LeituraPesquisaRadarFramework de investimento
Entrar / Cadastrar
TudoTecnologiaProdutosInvestimentoBolsaOutros

Leitura

2026-11-0499 posts

Atualizar

PagerDuty Launches Paige AI Agent, Rebrands Operations Cloud to PD Reliability Platform

Tópico · PaigeMaterial
2026-11-03 21:00 GMT+8

Operations management software company PagerDuty launched Paige, an artificial intelligence agent designed for site reliability engineering (SRE) to work as a member of operations and engineering teams, on November 3, 2026. Concurrently, the company renamed its flagship product, previously known as PagerDuty Operations Cloud, to the PD Reliability Platform, and opened PD Labs to provide customers with early access to features under development.

Paige functions as a "virtual responder" within escalation policies. According to CEO John DiLullo, as AI coding tools generate more production software with less human review, system failures are evolving, necessitating a strategy to "fight fire with fire" using AI itself. Teams can configure Paige's permissions: it provides faster triage with remediation held for approval in human-in-the-loop setups, but acts autonomously in explicitly authorized, well-understood, low-risk scenarios. This means some lower-priority incidents may never reach a human responder.

The updated platform includes a new console that integrates incidents, on-call rotations, and AI agents into a single view, allowing incident commanders to monitor which agents are active and their actions. Additionally, PagerDuty introduced three preview experiments via PD Labs: Topology (automated service mapping), Deep Investigation (parallel root-cause hypothesis testing), and SRE Workbench (a collaborative workspace for humans and agents).

PagerDuty claims in a benchmark report that Paige outperformed leading AI coding agents in diagnosing simulated IT incidents; however, this result is self-reported and has not been independently verified. Paige and the PD Reliability Platform are generally available today, while SRE Workbench is open for trial through PD Labs.

Fontes:SiliconANGLE

Pesquisa

Claude Code vs. Codex Code Review: One Pursues Deep Verification, the Other Filters to High-Risk Items Only; Neither Includes a Merge Gate by Default

Visão rápida
2026-11-04 03:02 GMT+8

Anthropic's Claude Code and OpenAI's Codex now both ship with built-in code review capabilities, but their design directions are opposite. Claude Code's managed review service runs multiple parallel agents on Anthropic infrastructure to search diffs and surrounding code, then discards unverified findings through a verification step, deduplicates and ranks them by severity, and finally posts inline comments marked with three tiers (🔴 Important, 🟡 Nit, 🟣 Pre-existing). Codex on GitHub flags only P0 and P1 level issues, deliberately excluding low-priority findings to ensure every comment carries weight.

This difference is directly reflected in how they are configured. Claude Code reads REVIEW.md as review-specific instructions (read from the default branch, so a PR cannot rewrite its own review rules) and CLAUDE.md as project context; Codex looks for the ## Code Review Rules section in the AGENTS.md file closest to the changed code. OpenAI's guidance is to start with two or three rules that encode checks reviewers often explain, state the safe path, and leave formatting and linting to CI.

A commonly overlooked shared limitation: neither prevents merging by default. Claude Code's check run conclusion is always neutral rather than a failure, and the review itself does not approve the PR; Codex only posts comments, leaving enforcement to tests, branch protection, and required approvals. Teams wanting a merge gate need to parse the severity counts from the Claude Code check run output themselves, or rely on Codex comments combined with branch protection rules.

Teams using both agents therefore face configuration file fragmentation: AGENTS.md only guides Codex, REVIEW.md only guides Claude Code, and standards must be synchronized manually. The article suggests a compromise: place rules that are expensive to fix later (architecture, security, forbidden patterns) in the instruction file read by the coding agent, and hand the complete list of team standards to an independent reviewer, keeping the instruction file short enough for the agent to actually follow.

It should be noted that this article comes from Qodo—a vendor providing an independent code review platform. The technical descriptions of the two agent review mechanisms in the text align with their respective product documentation, but framing both as having structural flaws (self-review blind spots, rule file vendor lock-in, context ending at the repository) serves Qodo's commercial positioning to promote its own cross-repository independent reviewer.

Dana Fine is Qodo's Global Community & Open Source Manager, not an engineering or research role. The Qodo Code Review Benchmark cited in the text claims Qodo scored 12 F1 points higher than Claude Code Review across 100 PRs and 580 injected issues; these are vendor-reported results with no independent reproduction.

Fontes:Qodo

Pesquisa

Transaction Structurer Steven Kopits: Iran’s Goal at Hormuz Is to Tax, Not Blockade; Gulf Oil Producers Face Pressure to Increase Output

Tópico · 伊朗霍尔木兹海峡征税策略Material
2026-11-03 10:36 GMT+8

Transaction structurer Steven Kopits offered a specific assessment on Econbrowser regarding Iran’s strategy at the Strait of Hormuz: Iran’s objective is not to block trade but to tax it. He drew an analogy to the 'robber barons' of the medieval Rhine—those who built castles one after another along the river, all sustained by tolls. Kopits believes Iran has signaled that it wants to 'take the money and leave,' rather than create shipping disruptions.

If this assessment holds, it alters the production incentive structure for Gulf oil producers. Kopits projects that Iran may compress the netbacks for Gulf producers to $40–50 per barrel. At this level, profit per barrel is insufficient to sustain the accustomed standard of living in Gulf states, compelling them to increase output to offset the decline in unit revenue. Kopits notes this is not a new pattern—there are historical precedents of OPEC producers increasing output during periods of low oil prices. He suggests Iran could become an 'OPEC of One,' influencing the market by controlling tolls rather than production volumes.

Kopits specifically referenced an open letter recently published by the Islamic Revolutionary Guard Corps (IRGC). He views the letter as coming from 'the institution with the greatest stake in the existing system'; its length and sincerity suggest that internal Iranian leadership understands a structural contradiction: it is impossible to simultaneously be a feared power and a toll-collecting landlord. In his view, this opens a window for reform in Tehran—the Iranians themselves recognize that the current system is outdated and decaying, similar to how Eastern European communist leadership viewed its own systems in the late 1980s.

From a transaction structuring perspective, Kopits argues that the negotiations are not technically difficult. He states that if one has experience designing structures for public offerings, private equity, and venture capital, this counts as a moderately challenging project: funds are ample, parties have incentives to participate, and there exists a framework all can agree to and abide by. The true difficulties are limited to two issues: Iran’s relationship with Hezbollah and Hamas, and the conditions Israel will ultimately accept. Everything else is a matter of valuation and structuring.

It should be noted that Kopits’ assumptions regarding netbacks and his 'OPEC of One' judgment are personal projections, lacking corroboration from official Iranian policy documents. His transaction structuring analogy also presupposes that all parties act as rational economic agents, whereas security fears and ideological commitments in geopolitics often do not follow such logic. Furthermore, his assessment of Iran’s nuclear program—that it is unclear whether nuclear weapons constitute a net benefit for Iran—is likewise an analytical inference rather than a confirmed policy shift.

Fontes:econbrowser.com

Pesquisa

Former Kleiner Perkins Partner Ted Schlein's Ballistic Ventures Files to Raise $490M Amid AI-Driven Cybersecurity Boom

Material
2026-11-04 05:43 GMT+8

According to TechCrunch, Ballistic Ventures, a cybersecurity-focused venture capital firm, has filed with the U.S. Securities and Exchange Commission (SEC) to raise a total of $490 million. The fundraising plan involves $360 million for its third flagship fund and another $130 million for a separate vehicle called "Ballistic Ventures Fund B." Neither fund had recorded any sales at the time of filing.

Ballistic Ventures was co-founded in late 2021 by former Kleiner Perkins partner Ted Schlein, former AT&T Cybersecurity president Barmak Meftah, and Roger Thornton. This move comes as the growing use of artificial intelligence raises concerns about more sophisticated cyberattacks. The rise of AI agents has also created new opportunities for cybersecurity startups and the venture capital firms backing them.

A recent report by JPMorgan Chase, citing PitchBook data, noted that U.S. cybersecurity startups raised $11.5 billion in 2025, the highest since 2022. The report stated that AI-enabled companies accounted for about 72% of U.S. cybersecurity venture deals through May this year, compared with 36% in 2019. Separately, Microsoft's annual Digital Defense Report warned that hackers are increasingly using AI for activities ranging from phishing and social engineering to malware development.

Fontes:TechCrunch

Pesquisa

Y Combinator CEO Garry Tan Reports Spending Up to $200K Annually on AI Agent Tokens

Tópico · Garry TanVisão rápida
2026-11-03 08:35 GMT+8

Y Combinator CEO Garry Tan stated on Walter Isaacson's podcast "The Builders" that he spends approximately $100,000 to $200,000 a year on tokens for his AI agents. Tan described suffering from "cyber psychosis," noting that he often barely sleeps due to excitement about working with his agents.

According to TechCrunch, Tan characterizes AI agents as a "Jiminy Cricket on your shoulder," capable of autonomously handling almost any daily task. His agents have access to his emails, Dropbox files, Apple Notes, and even therapy notes from the past 20 years. He also open-sourced a set of 23 Claude Code skills called gstack on GitHub to demonstrate agent capabilities.

Tan acknowledged that the current cost of personal AI agents is a "crazy privilege" that most people cannot afford. He argued that the future lies in user-controlled personal AI rather than free agents controlled by Big Tech. This disclosure offers a concrete example of the economic and privacy implications of high-level AI agent adoption.

Fontes:TechCrunch

Pesquisa

AI Interconnect Chipmaker Astera Labs Reports Record Quarterly Revenue as Scorpio Switches Become Top Product Line

Tópico · Astera LabsMaterial
Verificado 2026-11-04 05:44 GMT+8

U.S.-based semiconductor connectivity solutions provider Astera Labs (Nasdaq: ALAB) reported financial results for the third quarter ended September 30, 2026, on November 3, 2026. The company posted record quarterly revenue of $584.9 million, representing a 49% increase quarter-over-quarter and a 154% increase year-over-year. This strong growth was primarily driven by the volume production ramp-up of its Scorpio™ fabric switches, which have now become Astera Labs' largest product family by revenue.

In terms of profitability, Astera Labs reported a GAAP gross margin of 72.3%, GAAP operating income of $179.1 million, and GAAP net income of $182.9 million for the third quarter, resulting in GAAP diluted net income per share of $0.99. Non-GAAP metrics showed adjusted operating income of $261.6 million and non-GAAP diluted net income per share of $1.32. CEO Jitendra Mohan stated that design activity accelerated with wins across multiple customers, and highlighted new product introductions including Aries 7, Leo 2, and Orion, aiming to expand dollar content per accelerator and total market opportunity.

For the fourth quarter outlook, Astera Labs estimates revenue between $730 million and $750 million, with a GAAP gross margin of approximately 72%. This guidance suggests continued expansion momentum in the AI infrastructure sector, particularly as its rack-scale reference architecture Orion is on track for availability through partners in Q1 2027, and Aries 7 products are expected to begin sampling to select hyperscale customers in Q4 2026.

Fontes:sec.gov

Pesquisa

AMD Reports Record Q3 Revenue of $13.4 Billion as Data Center Segment Grows 97%; Guides Q4 to ~$17 Billion

Tópico · 台积电财报Material
Verificado 2026-11-04 05:44 GMT+8

U.S. semiconductor firm Advanced Micro Devices (AMD) reported its third-quarter 2026 financial results on November 3, 2026. The company achieved record quarterly revenue of $13.4 billion, a 45% increase year-over-year. The Data Center segment was the primary driver, with revenue reaching $8.5 billion, up 97% year-over-year, fueled by strong demand for AMD Instinct GPUs and EPYC processors.

On a non-GAAP basis, diluted earnings per share rose 74% year-over-year to $2.09, while operating margin expanded to 29%. Dr. Lisa Su, AMD Chair and CEO, stated that growth is expected to accelerate further in the fourth quarter as production scales for Helios rack-scale solutions and 'Venice' CPUs.

Looking ahead, AMD provided an optimistic outlook for the fourth quarter of 2026, expecting revenue to be approximately $17 billion, plus or minus 3%. This midpoint represents a year-over-year growth of about 66% and a sequential increase of roughly 27%. Non-GAAP gross margin is projected to remain steady at approximately 55%.

The report also highlighted strategic moves, including announced agreements to acquire World Labs and Taalas. These acquisitions aim to enhance AMD's model research capabilities and add specialized inference silicon technology to differentiate its system-level solutions.

Fontes:sec.gov

Pesquisa

Arista Networks Reports 58.5% YoY Revenue Growth in Q3, Highest Since IPO; Guides Q4 to $3.9B-$4.0B

Material
Verificado 2026-11-04 05:44 GMT+8

U.S. networking equipment manufacturer Arista Networks released its financial results for the third quarter ended September 30, 2026, on November 3. The report showed total revenue of approximately $3.66 billion, representing a 20.5% increase from the second quarter and a 58.5% increase from the same period last year. This growth rate significantly exceeded the previous guidance of $3.3 billion and marks the highest year-over-year revenue growth since the company's initial public offering (IPO).

In terms of profitability, Arista Networks' GAAP operating margin was 44.9%, and non-GAAP operating margin was 49.1%, both higher than the 42.4% and 48.6% recorded in the same quarter last year. Diluted earnings per share (EPS) were $1.22 on a GAAP basis and $1.20 on a non-GAAP basis, compared to $0.67 and $0.77 respectively in the prior year. CEO Jayshree Ullal stated that this momentum highlights the company's increased relevance in the generative and agentic AI era.

Looking ahead, Arista Networks expects fourth-quarter revenue to range between $3.9 billion and $4.0 billion, with a non-GAAP operating margin of 48-49% and non-GAAP diluted EPS of $1.26 to $1.30. This guidance suggests continued optimism regarding the capital expenditure cycle in the sector.

Fontes:sec.gov

Pesquisa

CNBC Options Columnist Julia Spina: Capturing Volatility Decay with Iron Condor Strategy Ahead of AMD Earnings

Material
2026-11-04 04:40 GMT+8

Over the past week, AMD's stock movement has largely tracked news from other companies. Alphabet suffered a record drop, Amazon reported weak results, and longtime rival Intel significantly lowered its fourth-quarter revenue guidance below consensus expectations and predicted that the PC market slowdown will persist at least through the first half of next year. Intel's stock plummeted 29% on Friday, and AMD fell 10.9% on the same day, followed by another 7.5% decline on Monday. Now, it is time for AMD's own numbers to be judged.

Julia Spina, writing in her CNBC options column, noted that while AMD's stock reactions to the last three quarters' earnings have been mixed, the company remains highly sensitive to the gap between guidance and investor expectations. After Q4 2025 results, the stock fell 17%, despite beating estimates and guiding above them, as some analysts remained disappointed. It rose 18.6% after Q1 2026 results, when guidance exceeded expectations by about 6.5%. However, it fell 7% after Q2 2026 results, where guidance above consensus still failed to meet the $14 billion target desired by some analysts. This unpredictability makes directional calls difficult, but the options market allows investors to construct trades that do not require picking a side.

The strategy she proposes is an iron condor spread. Based on Monday's closing price, she sells the November 4 expiration 425/415 put spread and the 510/520 call spread, collecting a credit of approximately $3.65. The logic is that AMD's 30-day implied volatility is elevated ahead of earnings, sitting above the 81st percentile compared to the past year, meaning sellers are collecting richer premiums than usual. Once earnings are released and uncertainty dissipates, these volatility premiums typically recede. This trade does not bet that options are mispricing the move or predict direction; rather, it aims to profit from the decay of the elevated volatility premium that tends to occur post-earnings.

Specific parameters include placing the short legs at the edges of the expected volatility range. These November 4 weekly contracts imply a move of approximately 9%, consistent with AMD's average close-to-close volatility of 8.4% over its last 10 earnings reports. The long legs are placed $10 outside the short legs to cap maximum capital risk entering the earnings announcement. If AMD closes between $425 and $510 on Wednesday, the strategy achieves its maximum profit of $365. The breakeven points are $421.35 and $513.65, located approximately 9.7% below and 10.0% above Monday's closing price, respectively, just outside the expected move range.

However, Spina honestly outlines the failure boundaries: a maximum loss of $635 occurs if AMD closes below $415 or above $520, representing a move of roughly 11% in either direction. This scenario has occurred in 11 of the last 40 earnings reports, and in the most recent four, moves have reached as high as 19%. This underscores the importance of the long wings; without them, naked short straddles or strangles on a high-volatility stock like AMD could result in losses far exceeding the collected premium.

Fontes:CNBC

Pesquisa

German Startup Orbital Matter 3D Prints First Rigid Beam in Vacuum of Space

Visão rápida
2026-11-04 04:57 GMT+8

German startup Orbital Matter announced that its Replicator-2 satellite successfully 3D printed a half-meter long rigid cylindrical beam using a custom plastic material while in orbit. Company executives described this as a first for a free-flying spacecraft and a significant step toward constructing large structures in space.

The satellite was launched on a SpaceX rideshare mission this summer, carrying four 3D printers. Each printer creates a rigid beam by extruding coils of resin cured by ultraviolet light. Jakub Stojek, CEO of Orbital Matter, stated: "Replicator 2 is the first proof that we can simply build the array in space, at whatever size the mission demands."

This technology is critical for future spacecraft, particularly if they serve as orbital data centers. These facilities will need to generate more electrical power and radiate more heat into the vacuum of space while fitting within the tight confines of rocket fairings. If equipment can be constructed in orbit, power and heat-dissipation challenges may become easier to solve.

Current large structures in space typically use folded deployment methods (such as solar arrays or telescope mirrors), which are engineering-intensive. NASA previously funded Redwire to 3D print structural beams in a vacuum chamber in 2023, but these were not tested in orbit. Orbital Matter's test occurred in the actual space environment.

Although one of two printers designed to deploy a flexible solar array failed due to an anomaly, the successfully printed boom completed the solar array deployment task, serving as an important proof of concept. Previous 3D printing efforts in microgravity were limited to the pressurized environment onboard the International Space Station.

The beam's length is limited by the amount of resin carried onboard, but engineers believe 100-meter booms are feasible. For comparison, SpaceX's Starlink V2 satellites use solar arrays that are 20 meters long. Orbital Matter is partnering with the European Space Agency (ESA) and Thales Alenia Space, planning a larger demonstration mission in 2028 aimed at achieving over 100 kW per wing by 2030.

Fontes:TechCrunch

Pesquisa

NYU Finance Professor Damodaran: The Most Fragile Element in Anthropic's Valuation Is Not the Cost of Capital, but the Trillion-Dollar Revenue Assumption

Tópico · AnthropicEstrutural
2026-11-04 04:11 GMT+8

Aswath Damodaran, a finance professor at the Stern School of Business at New York University, received a large volume of feedback after publishing an intrinsic valuation for Anthropic. In an article posted on November 3, he subjected his own valuation to self-criticism, ranking input variables along two dimensions: the magnitude of their impact on value and their degree of uncertainty. This analysis led to a clear conclusion: the most critical variable driving the $1.45 trillion valuation is not the cost of capital, but his assumption that Anthropic will reach one trillion dollars in annual revenue within ten years. If this revenue target is missed, profit margins will decline, compute assets will sit idle, and the risk of failure will rise, causing more drastic shifts in valuation.

Damodaran selected a cost of capital of 9.74%, which incorporates the cost of equity for high-risk technology companies and the cost of debt based on a BB+ credit rating. This places Anthropic close to the 90th percentile of U.S. companies in 2026 regarding capital costs. In the latter part of his article, he dedicates three full sections to arguing why this figure is not the central controversy of the valuation, and why claims that 12%, 15%, or even 20% are reasonable discount rates contain internal contradictions.

His first check is historical. The U.S. stock data he maintains spans from 1928 to 2025: the average annual return from 2001 to 2025 was 8.74%, from 1976 to 2025 it was 11.92%, and from 1928 to 2025 it was 10.13%. Looking solely at the equity risk premium (the excess return of stocks over government bonds), the premium relative to 10-year Treasury bonds ranges from 5.5% to 5.75%, and relative to short-term Treasuries, it ranges from 7% to 7.5%. Using the 10-year Treasury rate of 4.9% on October 26, 2026, as a baseline, the cost of equity for an average-risk company would be approximately 10.4% to 10.65%. He specifically notes that these historical figures come from the United States—the most successful stock market globally—and suffer from selection bias, potentially overstating future returns; therefore, using a higher discount rate than what history provides is even less defensible.

The second check is market-implied. Damodaran calculates the implied expected return for the S&P 500 monthly: in early October 2026, this figure stood at 8.99%, and it is independent of any specific asset pricing model—whether you believe in CAPM or not, the return implied by current market pricing is 8.99%. He further points out that almost any skepticism regarding cash flows or growth rates would drive this number lower, not higher. Therefore, claiming that the cost of equity for a median U.S. company should be 12% or 15% is equivalent to asserting that you cannot earn that return on an investment of equivalent risk in this market, yet still insisting on that number. Since 1994, the implied expected return on the U.S. stock market has never exceeded 10%.

The third check is consistency with pricing multiples. If one assumes a cost of equity for U.S. companies of 15%, combined with roughly 5% nominal economic growth and an average return on equity of 20%, the market P/E ratio should be only 7.5x. However, the actual forward P/E ratio is closer to 20x. This means you either sell all your stocks or revisit your discount rate assumptions. Similarly, a 12% cost of capital corresponds to enterprise value/operating income multiples of roughly 9x and EV/EBITDA multiples of 6 to 7x, which are far below current market levels. Damodaran concludes: if you claim the cost of equity is 15% while simultaneously arguing that a market P/E of 15x is reasonable, you are being internally inconsistent.

Fontes:aswathdamodaran.blogspot.com

Pesquisa

NVIDIA Open-Sources Distributed AI Workload Failure Attribution Method: Separating Root Causes from Cascading Symptoms

Material
2026-11-03 07:39 GMT+8

In large-scale distributed AI workloads, a hard fault in a single GPU or an NVLink link often leaves traces not just on one node. It can cause collective NCCL communication failures across other ranks within the same workload, trigger state transitions recorded by the scheduler, and prompt monitoring systems to report network metric anomalies. Operations teams face error messages scattered across dozens or even hundreds of nodes, while the actual issue requiring repair might be just a single port on one machine. NVIDIA is open-sourcing its long-used internal failure attribution system under the name NVIDIA Attribution, aiming to transform this troubleshooting process from manual, per-node log reading into reproducible, rule-based diagnostics.

The core design decision of Attribution is workload-centricity rather than node-centricity. It does not merely answer what is wrong with this specific GPU, but answers why this entire workload failed. The scope of analysis is determined by scheduler identifiers or explicit execution windows and node allocations, after which observations related to that workload are collected from application logs, system logs, GPU driver messages, Fabric Manager outputs, scheduler history, and node network metrics.

The attribution process consists of four stages. Stage one defines the workload boundary. Stage two collects observations. Stage three involves the attribution engine extracting events and evaluating their timing and interaction relationships through deterministic rules, separating downstream symptoms from the initial event that caused the failure. Stage four returns structured results, including the workload terminal state, attribution domain, root cause node, list of supporting symptoms, and collector coverage records.

The article provides a synthetic example to demonstrate this workflow. In a multi-rank training workload, application logs contain three distinct NCCL errors, while the system log on node-02 contains XID 149, with its error word decoded as an NVLink signal integrity issue. During the extraction phase, raw logs are converted into typed events with source provenance. In the filtering stage, an EARLY rule named HandleXidBurstEvents reads the XID burst evidence, decodes the error words, and emits an attribution event for signal integrity issues on node-02, leaving the three NCCL events unchanged. In the final refinement stage, the Xid149AndNvlinkFiltering rule retains the signal integrity symptom while demoting symptoms from other nodes. The Gb200PriorityRootCause rule further reduces the active symptom set, resulting in a hardware failure attribution mapping. The NCCL errors are not deleted but preserved as interesting evidence, remaining available for inspection.

A key boundary of this method lies in telemetry quality. The article explicitly states that when multiple causes fit the existing evidence, the attribution domain is marked as ambiguous; diagnostic quality depends on the completeness of underlying telemetry, and the system does not force incomplete evidence into definitive conclusions. Production-scale data—analyzing millions of workloads weekly, covering hundreds of thousands of GPUs, and identifying failure patterns around specific host and switch ports in large RDMA clusters—are self-reported by NVIDIA, with no independent reproduction observed.

Fontes:NVIDIA Technical Blog

Pesquisa
Próxima →
Entrar / Cadastrar
LeituraPesquisaRadarFramework de investimento

If AMD closes between $415 and $425, the short put leg may be assigned, creating an obligation to buy 100 shares per spread at $425. After accounting for the premium received, the effective cost basis is approximately $421.35, which is about 35% lower than the October 6 closing high. For investors bullish on AMD's long-term prospects, this offers a way to acquire the stock at a discount. Similarly, a close between $510 and $520 could trigger assignment on the short call, leaving a short stock position. Spina advises investors who wish to avoid these risks to manage their positions before Wednesday's close.

Spina discloses at the end of the article that she holds this trade. She emphasizes that this is not a judgment on AMD's fundamentals nor a prediction of the earnings result, but rather a neutral trade designed to capture the volatility premium that typically recedes after earnings announcements.

For readers concerned about Anthropic's default risk, Damodaran offers a key distinction: the cost of capital is a going-concern concept and should not absorb all default risk into the discount rate. He estimates Anthropic's total contractual commitments at approximately $450 billion, with a present value of about $373 billion, of which roughly $70 billion comes due next year and $76 billion the year after. The company currently holds $114 billion in cash and is expected to hold over $200 billion post-IPO. He has already embedded a 10% probability of failure in his valuation—that is, scenarios where Anthropic encounters headwinds, fails to meet contracts, and is forced to fire-sale assets. He argues that if your concerns exceed this level, the variable to adjust is the probability of failure, not the cost of capital. In fact, if one ignores failure risk and jams all default concerns into the discount rate, a cost of capital of about 11% would be required to arrive at a valuation similar to his.

Damodaran also criticizes the 20%, 30%, and even 50% target returns commonly used by venture capital and private equity industries. Citing data from Cambridge Associates, he notes that the actual long-term returns of these funds are not only far below their stated targets but also surprisingly close to those of public stock index funds. He believes these fabricated numbers are not merely marketing fluff; they feed back into how much equity founders must surrender and explain why VC and PE investors consistently fail to create value for their limited partners.

In the feedback section, Damodaran tallied the distribution of valuations from readers in a shared spreadsheet: most believed he overvalued Anthropic, with the group consensus closer to $1 trillion rather than $1.5 trillion. Among the dissenters, aside from intuitive judgment, the single largest reason cited was that a 9.74% cost of capital is too low. He expressed disappointment, though not surprise, observing that appraisers and analysts have long treated the cost of capital as a dial to be turned arbitrarily to express optimism or pessimism, lacking an understanding of its economic meaning and real-world constraints.

His final advice is to shift the discussion focus back to what truly matters: Can Anthropic achieve one trillion dollars in revenue within ten years? If this assumption does not hold true, profit margins, reinvestment needs, and risk will all deteriorate in the same direction. For investors dealing with this uncertainty, he outlines four paths—complete ignoring (the gambler approach), complete avoidance (holding only mature assets), setting a margin of safety, and utilizing the law of large numbers to diversify holdings. Which path to choose depends on your risk tolerance and portfolio structure, rather than a single unified discount rate number.

Architecturally, the system comprises two deployable components: the Attribution Service and the Attribution Blueprint. The Service is an HTTP wrapper around the core attribution engine, receiving standardized observations and returning a single verdict. The Blueprint is a Python reference orchestrator responsible for discovering workloads, parsing execution context, running collectors, calling the Service, enriching verdict results, and writing output. The Blueprint includes built-in execution context connectors for Slurm, Kueue, and static inputs, as well as log and metric connectors for Loki, local files, Kubernetes pod logs, and PromQL metrics. Rules are checked into the code repository in declarative configuration and Python formats, released alongside the Service, and are not dynamically uploaded via HTTP API.

For infrastructure teams, the significance of this open-source release lies in the possibility of encoding failure patterns once and reusing them across workloads and clusters. Developers can add rules for new failure modes, implement connectors for their own schedulers and telemetry backends, and route attribution results into incident management, recovery, and fleet management systems. Downstream systems consume the terminal_state and sub_domain fields, eliminating the need to parse product-specific log formats.