Reading

As power density in AI data centers increases, industry teams are exploring and validating 800 volt direct current (800 VDC) as a new power distribution method. DatacenterDynamics (DCD) published a podcast episode titled "Scaling Data Center Power with 800 VDC" on October 28, 2026, featuring contributors from Schneider Electric and NVIDIA who discuss how to design, protect, and test 800 VDC architectures as data center power scales.

The episode highlights several key learning points, including three 800 VDC architectural blueprints (A, B, and C) suited to different use cases, and a system-level approach to DC power distribution. Notably, it addresses personnel safety and arc-flash mitigation in 800 VDC systems, along with proof-of-concept testing on real equipment. According to the publisher's description, lab results showed that traditional circuit breakers cleared faults faster than initially estimated. This specific technical finding offers practical reference value for addressing safety constraints in high-voltage DC environments.

The discussion also covers fault detection, protection coordination, the challenge of defining system boundaries and interfaces, and how collaboration with the Open Compute Project (OCP) and IEEE supports alignment on objectives and standards. It is important to note that Schneider Electric sponsored the episode and contributed insights, so readers should consider potential vendor perspectives when evaluating the architectural recommendations.

Visa, the world's largest international card association, released its latest earnings on October 27, 2026, reporting that full-year revenue for fiscal 2026 (ending September 30) reached $46 billion, setting a new historical record. This performance was driven by sustained growth in consumer spending on travel, shopping, and dining, reflecting the resilience of global consumption momentum. According to Reuters, Visa's fourth-quarter revenue rose 13% year-over-year to $12.1 billion; adjusted earnings per share came in at $3.48, surpassing the market consensus of $3.44.

Transaction data showed that Visa's total volume increased 9% year-over-year in the fourth quarter, with transaction counts up 10%. Notably, cross-border transaction volume (excluding intra-Europe transactions) grew significantly by 14%, boosting interchange fee income. Visa Chief Financial Officer Chris Suh noted that spending growth remained robust across all consumer segments, with high-spending groups growing fastest and no obvious signs of weakness among lower-income consumers. Looking ahead to fiscal 2027, Visa projects full-year revenue growth of 10% to 12%, with adjusted EPS expected to grow 10% to 13%.

Despite both revenue and profit exceeding expectations, investor concerns over rising costs led to stock volatility. The report revealed that Visa's 'Client Incentives'—payments made to partner clients such as issuing banks—increased 21% year-over-year to $5.1 billion in the fourth quarter, higher than market forecasts. The rapid escalation of this key cost item raised fears that future profitability could be compressed. Consequently, Visa's shares fell approximately 1.3% in after-hours trading on October 27.

U.S. tech giant Meta has indicated that it may indefinitely delay the implementation of its controversial new business terms, originally scheduled to take effect on November 20th. This development comes amidst intense backlash from publishers concerned about the company's rights to scrape their content for AI training.

In a post on X, Meta communications head Andy Stone stated that the company does “not plan to make policy changes without a full consultation with industry groups and leaders.” Stone added, “We’ve heard publishers’ feedback on our proposed ads and business terms update... We’re committed to getting this right, and as we expect this consultation could extend beyond November 20, in which case we’d delay implementation of the updated terms. Until then our existing terms remain in place.”

The new business terms were initially set to go into effect on October 30th but appeared to grant Meta permission to freely scrape news sites using Facebook and Instagram for AI model training. Although Meta revised the language to remove the specific word “train,” legal experts told Nieman Lab that the update was insufficient to rule out such usage. Meta had previously delayed the effective date to November 20th to address these concerns.

Beyond opposition from the News/Media Alliance, lawmakers from both U.S. political parties have spoken out. Sen. Marsha Blackburn (R-TN) described the terms as an attempt to “exploit American creators and businesses,” while Sen. Adam Schiff (D-CA) urged Meta to reverse the changes. Additionally, digital advertisers and publishers in the European Union and the UK have called on local regulators to intervene and halt the implementation of the new terms.

Microsoft-owned code hosting platform GitHub announced at its GitHub Universe conference on October 28 that it is now processing 7.4 billion commits per month. This figure represents a significant jump from the 2.9 billion reported in August and far exceeds the 65 million monthly average seen in 2024. GitHub attributes this unprecedented growth primarily to "agentic demand," referring to code submissions generated automatically by AI programming agents.

To support this load, GitHub CTO Vlad Fedorov stated that the company has added 4 million CPU cores, mostly on Azure, along with 160 petabytes of high-speed storage and increased compute power for GitHub Actions. Network capacity was also expanded by 50 Tb/s. These hardware investments aim to alleviate recent service outages caused by traffic surges.

Beyond vertical scaling, GitHub reiterated its commitment to rearchitecting its platform to separate compute and storage. In an early October interview, GitHub COO Kyle Daigle revealed that the migration to Azure is roughly 60% complete. Internal benchmarks for the new architecture indicate up to 35x higher write throughput. This shift underscores that traditional monolithic code hosting architectures are hitting physical limits as AI agents become major participants in software development, necessitating reliance on hyperscale cloud infrastructure capabilities to maintain stability.

Taiwan-based memory manufacturer Macronix Electronics reported its third-quarter 2026 financial results on October 27, 2026. The company posted consolidated revenue of NT$24.237 billion for the quarter, representing a 27% increase from the previous quarter and a substantial 195% rise year-over-year. This performance marks the highest single-quarter revenue in the company's history.

Profitability saw a dramatic improvement, with operating gross margin surging to 70.5%, up from 64.4% in the previous quarter and 13.5% in the same period last year. Approximately NT$750 million of this margin boost was attributed to inventory write-back gains. Operating net margin reached 53.4%, while net income after tax hit NT$10.951 billion, a 1,370% year-over-year increase, resulting in basic earnings per share (EPS) of NT$5.52.

The primary drivers were strong demand for high-density NOR Flash in AI data centers and severe supply shortages in the SLC NAND and eMMC markets. In Q3, NAND Flash (including eMMC) accounted for 47% of total revenue, growing 768% year-over-year. NOR Flash contributed 46% of revenue, up 158% year-over-year, with computer and data center applications making up 47% of NOR Flash sales.

Macronix Chairman Wu Min-chiu stated that limited semiconductor capacity and the long lead time for new equipment investments mean short-term growth relies primarily on price increases and product mix optimization. He highlighted that beyond AI data centers, the rise of satellite and space communications is rapidly boosting demand for eMMC and Flash, positioning the company favorably in this emerging segment.

At the ElevenLabs Summit Bengaluru, Rishabhdhwaj Singh, SVP of Engineering at Indian home services platform Urban Company, played real production call recordings featuring AI relationship managers and AI trainers built on top of ElevenLabs voice models. These demonstrations addressed core pain points in managing large-scale service professional networks.

Urban Company connects thousands of service professionals with millions of customers across India, the UAE, and Singapore. Unable to assign a dedicated human relationship manager to every professional, the team developed AI solutions. The demo included guiding a first-time professional to a customer's home, rescheduling bookings mid-call via backend tool integration, and collecting referrals immediately after reversing a cancellation.

Beyond external relationship management, Urban Company deployed AI trainers to guide technicians through diagnosing faulty water purifier pre-filters and conducting best-practice training sessions, with some calls lasting up to 30 minutes. According to Singh, the platform now handles close to 5 million minutes of such AI calls per month.

Data indicates that the AI relationship manager independently resolves 80% of queries, routing the remaining 20% to humans, and collects most referrals in cities where it is live. Singh also explained why the trainer and the relationship manager use different voices and personas from the ElevenLabs voice library to suit their respective roles and user expectations.

According to Nikkei Asia, major personal computer manufacturers HP, Asus, and Acer are moving some of their manufacturing capacity back to China as skyrocketing memory and storage chip prices, driven by artificial intelligence demand, squeeze their margins. This shift aims to leverage China's mature supply chain ecosystem and relatively lower production costs to offset rising bill-of-materials (BOM) expenses.

Notably, these brands are not returning to traditional contract manufacturers like Foxconn, which have pivoted their capacity toward AI data center servers. Instead, they are signing contracts with smaller Chinese contract manufacturers such as Luxshare, Huaqin Technology, and Longcheer Technology. These emerging players are seeking to establish themselves in PC manufacturing and claim better access to domestic memory chips from suppliers like CXMT.

Previously, many PC brands gradually moved production out of China in 2024 and 2025 to adopt a 'China+1' diversification strategy amid U.S. sanctions and tariffs. However, with AI-induced memory prices climbing 500% in 12 months and costs for other components like PCBs and fiberglass cloth also rising, the economic case for maintaining dispersed production has weakened significantly.

This reversal is impacting newly established supply chains in Southeast Asia. A manager at a PC component supplier revealed that two new production lines built in Thailand for PC clients are now idle as substantial orders move back to China, with no clear timeline for recovery. This suggests that despite ongoing geopolitical risks, immediate economic realities are reshaping the geographic landscape of hardware manufacturing.

Former Yahoo CEO and early Google engineer Marissa Mayer recently found herself in a public relations storm due to an interview. While introducing her newly developed AI assistant, Dazzle, she showcased a feature she considered 'impressive': the AI, which learns from user photo albums, generated three descriptions for her ten-year-old daughter’s school birthday presentation—'empathetic, creative, and loves animals.'

This description resembles a generic horoscope, likely applicable to any ten-year-old child. Progressive activist Leah Greenberg commented that Mayer voluntarily showing this to reporters 'perfectly illustrates how deeply the culture of Silicon Valley has diverged from the rest of humanity.'

However, Silberling’s argument goes beyond mocking a CEO’s PR misstep. She points out that Mayer triggered such significant backlash because the public has been repeatedly conditioned to frustration—from OpenAI CEO Sam Altman stating last year on the Jimmy Fallon show that he 'cannot imagine raising a newborn without ChatGPT,' to his claim months ago on X that a 'cool use case' for ChatGPT Work was generating a podcast to 'explain your child’s interests' (which drew numerous replies asking, 'Why don’t you just talk to your kids?'), to Meta CEO Mark Zuckerberg predicting that outsourcing emotional needs to AI therapists and AI romantic partners will become more socially acceptable.

Silberling suggests that part of the reason behind these executives’ remarks may lie in the tech industry’s efficiency-centric culture—where people optimize sleep with health trackers and inject unregulated 'peptide stacks' to maximize productivity. If you are building electric vehicles or chips, efficiency is indeed a clear direction. But human relationships are not a tech product required to be delivered with minimal capital as quickly as possible.

She cites data from the Pew Research Center as an anchor: 50% of U.S. adults believe AI will weaken people’s ability to form meaningful relationships, while only 5% think AI will make these connections easier. This indicates a huge gap between the 'frictionless future' envisioned by executives and the intuitive judgment of the majority of the public—yet executives tend to view opponents as a small, angry fringe group.

Silberling concludes that human relationships are inherently messy, but it is precisely the investment within this chaos that gives life meaning. When we delegate these interactions to computers, we surrender not only efficiency but also the difficult yet fulfilling experience of being human.

Mayer later defended herself on X, claiming she was merely testing prototype capabilities rather than outsourcing parenting—Dazzle did retrieve photos of 'forgotten kindergarten projects' from the album, such as a 'snake devenomer' and a 'sandcat playground,' showcasing her daughter’s creativity. However, in the same interview, she did propose using AI to plan children’s birthday party themes, arrange playdates, decide on Halloween costumes, and even have her AI assistant coordinate plans with her husband’s AI assistant. When asked if imagining this 'frictionless future planned by AI' was depressing, she replied, 'I haven’t really thought about that.'

San Francisco-based cloud and artificial intelligence security startup Upwind Security announced on October 28, 2026, that it has raised $400 million in a Series C funding round led by Bessemer Venture Partners. This round values the company at $3.8 billion, a significant increase from its $1.5 billion valuation during its Series B in January. Total funding now stands at approximately $830 million.

Beyond the financing details, Upwind disclosed key performance metrics: the company stated that its revenue has quadrupled so far in 2026. This figure offers investors concrete evidence of demand in the niche market for runtime security tools designed to protect cloud workloads, APIs, and AI agents. Upwind's platform aims to address the lack of context in traditional security alerts by unifying signals across these environments.

New investor Canapi Ventures joined the round, alongside returning backers including Cyberstarts, Greylock Partners, Craft Ventures Management, Salesforce Ventures, and Picture Capital. Upwind plans to allocate the new funds toward product development and expanding its go-to-market efforts.

The announcement comes shortly after reports from Israeli business daily Globes, citing The Information, indicated that Upwind had rejected a $4 billion acquisition offer from Datadog Inc. While Upwind did not mention this in its official release, the reported interest underscores the strategic value and potential exit scenarios for companies operating in the rapidly growing AI security sector.

Amazon Web Services (AWS) released a sample SAP modernization agent built on its AI development tool "Kiro" on October 28, 2026. The agent aims to automate the most labor-intensive parts of Enterprise Resource Planning (ERP) system migrations: analyzing, converting, and testing legacy ABAP code to fit the latest SAP S/4HANA systems.

According to AWS's disclosed internal test results, the agent converted 88 programs consisting of 5,854 lines of SAP code in approximately 4 hours and 30 minutes, with all generated unit tests passing. AWS claims this process reduces workload by up to 87% compared to traditional manual methods. It is important to note that these figures come from AWS's own controlled test environment and have not been independently verified in real-world enterprise production settings.

Despite the limitations of self-reported data, the tool has attracted industry attention. Japanese telecom giant NTT Data applied the agent in its new migration service "Neo i-KOU," reporting up to a 95% reduction in ABAP code conversion workload. Consulting firm Capgemini also indicated that using Kiro-based AI agents could shorten the analysis and build cycle for system integration projects by 40% to 60%. Additionally, Toyota Chile utilized the agent to analyze hundreds of existing SAP programs, thereby narrowing the scope of code requiring migration.

The sample agent is currently available via application through the AWS registration page and distributed through GitHub repositories. AWS stated it will collaborate with partners such as Accenture, Deloitte, and Kinandil to support enterprises in their SAP system transformations. For large companies undergoing migration from on-premise deployments to cloud or S/4HANA, such AI-assisted tools may significantly alter the cost structure and project timelines of IT services outsourcing.

Korean AI assessment solution provider Grap announced that its "Precision Glasses Verification" feature for the online exam proctoring system Monito has been adopted by 21 major clients, including large corporations and public institutions, within two months of its release. The feature addresses the growing challenge of smart glasses, which are visually indistinguishable from regular eyewear and evolve rapidly, making them difficult to detect based on specifications alone.

To balance detection efficiency with accuracy, Grap implemented a system of "pre-exam multi-angle capture and direct proctor judgment." Candidates must photograph four specific areas of their glasses—front, side, inner surface, and temple tips—via the Monito app before the exam begins. Online proctors then compare these images against pre-registered photos and anomaly indicators to verify potential misconduct. This approach aims to minimize the risk of false positives associated with standalone AI judgments.

Grap CEO Lim Sung-soo stated that the system was designed from the outset to prevent candidates from being unfairly penalized by technical errors, hence the use of AI-assisted photography to provide clear evidence for proctors. The company plans to update the system in November with a Vision-Language Model (VLM) feature capable of analyzing behavioral context in real-time to detect anomalies such as switching to smart glasses during an exam. Previously, Grap launched the security browser Monito Plus in November last year and integrated an LLM-based agent for summarizing suspicious activities in May this year.

The Australian government's Digital Duty of Care bill passed the House of Representatives on October 28, 2026. The legislation aims to ban AI chatbots that seek to establish 'emotional intimacy' with users and mandates new online safety measures across social media, games, apps, and chatbots to protect Australians from harmful content.

According to TechRadar, Communications Minister Anika Wells stated that the framework will also target content encouraging misogyny, terrorism, suicide, self-harm, and eating disorders. It requires social platforms to provide an easy-to-access 'opt-out' feature for algorithm-driven suggested content. Prime Minister Anthony Albanese emphasized in a speech that stories of people identifying chatbots as their best friends should serve as a wake-up call for parliament to act.

While Labor hopes to pass the bill officially by the end of the year, it faces significant hurdles. The bill is set for debate in the Senate, where the Greens support the proposal but the Coalition and One Nation oppose it, accusing Labor of creating an 'Orwellian' system of censorship. Additionally, Labor rejected an amendment during the lower house debate, increasing pressure to enlist a telecommunications ombudsman as a watchdog for social media and AI.

This move follows similar regulatory actions globally. Months prior, China's Cyberspace Administration (CAC) and four other departments introduced regulations banning AI-generated romantic companions, citing concerns over tools that excessively cater to users or induce emotional dependence.