LeituraPesquisaRadarFramework de investimento
Entrar / Cadastrar
TudoTecnologiaProdutosInvestimentoBolsaOutros

Leitura

2026-11-19129 posts

Atualizar

OpenAI Reports Silicon-Proven PPA Gains in Jalapeño and AMD Designs, Announces Synopsys Partnership for EDA Integration

Material
2026-11-18 08:00 GMT+8

On November 18, 2026, OpenAI published research updates stating that its frontier models, including GPT-6 Astra, have delivered "silicon proven" power, performance, and area (PPA) optimizations in both its custom inference chip, Jalapeño, and AMD's flagship GPU and CPU designs. This follows an initial August disclosure regarding Jalapeño's early results.

For the Jalapeño chip, OpenAI's team estimates that model-proposed changes reduced standard-cell area by approximately 13% in the vector unit and 10% in the matrix unit. Specific examples include a multiplier redesign that cut cell area by about 15% while reducing latency from two clock cycles to one, and an adder redesign that also saved 15% area while maintaining two-cycle latency. These optimizations were verified during Jalapeño's silicon bring-up.

In collaboration with AMD, OpenAI reported that its models achieved up to 14% PPA improvement relative to heavily optimized baselines on AMD's flagship GPU IP by rewriting SystemVerilog RTL code. AMD has incorporated some of these changes into recent manufacturing submissions and added others to production files for future versions. Additionally, AMD verified single-digit percentage PPA gains in key blocks of its upcoming next-generation flagship CPU using designs written by OpenAI's models.

To measure capability progression, OpenAI benchmarked successive model generations on two open-source RISC-V cores: PicoRV32 and Ibex. Results showed that GPT-6.2 Astra achieved the highest average scores at comparable token budgets (~112,000 tokens). Notably, all tested models performed better when designing cores from scratch based on specifications than when optimizing existing RTL, suggesting strong architectural exploration capabilities.

Looking ahead, OpenAI announced a partnership with Synopsys, a leader in electronic design automation (EDA). They are training a new model, GPT-Synopsys, designed to use Synopsys EDA tools as expert chip designers do, aiming to integrate frontier AI directly into established engineering workflows.

Fontes:OpenAI

Pesquisa

Ant Group and Westlake HeartChen Open-Source Duo v1: A Reciprocal World Model Where Virtual Agents Autonomously Decide When to Speak or Listen

Tópico · Duo v1 Reciprocal World ModelMaterial
2026-11-18 23:00 GMT+8

According to a report by Zhidongxi on November 18, Ant Group's Technology Research Institute and Hangzhou-based large model company Westlake HeartChen released and open-sourced the Reciprocal World Model (RWM) Duo v1 on November 17. The model integrates an intelligent agent into the world through two tightly coupled modules: Intent and Realization. It allows virtual agents to continuously receive user audio and video inputs, autonomously decide when to speak or listen, and generate corresponding sounds and visuals in real time for interaction.

Traditional world models typically simulate only physical scenes such as lighting, materials, and dynamics, lacking an internally responsive agent. Duo v1 covers both physical reality and the intelligent agent, sharing the same interactive spatiotemporal space. Within each approximately 0.7-second chunk, the Intent module autonomously decides whether to speak or wait based on user input and interaction history, generating corresponding language content. The Realization module then combines the hidden state from Intent, user input, and its own generated historical content to produce matching audio and video.

In human evaluations across 30 sets of real interactions, Duo v1 passed 168 out of 180 checks, achieving an overall pass rate of 93.3%. Specifically, the pass rates for language adherence, speech intelligibility, and visual identity stability were all 100%, while content responsiveness was 73.3%. Content responsiveness and long-term consistency remain the main challenges for the current version. The technical report, training code, and model weights for Duo v1 are now publicly available for developers to download and build upon.

Fontes:zhidx.com

Pesquisa

Databricks launches Unity Gateway Insights to auto-analyze AI agent costs

Material
2026-11-19 06:42 GMT+8

US data intelligence company Databricks announced the launch of Unity Gateway Insights on November 18, 2026, a new feature designed to automatically analyze millions of AI agent traces alongside business data to identify cost inefficiencies. The tool aims to move beyond simple budget caps by using Genie, Databricks' AI assistant, to pinpoint where tokens are being wasted in enterprise coding workflows.

According to the announcement, trace analysis at Databricks revealed several specific opportunities for savings. For instance, fixing seven small bugs in their Model Context Protocol (MCP) servers recovered an estimated $1.2 million in costs annually. These bugs caused agents to retry failed tool calls repeatedly, leading to significant wasted tokens and engineer waiting time. Additionally, the analysis showed that simple prompts were often routed to expensive frontier models, prompting the adoption of Smart Routing to direct routine tasks to cheaper alternatives.

Unity Gateway Insights integrates with the existing Unity Gateway control plane. When developers use coding agents like Claude Code or GitHub Copilot via the ug CLI, the system automatically captures OpenTelemetry traces into a unified table. The Insights feature then generates reports answering questions such as which workstreams consume disproportionate tokens and which repeated workflows could be codified into shared skills. This allows teams to see exactly how much each type of waste costs them.

Fontes:Databricks

Pesquisa

Neo4j Product Lead: Load Git History as a Graph Database, So Coding Agents Can Find Unreferenced File Coupling in Source Code

Material
2026-11-19 05:00 GMT+8

The security directory in FastAPI contains four modules: api_key.py, oauth2.py, http.py, and open_id_connect_url.py. They do not import each other, and there are no clues in the source code pointing to one another. However, git history reveals that among the 28 commits modifying api_key.py, oauth2.py changed in 18 of them, http.py in 17, and open_id_connect_url.py in 16. Every time someone added behavior to one security scheme—auto_error, description parameter, 401 status code, or WebSocket support—all four modules moved together.

William Lyon, Neo4j’s product lead, used Relate’s git-to-graph extension to import FastAPI’s 8,102 commits into a graph database, then used fs-to-graph to import the working tree, exposing it to coding agents via MCP in read-only mode. He tasked Codex, OpenCode, and Pi with two assignments under two conditions: adding an optional constructor argument to APIKeyHeader, and adding a keyword argument to APIRouter.add_api_route. For each task, the agent had to first answer which other files needed changing, which tests to run, and who should review. When able to read only the source code tree, none of the three agents mentioned the three sibling security modules; after accessing the graph, all three found them and provided specific counts.

This gap is not an isolated case. Lyon queried all file pairs within the FastAPI package that co-changed at least five times, finding 126 pairs, of which 42—about one-third—contained no textual reference to each other in their file contents. This means that regardless of following import chains or searching for filenames, a static source code tree cannot expose this coupling. A graph database turns historical patterns into questions answerable by Cypher queries: What files change along with mine? Has anyone made this kind of change before? Who recently touched these files?

Boundaries to retain: This is a controlled test built, executed, and reported by Neo4j itself, constituting vendor self-testing without independent replication. The repository contained only FastAPI, and the model used was only gpt-6.1-sol. The count of 42 pairs is based on filename matching in text rather than a true import graph, and Lyon himself acknowledges this number is approximate. It proves that for three categories of problems—file coupling, change precedents, and reviewer identification—the answers exist in commit history rather than code text. The source code tree structure cannot provide this information not because it hides it, but because the code itself does not record who changed things together.

Fontes:Neo4j Graph Intelligence Platform

Pesquisa

Anthropic's Quarterly Compute Spending Nears $10 Billion, with Roughly Half Flowing to SpaceX

Tópico · Anthropic Compute & SpaceXMaterial
2026-11-18

Anthropic spent nearly $10 billion on compute and infrastructure in the third quarter of 2026, or close to $40 billion on an annualized basis. This figure comes from the S-1/A filing submitted to the U.S. Securities and Exchange Commission (SEC) for its initial public offering (IPO), marking the first time that a frontier AI laboratory has publicly disclosed compute spending details with quarterly granularity.

More noteworthy is the structure of the expenditure: Anthropic paid approximately $5 billion to SpaceX during the quarter, representing nearly half of its total compute spending. This ratio was confirmed by Bloomberg in late October 2026—Anthropic accounted for 34.5% of SpaceX's total revenue in Q3 2026, corresponding to about $5 billion. This means that Anthropic's reliance on a single supplier far exceeds previous estimates based on annual aggregate data.

In terms of growth trajectory, Anthropic's compute spending rose from $433 million for the full year of 2023 to $9.863 billion in Q3 2026 (annualized at $39.45 billion), representing an average annual growth rate of approximately four times over three years. Epoch AI's comparative analysis shows that OpenAI maintained a roughly threefold annual growth rhythm during the same period: approximately $5.8 billion for the full year 2024, $16.3 billion for 2025, and President Greg Brockman testified in May 2026 that he expected full-year spending to reach about $50 billion.

The absolute scale gap between the two companies is narrowing. In 2024, Anthropic's annualized compute spending was approximately $2.5 billion, only 43% of OpenAI's; by Q3 2026, Anthropic's annualized spending had reached approximately $39.5 billion, approaching 80% of OpenAI's full-year forecast. If Anthropic maintains its current quarterly growth rate, its annualized spending could equal OpenAI's by the end of 2026 or early 2027.

Future commitments indicate that competition will not slow down. Anthropic's S-1 disclosure revealed compute contract commitments of $447.5 billion (from September 2026 through the end of 2030), averaging about $100 billion per year. On the OpenAI side, according to Yahoo Finance, it plans to invest approximately $750 billion over the same period. These figures far exceed current actual expenditures, meaning both companies are locking in supply for compute expansion over the next several years.

For industry observers, this set of data reveals a structural feature that had previously been vaguely handled: the compute spending of frontier AI laboratories is highly concentrated among a few suppliers, and this concentration is still rising. Anthropic's suppliers include Amazon, Google, Microsoft, and SpaceX, but SpaceX alone accounted for nearly half of its spending in Q3 2026. Epoch AI points out that a significant portion of Anthropic's compute spending growth can be attributed to the expansion of transactions with SpaceX, while SpaceX itself is also rapidly entering the compute rental market.

Fontes:Epoch AI

Pesquisa

Analyst Zeus Kerravala: Multi-Availability Zones Prevent Component Failures, Not Total Service Outages

Tópico · Zeus Kerravala Resilience ArchitectureMaterial
2026-11-19 09:21 GMT+8

Between November 12 and 17, 2026, six major IT outages occurred over six days: Microsoft 365 experienced a global outage lasting nearly 24 hours; Workday inadvertently connected its disaster recovery environment to production databases; the U.S. Federal Aviation Administration issued a nationwide ground stop due to router overload; United Airlines delayed flights because it could not obtain aircraft weight data; JPMorgan Chase suffered customer transfer failures due to local hardware and software defects; and Apple experienced outages affecting iCloud and Apple Pay due to an internal update. While the root causes varied—a dormant bug, overloaded routers, a faulty internal update, and local hardware failure—the pattern was identical: a critical dependency failed, and business operations stopped. ZK Research Principal Analyst Zeus Kerravala argues from this that availability architecture and resilience architecture are not the same thing. The former keeps services running when components fail; the latter keeps businesses running when an entire service fails. Most companies possess the former but mistakenly believe they have the latter.

Kerravala points out that cloud computing has caused the industry to forget disaster recovery discipline. In the financial services sector, maintaining both a production data center and a disaster recovery (DR) data center is non-negotiable standard practice. As the world shifted to the cloud, many assumed disaster recovery became the responsibility of the cloud provider; however, it remains a shared responsibility. Availability zones and multi-region designs protect against hardware failures and localized events, but these environments share software, control planes, configuration pipelines, and DNS. When a bad configuration or a dormant bug propagates, it spreads everywhere—as seen with Microsoft’s July network gateway bug, which cascaded across 74 countries and territories.

Kerravala cites Zscaler’s Business Continuity Cloud as a concrete example of what fate separation requires. This environment operates in third-party data centers rather than Zscaler’s own facilities, utilizing independent networks, autonomous system numbers, DNS, and cloud domains. Software and threat intelligence updates are deliberately staggered so that a defective production release does not simultaneously land in the backup. Policy and identity data sync unidirectionally into a read-only, last-known-good state. Spanning nine regions, the architecture is sufficient to maintain domestic operations if a nation were cut off. Zscaler also encourages customers to route synthetic or trusted user traffic to the backup environment continuously during peacetime, validating the failover path daily rather than testing it only during annual tabletop exercises. Failover can be triggered automatically or manually by the customer, ensuring functionality even if Zscaler itself becomes unreachable.

Kerravala’s conclusions are directed at buyers: define the minimum viable business, identifying three to five workflows that must continue running along with all their dependencies; interrogate each key vendor’s continuity architecture—if the answer is 'multiple availability zones,' that indicates availability, not continuity; test continuously rather than relying on annual exercises; keep security capabilities online during crises, as attackers will act immediately when defenses are relaxed; and leverage buyer power at renewal time by making fate-separated continuity a contractual requirement. He acknowledges that Zscaler cannot keep Microsoft mailboxes online or restart FAA routers; its architecture serves merely as a continuity model for its own platform, not a panacea.

Fontes:SiliconANGLE

Pesquisa

Palo Alto Networks Survey: Over 70% of Enterprises Allow AI Agents to Modify Code, but Less Than 10% Can Fully Detect Unmanaged AI Activity

Tópico · AI Agent Identity ManagementMaterial
2026-11-19 08:00 GMT+8

Cybersecurity firm Palo Alto Networks recently published its "2026 State of Cloud-Native Security" report, revealing that while AI agents are ubiquitous in enterprise software development, security measures have failed to keep pace. Nearly all surveyed organizations use AI agents in their software development lifecycle, and more than 70% allow these agents to modify code, adjust Continuous Integration/Continuous Delivery (CI/CD) settings, or make changes to production environments. However, less than 10% of respondents can comprehensively detect and classify unmanaged AI activity across different cloud environments.

The introduction of AI-assisted development has significantly increased the frequency of infrastructure changes. Over 80% of surveyed organizations reported that AI agents lead to more frequent cloud infrastructure modifications, yet fewer than 40% enforce mandatory manual security reviews. Additionally, only about 30% of organizations can track the complete process from code commit to deployment execution for most production workloads. Regarding Infrastructure as Code (IaC), more than 40% of organizations use AI to generate IaC templates, but only about 30% require mandatory security scanning before deployment. Data shows that during a 90-day period, an average of nearly 20% of high-severity application security issues made it into production, with the average time from discovering a code security issue to deploying a patch exceeding 18 days.

Identity management for AI agents presents another critical vulnerability. Nearly 40% of surveyed organizations admitted that their AI development tools or agents can access production cloud environments through development or deployment pipelines. More concerning is that over half of the responding engineers stated that AI agent operations are attributed to the engineer's own identity or personal access tokens, rather than a distinct AI agent identity. This makes it difficult for enterprises to distinguish human actions from machine behaviors during audits, and 40% of organizations cannot reconstruct what operations an AI agent performed, hindering their ability to confirm whether the agent accessed sensitive resources or exceeded its intended scope.

Credential management also suffers from significant gaps. 30% of surveyed organizations discovered that credentials used by AI agents in CI/CD pipelines were not included in enterprise management systems. Over 30% reported difficulty in revoking or rotating credentials used by AI agents during security incident investigations, while 60% still require manual intervention to revoke access rights for Nonhuman Identities (NHI). Data leakage risks are also rising: about 20% of organizations only discover shadow AI after an incident or audit, 40% have found managed data copied to unmanaged AI environments, and nearly 40% have identified sensitive or regulated data being used for AI model training or fine-tuning without authorization.

Fontes:iThome

Pesquisa

Cloudflare Releases Open-Weight Clairic and Clairic-flash Extraction Models with Parallel Slot Autoregressive Architecture

Material
2026-11-19 06:00 GMT+8

US cloud computing company Cloudflare announced on November 18, 2026, that its Workers AI Superintelligence team has released two new open-weight extraction models: an updated version of Clairic and a new lightweight variant called Clairic-flash. Both models are now hosted on Workers AI, with full open weights available for download on Hugging Face.

This release marks Clairic's graduation from preview status. Cloudflare stated that the new version of Clairic features a redesigned architecture using "parallel slot autoregressive generation" instead of the previous single forward pass. In this new mechanism, each field uses tokens it has already written to choose its next token, stopping at its own end marker, while all fields advance together in the serving batch. This change aims to improve accuracy in text and image extraction while maintaining low latency.

In addition to the architectural upgrade, Cloudflare introduced Clairic-flash, a lighter and faster version with 9 billion parameters designed for compact deployment. According to internal evaluation data published by the company, Clairic-flash performs strongly across various extraction tasks; for example, in mixed-input field internal evaluations, Clairic-flash scored 213/226, compared to 156/226 for the older preview model. It should be noted that these benchmarks are vendor self-reported results and have not yet been independently reproduced by third parties.

Regarding pricing, Cloudflare adjusted its "System One" model billing strategy. The 27B-parameter Clairic model is priced at $0.24 per million input tokens, while the 9B-parameter Clairic-flash is priced at $0.10 per million input tokens. To facilitate migration, Cloudflare offers a limited-time promotion: until December 1, 2026, both models will be billed at $0.10 per million input tokens. After this date, Clairic will return to its standard price.

Cloudflare emphasized that despite the architectural changes, the API interface remains unchanged, continuing to return structured information such as confidence scores, character spans, and probabilities. The team stated that releasing open weights is intended to encourage the community to experiment and build upon their ideas, helping developers supplement expensive large language model usage with smaller, faster, and cheaper purpose-built models.

Fontes:Cloudflare Blog

Pesquisa

US and Philippines Plan Framework Agreement for AI Chip and Data Center Hub

Tópico · Pax Silica PhilippinesMaterial
2026-11-19 03:40 GMT+8

According to The Verge, US President Donald Trump is advancing an international initiative called Pax Silica, designed to secure chips and critical minerals used in AI hardware. Facing growing opposition at home, the plan is turning its eyes to the former US colony of the Philippines to build chip fabs and data centers.

Philippine and US officials are expected to sign a framework agreement by the end of 2026 for the jointly developed Pax Silica Economic Security Zone. Located at the former Clark Air Base, the project spans approximately 1,620 hectares (4,000 acres) and plans to house chip factories, processing plants for critical minerals, and data centers. The Philippines' Bases Conversion and Development Authority (BCDA) oversees the plan, expecting an economic boom from minerals processing and chip manufacturing, and potentially attracting tech giants like Nvidia, TSMC, and Apple.

This move reflects a geopolitical shift in global AI infrastructure placement. With Americans increasingly opposed to data centers and chip factories being built in their communities, the US government is recruiting less affluent nations to take up some of the slack. Through Pax Silica, the US aims to insulate itself from China's dominance in manufacturing semiconductors and processing critical minerals needed to make those chips. At least two dozen countries have signed on to the initiative, and Trump was expected to pressure more to join during his visit to the Philippines.

However, the project has raised serious human rights and environmental concerns. Human Rights Watch warns that the plan will displace Indigenous residents, specifically the Aeta Hungey tribe, cutting them off from water resources they rely on. Petronila Capiz-Muñoz, chieftain of the Aeta Hungey tribe, stated: "We do not want to give up our community... How do we give up our burial grounds, our church, our livelihood?" Additionally, an American energy company is conducting a feasibility study into building a liquefied natural gas line and plant to meet the hub's enormous electricity demand. While LNG is touted as a clean fuel, its primary component methane is an even more potent greenhouse gas than carbon dioxide.

Details of the project remain murky and shrouded in non-disclosure agreements, exacerbating public distrust. The BCDA estimates the AI hub will create up to 190,000 direct jobs, but critics point out that historically, chip manufacturing moved to Asia where similar health problems stemming from toxic chemicals cropped up decades later, raising fears that history might repeat itself.

Fontes:The Verge

Pesquisa

Microsoft CEO Nadella Explains 60K Buyout Plan: A Shift to Managing AI Agents, Not Replacement

Material
2026-11-19 08:42 GMT+8

Microsoft CEO Satya Nadella explained at the New York Times Dealbook Summit why the company offered voluntary buyouts to approximately 60,000 employees. He emphasized that the move is not about replacing software engineers with artificial intelligence, but rather reorganizing Microsoft into a company where everyone uses AI to perform their work, shifting their focus to managing AI agents.

Nadella argued that software engineering is thriving but has reached a new operating point. He cited the team behind Cowork, an AI agent in Microsoft 365 apps, which consists of just 11 developers who build a "software factory" that produces the product, rather than coding it directly. This model requires employees to direct and coordinate AI capabilities. For those struggling with this structural change, the buyout offers a way to retain "agency" in choosing their path.

While previous reports from The Verge highlighted the plan as a method to cut labor costs while sustaining heavy AI infrastructure investment, Nadella's latest comments add an organizational philosophy layer. He acknowledged the anxiety caused by the AI revolution and stated that structural changes are necessary to help employees regain control over their careers. As Microsoft's global headcount declines year-over-year for the first time in a decade, this shift signals a broader trend among tech giants moving from labor-intensive growth to capital-intensive, leaner operational models centered on AI management.

Fontes:TechCrunch

Pesquisa

Qdrant Team Benchmarks: Retrieval Quality Surpasses 0.6B Original Precision After Compressing Code Embedding Model from 4B to 1 Bit

Material
2026-11-18 08:00 GMT+8

The Qdrant team conducted an independent evaluation of Papr's papr-embed-code-1 code embedding model, separate from the model card. The model is based on Qwen3-Embedding and was further trained on code retrieval data, offering two sizes (0.6B and 4B) with open weights. The core finding of the evaluation is not merely that code fine-tuning is effective—both sizes outperformed their base models on code tasks they had not been trained on (HumanEval, MBPP, FreshStack, WikiSQL), with the 4B gaining an average of 0.034 nDCG@10 and the 0.6B gaining 0.026—but rather that quantization alters the logic for choosing model size.

The 4B model has a vector dimension of 2560, storing approximately 10 KB per vector in float32. Using Qdrant's 1-bit TurboQuant compression, each vector requires only 328 bytes, a reduction of 31 times. Across seven code retrieval tasks, 1-bit compression retained 99.7% of the 4B model's float32 score without rescoring, and all precision with rescoring at 2x oversampling. This means the compressed 4B model achieves higher retrieval quality than the uncompressed 0.6B (0.877 vs. 0.838 nDCG@10), while consuming only one-twelfth of the RAM. Based on this, the authors recommend: if embedding compute allows, select the 4B model stored at 1 bit, enable rescoring, and prioritize increasing hnsw_ef over adding quantization bits.

This conclusion has clear boundaries. On prose retrieval tasks, code fine-tuning actually caused both sizes to score slightly lower than their base models (the 0.6B dropped by 0.003, the 4B by 0.016); on TREC-COVID, the 0.6B even dropped by 0.065. BM25 hybrid search failed to provide gains across the tested query types (questions, descriptions, code); equal-weight fusion decreased performance on all tasks by 0.02 to 0.17, and even reducing the BM25 weight to 0.1 only brought results close to parity. The entire evaluation was completed by the Qdrant team on its own infrastructure, constituting vendor self-testing, and has not yet been independently reproduced.

Fontes:qdrant.tech

Pesquisa

Microsoft CEO Nadella Predicts AI Agents Will Reshuffle Web Economics, Warns Rejectors Like Amazon Face Disadvantage

Tópico · Microsoft Nadella AI Agent EconomyMaterial
2026-11-19 06:58 GMT+8

Microsoft CEO Satya Nadella stated at The New York Times' DealBook Summit that the rise of AI agents will reshape the economics of the internet, impacting business models reliant on advertising and "human web" traffic.

Nadella argued that current aggregators monetizing human eyeballs will face disruption. As traffic shifts to AI agents, revenue models may move from display ads to cost-per-click (CPC) or direct transactions. "If you don't accept the agent, somebody else will," he said.

He cited Amazon's decision to block AI agents from its retail site as an example of protecting legacy models. Nadella suggested such companies could lose ground to platforms like Shopify that embrace agentic commerce, highlighting openness as a key competitive advantage in the AI era.

Fontes:TechCrunch

Pesquisa
Próxima →
Entrar / Cadastrar
LeituraPesquisaRadarFramework de investimento