LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-11-1525 publications

Actualiser

GeekWire Editors Analyze Nvidia Succession: Company Needs a 'Crisis Manager' Rather Than a Technical Genius

Sujet · 英伟达CEO继任Matériel
2026-11-15 06:10 GMT+8

Nvidia’s board says it is too early to discuss who will succeed Jensen Huang. GeekWire editors Todd Bishop and John Cook clearly do not intend to follow this advice. In a podcast recorded on November 13 and an article published on November 14, they offered a clear judgment: what Nvidia needs now is not another technical genius, but a crisis manager capable of steering the company through turbulence.

The analogy for this argument draws on Alan Mulally’s move from Boeing to Ford, and Dara Khosrowshahi taking over Uber after Travis Kalanick was ousted. Cook stated in the podcast that Nvidia needs someone who can lead the company through chaos, not just a great technology expert. This judgment rests on a background fact: Jensen Huang has been resistant to succession planning. Arm CEO Rene Haas quoted Huang in a 2024 interview as saying, 'He didn’t like the idea; he believed it would create competition among executives and cause division.'

Huang himself said in an October 14 Bloomberg interview that he hoped his successor was already inside the company. However, the board stated it would consider both internal and external candidates, and its Friday statement indicated it was still selecting a search firm and had not formally engaged any candidates. The article lists 11 mentioned candidates spanning chip design, cloud computing, consumer electronics, and storage. Besides Haas, who had previously been reported by the Financial Times, the list includes AWS CEO Matt Garman, Amazon’s head of chips and AI Peter DeSantis, AMD CEO Lisa Su, Google Cloud CEO Thomas Kurian, Micron CEO Sanjay Mehrotra, Apple Executive Chair Tim Cook, and several Microsoft executives.

Several newly disclosed facts in the article are worth noting. Nick Parker is a veteran who spent 26 years in Microsoft’s sales system; he joined Nvidia in August this year as Executive Vice President of Global Field Operations, with a compensation package valued at over $40 million. Peter DeSantis currently oversees Amazon’s AI models, custom chips, and quantum computing business, reporting directly to CEO Andy Jassy. Dave Treadwell worked at Microsoft for 27 years before joining Amazon in 2016, and took over AWS compute and machine learning services in August this year.

These personnel changes themselves constitute a facet of the competitive landscape in the AI industry. The article points out that whoever ultimately succeeds will face a market where customers are trying to reduce their reliance on Nvidia. In his annual shareholder letter in April, Jassy wrote that if Amazon’s chip business sold to AWS and third parties like other chip companies, its annual revenue run rate could reach approximately $50 billion. DeSantis also told Bloomberg in June that Amazon was negotiating to sell Trainium AI chips to other companies.

Haas’s case involves a notable complex relationship. He served as Vice President of Sales at Tensilica, whose co-founder Harvey Jones was appointed as Nvidia’s first board chairman this week. Meanwhile, SoftBank appointed Haas as CEO of SoftBank International in April to coordinate its international chip and AI portfolio, which includes Project Izanagi, an initiative aimed at competing with Nvidia. The article lists this as a 'complicating factor' in Haas’s candidacy, rather than a confirmed conflict of interest.

Sources :GeekWire

Recherche

Technical Author's Real-World Test: Claude Fable 5.2 Has a Lower Unit Price, but Actual Coding Task Costs Are About Three Times That of GPT-6 Astra

Matériel
2026-11-14 22:00 GMT+8

When Anthropic released Claude Fable 5.2 on November 9, it set its standard API price at $8 per million uncached input tokens and $40 per million output tokens, which is lower than OpenAI's GPT-6 Astra ($10/$50). This price cut was widely promoted as a cost advantage for Fable 5.2. However, The New Stack technical author Jessica Wachtel conducted independent tests using three coding tasks and found that a lower list price does not necessarily mean lower actual spending.

Wachtel ran each model once via the respective vendor APIs with default settings to test three tasks simulating real development work: fixing 10 planted bugs in a 2,600-line billing codebase, reviewing an EU invoicing package for 45 manually planted defects, and writing validation code from scratch for VAT IDs in 27 countries and IBANs in 34 countries. Each task was scored out of 50 by an automated script.

In the bug-fixing task, both models fixed all 10 bugs and passed all 50 hidden checks, each requiring 12 conversation turns. However, the processes differed significantly: Fable 5.2 consumed over 17,000 reasoning tokens, while Astra used only 819. Fable 5.2 cost approximately $2.51 and took 326 seconds, whereas Astra cost about $0.98 and took 122 seconds. For the same score, Fable 5.2 cost about 2.6 times more than Astra.

Code review was the only task where scores differed. Fable 5.2 scored 49, while Astra scored 45. Fable 5.2 identified 44 of the 45 defects with no false positives, while Astra identified 41 and incorrectly classified one Spanish ID as valid. However, Fable 5.2 spent $6.05 and 993 seconds to achieve this, compared to Astra's $0.95 and 227 seconds. Fable 5.2's extra 4 points came at a cost exceeding six times that of Astra and four times the time.

In the code-writing task, both models achieved a perfect score of 50, yet Fable 5.2 still cost 2.3 times as much as Astra. Across all three tasks, Fable 5.2's total cost was approximately $13.39, while Astra's was about $3.99—despite Fable 5.2's lower unit price, the total cost was roughly 3.3 times higher. Wachtel noted that Fable 5.2 consumed about five times as many tokens as Astra across these three runs.

These results corroborate OpenAI President Greg Brockman's statement at the Astra launch that price per task is key. Wachtel concluded that for such tasks, she would start with Astra, which achieved the same scores on bug fixing and code writing at about 41% of the cost; she would only consider Fable 5.2 when careful code review was needed, as it found three additional defects in that specific instance. She also acknowledged that both models missed some defects and that each model was run only once, so the four-point gap could shift in another run.

Sources :The New Stack

Recherche

DeepSeek Opens Full Suite of Application APIs Including Voice and OCR, Enabling Low-Cost App Replication

Sujet · DeepSeekMatériel
2026-11-14 10:39 GMT+8

Chinese AI company DeepSeek announced on November 14, 2026, that it has officially opened its "full suite" application APIs. Features previously available only within its official App—such as web search, deep thinking, file upload, link reading, and newly added voice recognition (ASR) and speech synthesis (TTS)—are now provided as independent interfaces. Developers can combine these capabilities as needed and are billed separately for each service.

The key update lies in the comprehensive opening of multimodal capabilities. Although DeepSeek launched Search and Fetch interfaces in October, developers previously had to write custom code to orchestrate model decisions, search execution, and content summarization. Now, by simply enabling web search, the model automatically handles search timing, content retrieval, and result integration. Additionally, voice capabilities now have official independent interfaces; previously, developers building voice-enabled assistants typically needed to integrate third-party services. Currently, the V4.1-Flash model supports these extended functions across Messages, Responses, and Chat Completions interfaces.

In terms of pricing, DeepSeek has adopted an aggressive strategy to drive adoption. Voice recognition is discounted by 50% to 0.39 yuan per hour (from 0.78 yuan), and speech synthesis is also halved to 1.5 yuan per 10,000 characters (from 3 yuan). For comparison, OpenAI's GPT-Transcribe costs approximately 1.81 RMB per hour of audio, while Google Cloud's standard speech service reaches 6.43 RMB. Web search and fetching remain at 5 yuan per 1,000 successful calls, and Optical Character Recognition (OCR) is priced at 1.5 yuan per 1,000 pages.

DeepSeek has also open-sourced a lightweight chat frontend Demo that supports importing historical conversations, aiming to help developers quickly build products with functionality close to the DeepSeek App. The company stated that developers can even "replicate the complete DeepSeek App 1:1." This move indicates DeepSeek's attempt to attract developers to build next-generation AI applications on its platform by lowering basic tool invocation costs and providing a complete technology stack, rather than relying solely on its own consumer-facing chat product.

Sources :量子位

Recherche

South Korean Securities IT Firm DefiSoft Launches Paave 2.0 in Vietnam, Enabling User-Created AI Agents for Simulated Stock Trading

Prise rapide
2026-11-13 16:29 GMT+8

South Korean securities IT firm DefiSoft has officially launched its market analysis and investment platform, "Paave 2.0," in the Vietnamese market. The core innovation of this version allows users to build their own AI agents capable of automatically analyzing real-time quotes, financial statements, and public disclosures from the Ho Chi Minh City Stock Exchange (HOSE), Hanoi Stock Exchange (HNX), and UPCoM, and executing simulated stock trades using virtual funds.

Through the "Agent Studio" feature, users can define analytical perspectives and prompts to create researcher agents that write market reports or investor agents that execute buy/sell orders. All simulated trading results generated by these AI agents are recorded on the same leaderboard as human users, with trading logs made public to the community. This mechanism aims to lower the barrier for individual investors to use AI tools for market analysis through gamified competition.

Currently, Paave 2.0 supports only simulated trading with virtual funds, but DefiSoft plans to integrate real-money trading within the year, enabling users to place orders via their own accounts through partnerships with local Vietnamese brokers. Additionally, the platform includes a conversational analysis tool called "Paave AI" and supports connecting external large language models like Claude or ChatGPT via the MCP protocol to query market data directly.

Sources :스타트업 생태계의 모든 것 \'플래텀(Platum)\'

Recherche

NVIDIA Releases New cuBLAS Algorithms to Accelerate Double-Precision Scientific Computing

Sujet · 英伟达cuBLAS新算法Matériel
2026-11-14 06:13 GMT+8

NVIDIA has introduced two new algorithms, Ozaki-II and 2M, in the latest release of its CUDA Toolkit 13.4.2 for the cuBLAS math library. These technologies aim to resolve a long-standing contradiction in high-performance computing: how to leverage the faster integer Tensor Cores—designed primarily for low-precision tasks—while maintaining the full accuracy of double-precision floating-point (FP64) calculations. This update allows scientific simulations dependent on high-precision dense general matrix multiplications (GEMM), such as those in molecular chemistry and materials science, to automatically gain speed without requiring any changes to application code.

Traditionally, GPU FP64 units offer high precision but limited throughput, while Tensor Cores provide extreme speed for lower-precision data. The new cuBLAS methods employ "floating-point emulation" techniques, decomposing each double-precision number into multiple integer segments, performing exact integer multiplications on the Tensor Cores, and reassembling the result to meet strict FP64 standards. Specifically, the Ozaki-II algorithm utilizes the Chinese Remainder Theorem (CRT) to significantly reduce the number of required integer multiplications compared to its predecessor, Ozaki-I. Meanwhile, the 2M algorithm optimizes complex matrix multiplication by reducing the operation from four real multiplications to two, thereby lowering computational overhead.

According to NVIDIA's official technical blog, these algorithms demonstrate substantial performance improvements in real-world applications. In quantum chemistry, DGEMM (real double-precision matrix multiplication) using Ozaki-II runs 5.5x faster than native FP64 on NVIDIA VR NVL72 systems and 3.9x faster on GB200. For Coupled Cluster Singles and Doubles (CCSD) calculations—considered the "gold standard" in quantum chemistry—the end-to-end speedup reaches 2x on VR NVL72 and 1.4x on GB200. When employing mixed-precision modes, these gains extend to 5x and 3.5x respectively.

In materials science, ZGEMM (complex double-precision matrix multiplication) within the BerkeleyGW software package achieves a 5x speedup on GB200 when using the 2M algorithm. Tests on the BerkeleyGW Epsilon module show an end-to-end speedup of 6.6x on VR NVL72 and 3.9x on GB200. Notably, all tests verified that results remain within FP64 accuracy bounds, and users do not need to manually adjust APIs or rewrite kernels, as cuBLAS automatically selects the optimal path.

Furthermore, the upcoming NVIDIA Rubin GPU architecture introduces the new TI16 numeric format and the tcgen05.mma instruction, which will further accelerate Ozaki-series computations. Existing users can enable these features simply by downloading and installing the latest CUDA Toolkit. Developers can refer to the floating-point emulation section of the cuBLAS documentation for detailed API control options, such as cublasSetEmulationStrategy(), to fine-tune the balance between precision and performance for specific workloads.

Sources :NVIDIA Technical Blog

Recherche

Nvidia's Top Shareholders Urge Board to Consider Jensen Huang's Return

Sujet · 英伟达CEO继任Matériel
2026-11-15 01:35 GMT+8

Two major institutional investors in Nvidia—Brad Gerstner of Altimeter Capital and Gavin Baker of Atreides Management—published an open letter on Friday urging the chipmaker's board to pursue a "slow, measured, and deliberate" process for selecting a new CEO, explicitly stating that the option of founder and former CEO Jensen Huang's return should remain on the table.

The firms collectively hold approximately $2.13 billion in Nvidia stock. The letter acknowledges that interim CEO Colette Kress is capable of leading the company but argues that Huang, as the long-time leader, has earned the right to a careful evaluation before any hasty replacement decisions are made. This move serves as a direct counterweight to widespread media reports suggesting the board was preparing to permanently oust Huang.

Huang resigned on November 12 amid allegations involving personal relationships and an investigation into suspected Chinese spy Alice Huang. While the market largely anticipated his permanent departure, this shareholder intervention highlights a growing divide among investors who believe Huang's removal could inadvertently benefit China by weakening Nvidia's strategic advantage in AI hardware. The resulting governance uncertainty adds complexity to Nvidia's leadership transition in the coming weeks.

Sources :CNBC

Recherche

OpenAI Launches ChatGPT Wallet to Let Dots Agents Make Purchases with Biometric Approval

Sujet · ChatGPT钱包Matériel
2026-11-14 04:45 GMT+8

OpenAI launched "ChatGPT Wallet" on November 13, 2026, for ChatGPT Pro subscribers in the United States. The new feature allows OpenAI's Dots AI agents to complete online shopping tasks on behalf of users. Currently, access is restricted to users aged 18 and older with a US billing address who hold eligible US-issued credit cards from American Express, Chase, Mastercard, or Visa.

According to OpenAI's support page, once a card is added to the wallet, Dots agents can use it to make purchases based on user prompts, but they must check in with the user before checking out. Every purchase made through ChatGPT Wallet requires a security check involving biometric verification, such as Face ID or a fingerprint scan. OpenAI states that this mechanism uses payment tokenization similar to Apple Pay or Google Wallet: actual credit card details are not shared with the agent or the merchant, protecting user privacy.

While the current version mandates user approval for each transaction, OpenAI indicated future plans to allow users to set parameters like spending limits, purchase categories, or time periods, enabling fully delegated pre-approved purchases. However, widespread adoption faces hurdles. Some retailers restrict or block AI agents entirely. OpenAI noted that websites need to ensure agents are not inadvertently blocked, and payment systems must recognize legitimate agent-initiated purchases. Amazon previously blocked Meta's Muse agents, highlighting the cautious stance of major e-commerce platforms toward agentic shopping.

Sources :The Verge

Recherche

Russia's Communications Watchdog Deploys AI Filters to Hunt Down VPNs in 2.2-Billion-Ruble Crackdown

Sujet · RoskomnadzorMatériel
2026-11-14 01:31 GMT+8

According to TechRadar, Russia's communications watchdog, Roskomnadzor, has reportedly officially deployed machine learning filters to choke virtual private networks (VPNs) across the country. This move marks a sharp escalation in the Kremlin's digital barrier. The state censor poured 2.27 billion rubles into the project, according to government digital transformation documents reviewed by Forbes Russia and reported by The Moscow Times.

Unlike previous attempts to decrypt user content, the newly deployed AI tool is a predictive model that searches for behavioral fingerprints in connection metadata: packet sizes, transmission intervals, session lengths, and hosting server patterns. Dmitry Kuteynikov, an expert at digital rights organization RKS Global, explained that the tool helps staff decide what to block and warned that as the model comes into use, blocks are likely to come faster, making VPNs less reliable.

A developer working for Russian provider Amnezia VPN confirmed that this kind of model could affect a wide range of protocols because it targets the general characteristics of VPN traffic rather than any single one. While obfuscation hides a protocol’s signature, the traffic pattern of a tunnel stays visible, which is exactly what a behavioral model looks for.

The automated filtering is already causing disruption across Russian business networks. Although Russian law technically exempts internal corporate VPNs, enterprise connections look nearly identical to circumvention tunnels under automated inspection. As reported by Kommersant, telecom sources noted that traffic analysis algorithms misclassify traffic in 5% to 6% of cases, causing unrelated websites and corporate cloud applications to collapse when entire hosting subnets are blacklisted.

Earlier roadmaps showed that Russia drafted plans for AI that could block 95 percent of VPNs. While official targets were quietly revised, funding continues to surge. The Kremlin is preparing to double its internet censorship budget in coming years, with another 10.2 billion rubles allocated to upgrade deep packet inspection boxes in 2026 alone.

Sources :TechRadar

Recherche

Asian Governments React to Nvidia CEO Jensen Huang's Exit: Confirming Supply and Monitoring Investigations

Sujet · 黄仁勋离职与调查Matériel
2026-11-15 00:34 GMT+8

The departure of Nvidia CEO Jensen Huang has triggered widespread reactions across Asia. Some governments have moved quickly to confirm chip supply commitments, while others are monitoring the impact on local projects or addressing ongoing investigations.

In South Korea, the Ministry of Science stated that Nvidia had reaffirmed its plan to supply 260,000 advanced GPUs to the country and to continue joint AI projects. Samsung Electronics and SK Hynix, major suppliers of memory chips for Nvidia's AI processors, maintain close ties with the company. Additionally, South Korea's government AI strategy council held an emergency meeting on Friday to assess the potential impact of Huang's departure on the country's AI industry.

In Singapore, the Ministry of Home Affairs confirmed that the investigation into Alice Huang, former CEO of Megaspeed, remains ongoing. Authorities seized her passport in June 2025 to investigate possible false declarations regarding server purchases. In February 2026, police found insufficient evidence to establish offenses related to these declarations but noted she knowingly provided inaccurate information about her role at Megaspeed. She was issued a warning and barred from re-entering Singapore. Singapore also mentioned receiving limited assistance from U.S. authorities despite repeated requests for information about server maker Supermicro.

In China, the Foreign Ministry rejected allegations linking a Chinese businesswoman to espionage as "groundless." A Reuters review indicated that Huang's exit has raised uncertainty over more than $3 billion in Chinese AI computing projects reliant on Nvidia's restricted processors. Many of these chips are subject to U.S. export controls, making them difficult for Chinese companies to obtain.

In Taiwan, Economy Minister Kung Ming-shin said Huang's departure was unlikely to significantly change the growth of the global AI industry. The Taipei city government stated that Nvidia's planned overseas headquarters in the city remained on track despite the leadership change, noting the company has met its contractual obligations so far.

In Japan, the Industry Ministry said the government would closely monitor Nvidia, describing the chipmaker as an important player in Japan's AI semiconductor industry. Deputy Chief Cabinet Secretary Masanao Ozaki said it was too early to assess the impact on the global AI industry, while stressing Japan's commitment to ensuring a stable supply of semiconductors.

Sources :TechCrunch

Recherche

Roundhill Investments CEO Dave Mazza and Yorkville Ives Partner Dan Ives: Nvidia Turmoil and Data Center Restrictions Boost Value of Compute and Alternative Chips

Sujet · 英伟达CEO辞职市场影响Matériel
2026-11-14 23:50 GMT+8

In the week when Nvidia founder and CEO Jensen Huang resigned, AMD shares rose nearly 13%, Broadcom posted similar gains, and the Nasdaq Composite recorded its best single-week performance since August. When one company encounters trouble, the entire sector strengthens—a seemingly contradictory market reaction that two AI-themed ETF managers view as an investment thesis worth explaining, rather than simple panic-driven rotation.

Dave Mazza, CEO of Roundhill Investments, proposed a core mechanism on CNBC's "ETF Edge": if investors lose trust in Nvidia, they will not stop buying AI chips but will instead turn to other suppliers for compute power. This means Nvidia's trust crisis creates a window for demand transfer to competitors in the short term. The Magnificent Seven ETF (MAGS) managed by Mazza closed the week up less than 1%, while the Roundhill Neocloud ETF (NCLD), which focuses on small- and mid-cap specialized cloud computing companies, rose more than 1%, despite having fallen cumulatively by 39% since its launch in August 2026.

Mazza further pointed out that hyperscale cloud service providers may benefit from Nvidia's difficulties because they already possess available compute infrastructure. With demand remaining constant and new supply constrained, the value of existing compute capacity rises.

Dan Ives, partner at Yorkville Ives, added a political dimension to the analysis. Following the U.S. midterm elections on November 3, 2026, where the Democratic Party gained control of Congress, Ives estimated that this political shift could prevent 10% of planned data centers from being built, with high-end estimates ranging from 15% to 20%. However, his conclusion was not bearish on AI compute; rather, he reasoned inversely: given that demand still exceeds supply by approximately four to one, tighter regulation makes existing compute scarcer and therefore more valuable. Before Nvidia's Q3 earnings report, Ives lowered his price target on Nvidia from $300 to $225 but maintained a buy rating, arguing that a valuation of roughly 10 times 2028 earnings per share remains too low.

The applicability of this argument comes with important caveats. First, the 4:1 supply-demand ratio is Ives' personal estimate and has not been independently verified. Second, the Neocloud ETF fell 39% just three months after its launch, indicating that not all market participants accept the linear logic that compute scarcity equals value enhancement—balance sheet pressures and ongoing concerns about Nvidia's role in financing AI infrastructure building may offset premiums derived from scarcity. Third, Nvidia's Q3 earnings are due soon; if its data center revenue is not significantly impacted, the urgency of the substitution logic will be weakened. Essentially, the managers' argument is a bet on asset repricing under supply constraints, not a denial of Nvidia's long-term competitiveness. Ives explicitly stated that Nvidia remains about two to three years ahead of competitors, and its strategic roadmap and chip development pace driven by 42,000 employees will not change due to the founder's departure.

Sources :CNBC

Recherche

Anthropic Expands into Enterprise Software, Raising Developer Concerns That the Platform Provider Is Becoming a Competitor

Sujet · AnthropicMatériel
2026-11-14 23:15 GMT+8

Scott Stevenson, CEO of legal AI company Spellbook, recounted an experience in a November 2026 interview with Forbes: Anthropic employees had assured him that the company had no plans to compete directly in the legal software market, before Anthropic launched Claude for Legal, which includes contract review features overlapping with Spellbook's core business. Stevenson did not abandon Claude—Anthropic still showcases Spellbook on its customer page today, reporting that its Claude-powered agents perform approximately 530,000 contract reviews per month. However, he drew a conclusion: 'Anthropic has been a great partner, but it made us realize they are a massive company and two parts of the org can have conflicting mandates.'

Gerald Zingraf, President and COO of GlacierGrid, expressed similar concerns on LinkedIn, comparing Anthropic's strategy to Amazon's private-label business. His worry is that Anthropic can see which third-party integrations customers use most frequently and potentially leverage that information when deciding which capabilities to build itself. 'Makes me nervous for wrappers and pure SaaS companies,' he wrote, citing his own experience selling design products on Amazon and then watching the retailer launch competing products.

Todd Saunders, CEO of Dalton Mills, posted on X after Anthropic launched Claude Code Review, directly stating that 'Anthropic is the new Amazon,' warning developers building on the Claude SDK that the company could observe application usage and eventually incorporate competing functionality into Claude.

These concerns point to a specific mechanism: Anthropic controls the interface between users and third-party applications. The November 10, 2026 announcement introduced Claude Sheets, Claude Whiteboards, and Claude Data, while previewing Claude Meetings. Expanded interactive connectors allow partners to embed applications directly inside Claude, with software from Adobe and Atlassian already included. This means Anthropic acts simultaneously as a technology supplier, an owner of the application distribution channel, and a potential functional competitor—the tension among these triple roles is structural, regardless of any single decision-maker's goodwill.

However, the article itself remains cautious regarding the strongest allegations. Regarding Zingraf's concerns, the author explicitly states that 'he provided no evidence that the company uses integration data to identify competitors or guide its product roadmap.' Regarding Stevenson's experience, the author similarly notes that 'Stevenson’s account does not establish that Anthropic used Spellbook’s proprietary information to develop its legal product.' Anthropic’s commercial terms stipulate that customers retain rights to their inputs and own their outputs, and the company agrees not to use customer content to train models; its commercial privacy documentation also clarifies that feedback data excludes content pulled through connectors (including local and remote MCP servers) unless users paste it directly into a conversation.

Sources :The New Stack

Recherche

DuckLake Lead Developer: Treat Metadata as a Relational Database Problem, and Lakehouse Queries Become Just Catalogs and Parquet Files

Sujet · DuckLakeMatériel
2026-11-14 23:45 GMT+8

Traditional lakehouse formats split metadata into numerous manifest files scattered across object storage. For every query, the engine must traverse these files to determine which Parquet data files to read. Pedro Holanda, lead developer at DuckLabs, argues that this is essentially a relational database problem that does not need to be solved using file system methods.

His explanation is straightforward: you issue a query, and the server answers with which Parquet files to read. The Parquet files themselves do not change; only the catalog changes. This is why DuckLake stores table metadata in a standard SQL database rather than scattering manifest files. Users are already familiar with SQL and schemas, so they do not need to learn an additional logic for traversing metadata.

DuckLake 1.1 was released in October alongside DuckDB 2.0.0 LTS, bringing file compaction and inline/flush mechanisms tailored for streaming data lakes. Regarding file compaction, when the number of files becomes excessive, DuckLake can merge multiple files into one to reduce request counts; this operation can span multiple snapshots, further compressing the total file volume. Regarding inlining, actual data representations are stored directly as tables within the catalog. Holanda stated that while building DuckLake, they realized that since they already had a database system and a catalog, why not store the data itself in tables?

This resolves the small-file problem in streaming scenarios. When multiple data sources write simultaneously, traditional lakehouses generate a large number of fragmented files. Queries then require traversing metadata and reading many small Parquet files, slowing down both processes. DuckLake absorbs these fragments into the catalog through inlining and flushing, deferring compaction operations until they are needed.

Weston Pace, an engineer at LanceDB, supplemented DuckLake's design trade-offs from the perspective of concurrent writes in another technical blog post. He categorized table-format coordination mechanisms into two types: zero-infrastructure atomic operations (relying on cloud storage's 'put-if-not-exists' semantics, requiring no extra services but limited to approximately ten commits per second) and centralized coordination (where DuckLake uses a Postgres instance as a commit coordinator, synchronizing writers via mutexes or CAS primitives). Centralized schemes may achieve higher transaction rates and apply to filesystems that do not support atomic creation, at the cost of maintaining a long-running service.

Both DuckDB and DuckLake are open-source under the MIT license, overseen by the independent DuckDB Foundation. DuckLake can run without massive server clusters, partly because the Parquet and DuckDB storage formats are naturally similar—both are columnar and organized by row groups, minimizing data movement steps.

Sources :The New Stack

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement

Betting markets provide a rough thermometer of public opinion. On the Kalshi platform, by Saturday noon, Haas had surpassed the previously leading Ian Buck, creator of Nvidia’s CUDA platform (18%), with a probability of 25%, while interim CEO Colette Kress stood at 10%.

Two major investors, Brad Gerstner of Altimeter Capital and Gavin Baker of Atreides Management, wrote to the board requesting a prudent search and retaining 'the possibility of Jensen Huang’s return.' The 'Stand with Jensen' petition claims to have collected over 108,000 signatures, including more than 5,600 people identifying themselves as current Nvidia employees. Elon Musk also stated on X that he was signing the petition.

It is necessary to emphasize that this article is speculative analysis by GeekWire editors, not a confirmed succession process. The board has not formally engaged any candidates, and betting market odds reflect market sentiment rather than professional judgment. The value of the article lies in providing an evaluation framework—leadership type rather than technical capability—and the specific facts it reveals regarding talent mobility and competitive dynamics in the AI industry.

However, these terms do not answer a critical question: how Anthropic uses aggregated integration data. The privacy policy acknowledges the company's right to access information about how customers interact with third-party apps and services, but does not explain whether this data influences product decisions. The article concludes that developers accept a trade-off: gaining access to Anthropic’s expanding enterprise customer base comes at the cost of Anthropic controlling the interface, allowing it to introduce competing features without consulting the developers who depend on it.

For developers evaluating whether to build products on frontier models, this article provides not a conclusion but a framework: platform risk depends not on the supplier's current promises, but on its capability structure and incentive direction. Stevenson’s judgment that 'ultimately this is inevitable' is worth incorporating into decision-making considerations more than any specific product release.