customers. The team was subsequently used to build Graviton CPUs (which he claims are about 40% more price-performant than x86 processors, used by 99% of Top 1000 EC2 customers), and later Trainium AI accelerators (which he claims are 30% to 40% more price-performant than comparable processors). His judgment is that self-developed chips are the primary lever for reducing inference costs and thereby opening up the middle segment.

Boundaries to retain: These figures are estimates made by Jassy and self-reported data from Amazon, without independent verification. 'An order of magnitude' is his personal judgment, not a measured result. Reasons for the middle segment's lack of AI adoption may extend beyond cost—Jassy also cited talent familiarity with post-training and fine-tuning, and organizational ability to prioritize, as influencing factors. Jassy himself admitted it is 'hard to predict the rate at which this will happen.' Additionally, a16z, as a venture capital firm, has interests in Amazon and the companies discussed; while this does not invalidate Jassy’s observations, readers should be aware this is not neutral third-party analysis.

Nevertheless, the value of this framework for readers lies in breaking down the vague judgment that 'AI adoption is slow' into a trackable structure: both ends of the barbell are already growing (lab compute consumption, enterprise cost optimization), the middle segment is the future's largest increment, and inference costs are the observable variable connecting them. When pricing and performance data for Trainium or similar self-developed chips become public, readers can test whether this mechanism is being realized.

Quellen:youtube.com

Forschung

OpenAI Case Study: Benefit Street Partners Analyst Builds Financial Analysis Tools with Codex

Thema · OpenAI CodexKurzfassung
2026-10-24 02:20 GMT+8

OpenAI published a video case study on its official YouTube channel on October 23, 2026, titled 'How Benefit Street Partners is building financial analysis tools with Codex.' The video documents the work practices of Anant Kumar, an investment analyst and portfolio manager at the US-based alternative asset management firm Benefit Street Partners.

According to the publisher's description, Kumar faced a specific challenge: comparing how different firms valued the same loans, with this information spread across various Securities and Exchange Commission (SEC) filings. To address this bottleneck, Kumar utilized OpenAI's coding assistant, Codex, to develop a tool that automatically collects this data and transforms it into a website with visualizations.

The publisher notes that Kumar calls Codex the 'big unlock' in this process, mentioning that the first version of the tool took approximately one day to complete. The video illustrates the transition from a traditional analyst role to one incorporating development capabilities, and how building custom tools changed the speed and depth of his analysis. For readers seeking to understand the practical boundaries of AI coding tools in professional finance, this case offers a verifiable scenario, though it should be noted that the source is vendor-provided and lacks independent performance validation or full technical details.

Quellen:youtube.com

Forschung

Will Brown of Prime Intellect: Continual Learning Will Become Routine Like CI/CD

Kurzfassung
2026-10-23 23:46 GMT+8

Will Brown of Prime Intellect delivered a talk on continual learning at "The Learning Loop," an evening event hosted by LangChain in San Francisco. Brown's central thesis is that continual learning is about to get "boring," but in a good way: it is becoming the software development life cycle applied to models.

According to the episode description, Brown traces how agent workflows grow from terminal experiments into production software. He discusses why evaluations (evals) should capture the "vibe check" and lists the post-training methods he reaches for, including Supervised Fine-Tuning (SFT), on-policy distillation, context distillation, and online Reinforcement Learning (RL). He also expresses skepticism about jumping straight to On-Policy Self-Distillation (OPSD).

Furthermore, Brown covers build versus buy decisions for models, citing what quant finance shows about running many specialized models, and predicts that post-training will become as routine as Continuous Integration/Continuous Deployment (CI/CD). For readers focused on AI engineering implementation, this perspective offers a reference framework for translating abstract learning concepts into concrete software engineering practices.

Quellen:youtube.com

Forschung

Siemens Announces €500 Million AI Transformation Investment at German Erlangen Factory, Deploying Nvidia Tech to Cut Engineering Costs

Thema · 西门子AI工厂改造Wesentlich
Verifiziert 2026-10-24 02:22 GMT+8

German industrial giant Siemens has announced plans to invest approximately €500 million in the artificial intelligence transformation of its Electronics Factory in Erlangen, Bavaria, Germany, by 2030. This initiative is part of a roadmap between Siemens and Nvidia aimed at creating some of the world's first fully AI-powered manufacturing facilities.

According to reports, Nvidia is deploying systems including digital twins, AI-generated training data, and AI agents at the factory. One application uses AI to generate photorealistic virtual copies of printed circuit board (PCB) assemblies from digital twins, allowing inspection systems to be trained before physical prototypes are available. The companies claim this approach can cut engineering effort by up to 85% and reduce time-to-market by up to 40%.

A second application uses AI agents to model and predict production scenarios, validating proposed plans against a digital twin of the factory before implementation. Nvidia is providing its Cosmos world models to generate synthetic training data for quality inspection, while its Factory Operations Blueprint supports AI agents in production planning. Currently, more than 100 AI algorithms are already in use at the Erlangen site across production, logistics, quality inspection, and maintenance.

Samantha Mou, an analyst at Interact Analysis, noted that the significance lies in Siemens demonstrating the technology in its own operations rather than just promoting its potential, which helps encourage broader adoption among end users and integrators. However, Gartner analyst Jonathan Davenport cautioned that individual industrial AI point solutions will not scale; the key is developing a platform supporting multiple use cases. While efficiency gains are significant, these figures are based on a Siemens-owned, high-mix plant, and their reproducibility across multi-vendor brownfield sites remains to be proven.

Quellen:press.siemens.com

Forschung

Nvidia CEO Jensen Huang: AI Capital Demand Outstrips Supply as Neocloud Stocks Slide on Debt Fears

Thema · AI基建融资缺口Wesentlich
2026-10-23 19:00 GMT+8

Nvidia CEO Jensen Huang stated during an interview with CNBC at the GPU Technology Conference (GTC) in Berlin that "the demand of the capital is much greater than the availability of the capital." This comment comes amidst significant volatility in AI compute leasing companies, known as "neoclouds" such as CoreWeave and Nebius, highlighting the funding bottleneck facing AI infrastructure buildout.

Despite Huang's assertion that "the demand for the neo clouds is off the charts," market sentiment has been dampened. CoreWeave shares fell 10% and Nebius dropped 7% this week, driven by investor concerns over their heavy borrowing and lack of credit ratings. An executive at an AI infrastructure company told CNBC that the magnitude of unrated businesses was "freaking the market out."

To address revenue losses from deployment delays, Nvidia announced a new product packing seven Vera Rubin AI racks into a shipping container to speed up data center activation via the supply chain. Huang noted that a 1-gigawatt data center costs approximately $50 billion, while Goldman Sachs estimates total AI-related capital spending could reach $7.6 trillion between 2026 and 2031.

Huang believes the missing element is working capital to fuel growth but added that this scarcity "is going to get solved." However, one AI infrastructure financing executive told CNBC that very soon capital will become the bottleneck.

Quellen:CNBC

Forschung

Western Digital Reports Fiscal Q1 2027 Results: Revenue Up 55%, Guides for Over 66% Growth Next Quarter

Wesentlich
Verifiziert 2026-10-24 01:46 GMT+8

US storage device manufacturer Western Digital Corporation (Nasdaq: WDC) reported its financial results for the first quarter of fiscal year 2027 (ended October 2, 2026) on October 22, 2026. The report shows revenue of $4.38 billion, a 55% increase year-over-year. Non-GAAP gross margin reached 58.1%, significantly expanding from 43.9% in the same period last year. Non-GAAP diluted EPS was $4.64, up 161% year-over-year. Operating cash flow was $1.60 billion, and free cash flow was $1.38 billion.

CEO Irving Tan stated that results exceeded expectations due to "robust customer demand and disciplined execution," noting the company is executing well through technology transitions to meet growing storage demand. As a key supplier of storage infrastructure in the AI data economy, Western Digital partners with leading hyperscalers, cloud service providers, and enterprises.

The outlook for the second quarter of fiscal year 2027 (Q2FY27) is even more optimistic, with expected revenue of approximately $5.1 billion (range ±$100 million), implying a year-over-year growth rate of 66% to 72%. Non-GAAP gross margin is projected at 60%-61%, and non-GAAP EPS at $5.80 (range ±$0.20). Additionally, the Board declared a quarterly cash dividend increase of 17% to $0.175 per share.

Despite beating Wall Street forecasts on both results and guidance, Seeking Alpha reported that Western Digital shares erased earlier gains and fell about 1.5% in extended trading. This may reflect cautious sentiment toward high-valuation tech stocks or concerns about the sustainability of AI infrastructure spending.

Quellen:Western Digital

Forschung

OpenAI Engineers Describe the Multi-Tenant Cycle of Their Storage Platform: When Workloads Become Outliers for Cloud Providers

Wesentlich
2026-10-24 01:00 GMT+8

OpenAI’s storage platform, Habitat, underwent a complete cycle from multi-tenancy to single-tenancy and back to multi-tenancy. Three technical members—Jon Lee, Hrachya Hakobyan, and Chaomin Yu—documented the reasons and costs behind this cycle in an engineering blog post published on October 23, 2026. The core takeaway is not a specific performance metric, but a broader principle: when a company grows large enough that its own workloads become outliers in a cloud provider’s infrastructure, 'managed services' cease to be fully managed.

Habitat is the online storage platform relied upon by all OpenAI products (ChatGPT, API, Codex, and internal services). It must respond in milliseconds to over one billion users across 40 geographic regions, managing more than 500 petabytes of data. Its underlying storage uses Azure Cosmos DB. In the first phase, the team relied entirely on Cosmos DB’s native multi-tenant capabilities: each account used only a default container, allowing all products to share it. This quickly exposed three problems.

The first problem was logical partition key collisions. Because early API designs exposed the choice of logical partition keys directly to clients, different product use cases frequently used the same key names (such as user_id or workspace_id). This caused cross-product requests for the same user to share the same partition quota, constraining overall throughput. The second problem was noisy neighbors. Although the documented limit of 10K RU/s applied to logical partition keys, enforcement actually occurred at the physical partition level. Multiple logical partition keys coexisting on the same physical partition meant that background batch operations for niche product features could slow down core login and signup flows. The third problem was weak partitions: as data volume grew, physical partitions split, keeping total throughput constant but halving the available capacity per partition. For storage-intensive use cases, data growth alone reduced the throughput capacity for active workloads even if request rates remained stable.

In the second phase, the team moved to creating independent containers for critical new use cases to achieve workload isolation. This partially alleviated partition key collisions and noisy neighbor issues but introduced new organizational failures. Independent containers meant every product team had to handle capacity planning themselves. In a shared container, a use case growing from 0 QPS to 1 million QPS had negligible impact on total usage; in an isolated container, this might represent a 1,000-fold baseline jump. Product teams often forgot to pre-scale before launch, forcing the platform team to sit in incident response rooms on launch days to adjust container parameters last minute. More fundamentally, this model required every product team to have deep knowledge of Azure Cosmos DB configuration details, contradicting Habitat’s design goal of letting product developers avoid understanding the database.

In the third phase, the team returned to multi-tenancy, but this time as a result of joint engineering with the Azure team. Specific fixes included hot-key throttling (prioritizing throttling of the hottest keys to prevent any single key from exhausting a physical partition’s capacity), request tiering QoS (marking offline workloads like bulk inserts and data exports as low priority to ensure interactive requests are not crowded out), and an improved Serverless mode built by the Azure team specifically for OpenAI. This mode allowed each physical partition to scale up to 10K RU/s and introduced splits based on RU usage (in addition to storage usage), eliminating the need for manual monitoring and container scaling.

Quellen:OpenAI

Forschung

Google Updates Privacy Policy: Data Connected to AI Features May Be Used for Targeted Ads

Thema · 谷歌隐私政策AI广告争议Wesentlich
Verifiziert 2026-10-24 01:12 GMT+8

U.S. technology company Google updated its privacy policy on October 16, 2026, adding a key exception clause. Under the new terms, if users choose to connect personal data from services like Gmail, Google Drive, or Google Photos to Google’s “Personal Intelligence” features to enhance AI search experiences, summaries and inferences derived from this data may be used to show them personalized ads. Previously, Google promised not to use content from these services for advertising purposes.

This change implies that users seeking more precise AI recommendations may need to trade off some privacy boundaries. Google states that advertisers do not get direct access to raw emails or photos, and ads will not be based on sensitive categories such as race, religion, sexual orientation, or health. However, when AI systems process user-authorized data, they generate summaries, excerpts, and inferences, which Search and other services may use to personalize content, including ads. For example, AI might scan emails and photos to summarize travel preferences, potentially influencing the types of ads shown on YouTube or other platforms.

For Google users wishing to avoid this scenario, they can disconnect linked content apps or disable “Personalized Recommendations” entirely via the “Search Services Personalization” page in their account settings. Note that disconnecting an app does not clear information already collected by Google and stored in “Search Services History”; users must separately delete corresponding activity records to remove this data completely.

Quellen:policies.google.com

Forschung

US AI Agent Cloud Startup Daytona Raises $60M Series B at $1 Billion Valuation Led by FirstMark

Thema · Daytona融资动态Wesentlich
Verifiziert 2026-10-24 00:34 GMT+8

New York-based AI infrastructure startup Daytona Platforms announced on October 23, 2026, that it has closed a $60 million Series B funding round at a $1 billion post-money valuation. The round was led by FirstMark Growth, which also led Daytona's $24 million Series A in February 2026. This brings the company's total capital raised to $91 million.

Daytona is building a cloud platform specifically designed for long-running AI agents. Founder Ivan Burazin stated that traditional clouds are built for long-lived services, whereas agents spin up thousands of short, stateful jobs requiring specialized infrastructure. Daytona's solution provides each agent with its own virtual computer, offering sandboxed access to Linux, Windows, macOS, and GPUs.

Beyond institutional investors, the round attracted a notable group of angel investors from the tech industry, including Patrick Collison (Stripe), Tobi Lütke (Shopify), Aravind Srinivas (Perplexity), Eric Yuan (Zoom), Olivier Pomel (Datadog), and Adam D'Angelo (Quora). Qualcomm Ventures, Samsung Next, and Okta Ventures also participated in the financing.

Quellen:daytona.io

Forschung
Weiter →
Anmelden / Registrieren
LesenForschungRadarAnlageframework

The most notable judgment for readers appears in the section on 'Managed Services.' The authors wrote: 'When companies grow to OpenAI’s scale, managed services become less “managed.” Many workloads become outliers in the cloud provider’s infrastructure.' They cited two real incidents from late 2024: one where a regional power outage caused failover to take several hours (neither team had tested at this scale); and another during failback, where a defect in the gateway-mode client caused all connections to flood the original region, overwhelming the gateway. These incidents revealed not a bug, but a systemic mismatch in testing and operational models between the two teams.

Since then, Habitat and the Azure engineering teams established a deep joint operations model: shared test environments and dashboards, direct engineering channels (calling each other via Teams, bypassing support tiers), joint ownership of failure modes that only appear at OpenAI’s scale, and using production-sized Cosmos DB accounts in pre-production environments for change validation. Over the past year, both sides completed improvements to control plane and gateway scalability, metadata caching, connection pool optimization, workload colocation, and isolation of critical and offline workloads via different VPCs.

On the performance side, the team developed a layered read optimization strategy (read requests account for over 90% of Habitat’s total traffic): highly replicating the smallest, most latency-sensitive datasets (such as authentication metadata) across multiple regions; deploying Valkey distributed caches close to where user requests are processed; propagating cache invalidations globally within seconds via Azure Cosmos DB Change Data Capture (CDC) (following Meta’s TAO system pattern); and rewriting the Habitat service itself in Rust to eliminate proxy overhead. These measures reduced global read p50 latency from approximately 120 ms to roughly 10 ms.

It should be noted that all the above numbers, incident descriptions, and judgments are self-reported by the OpenAI team and have not been independently verified. Their scale—largest Cosmos DB account storing tens of petabytes, serving over 20 million QPS—is unreproducible for most engineering teams. The 'outlier' problem itself only arises when workloads are large enough to invalidate standard operational assumptions of cloud providers. For teams that have not reached this scale, the layered caching strategy and joint engineering model described in the article are more directional references than directly actionable solutions. However, the 'multi-tenant cycle' itself—from shared to isolated to shared with fixes—is an architectural decision pattern that repeatedly appears during organizational expansion, and its trigger condition (organizational cost of capacity planning exceeding the benefits of isolation) has broader applicability.