# capacity Published articles for capacity. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers DevFeed: [Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers]() Original publisher: [Read original article]() Author: Josh Parker Published: 2026-09-16T13:00:33Z Content type: news Language: en Sources: [NVIDIA Blog]() Topics: [data centers](), [AI Infrastructure](), [Google](), [Nvidia](), [Artificial Intelligence](), [datacenter](), [benchmarking](), [Requirements]() Tags: [ai](), [ai-data-centers](), [ai-infrastructure](), [capacity](), [corporate](), [data-centers](), [energy](), [google](), [infrastructure](), [launch](), [metrics](), [nvidia](), [performance-metrics](), [requirements](), [resource](), [systems]() ### AI overview Emerald AI, Google and NVIDIA announced the AI Energy Management Alliance, a coalition focused on flexible AI data centers that can dynamically adjust electricity use in response to grid conditions. The article describes technology-neutral, performance-based requirements covering response speed, duration, predictability and emergency behavior. ### Source excerpt AI factories are the infrastructure of the intelligence era. Scaling them responsibly will depend as much on innovation across the grid as inside the data center. Today, Emerald AI, Google and NVIDIA announced the launch of the AI Energy Management Alliance (AEMA), a first-of-its-kind coalition advancing data centers that can dynamically manage their electricity use [...] ## Dropbox Outlines How Focusing on Existing Infrastructure Efficiency Can Create Headroom for AI DevFeed: [Dropbox Outlines How Focusing on Existing Infrastructure Efficiency Can Create Headroom for AI]() Original publisher: [Read original article]() Author: Matt Foster Published: 2026-09-16T07:15:00Z Content type: news Language: en Sources: [InfoQ]() Topics: [Artificial Intelligence](), [cloud-infrastructure](), [Magic Pocket](), [data centers](), [networking]() Tags: [ai](), [ai-architecture](), [ai-ml-data-engineering](), [capacity](), [cloud](), [data-center](), [data-storage](), [devops](), [dropbox](), [dropbox-datacenter](), [efficiency](), [infrastructure](), [infrastructure-optimisation](), [magic-pocket](), [ml-data-engineering](), [networking](), [news](), [optimization](), [storage](), [sustainable-computing]() ### AI overview Dropbox describes how long-running infrastructure optimization helps it accommodate growing AI demand by improving forecasting, fleet utilization, storage density, hardware lifecycles, and rack-level power delivery. Its storage infrastructure has used more than 50% less power per petabyte since 2020. ### Source excerpt Dropbox has outlined how a decade of infrastructure optimization is helping it absorb growing demand from AI without treating new data-center capacity as the only answer. Its work spans forecasting, fleet utilization, storage density, hardware lifecycles, and rack-level power delivery, much of it predating the current AI boom. By Matt Foster ## Micron Shows off 512GB DDR5 RDIMM: 12TB per Dual-Socket Server at 9,200 MT/s, Volume Production in 2H 2027 DevFeed: [Micron Shows off 512GB DDR5 RDIMM: 12TB per Dual-Socket Server at 9,200 MT/s, Volume Production in 2H 2027]() Original publisher: [Read original article]() Author: Brian Beeler Published: 2026-09-15T20:18:17Z Content type: news Language: en Sources: [StorageReview.com]() Topics: [ddr5](), [servers](), [intel](), [cpu]() Tags: [capacity](), [ddr5](), [enterprise](), [generation](), [intel](), [memory](), [modules](), [performance](), [production](), [release](), [server](), [speed](), [volume]() ### AI overview Micron demonstrated a 512GB DDR5 RDIMM rated for up to 9,200 MT/s. The module can provide 12TB of memory in a 24-slot dual-socket server, with volume production scheduled for the second half of 2027. AMD and Intel are validating it for next-generation server platforms. ### Source excerpt Micron has demonstrated a 512GB DDR5 RDIMM running on multiple server platforms, which it calls the world's first module at that capacity, and says AMD and Intel are both validating it for their next-generation server platforms. The module is rated for speeds up to 9,200 MT/s, and in a 24-slot dual-socket server it puts 12TB The post Micron Shows off 512GB DDR5 RDIMM: 12TB per Dual-Socket Server at 9,200 MT/s, Volume Production in 2H 2027 appeared first on StorageReview.com. ## From Megawatts to Tokens: How NVIDIA Maximizes AI Factory Production DevFeed: [From Megawatts to Tokens: How NVIDIA Maximizes AI Factory Production]() Original publisher: [Read original article]() Author: Vishal Ganeriwala Published: 2026-09-15T16:55:59Z Content type: article Language: en Sources: [NVIDIA Blog]() Topics: [AI Factory](), [NVIDIA DSX](), [datacenter](), [Orchestration](), [Inference](), [High-Performance Computing]() Tags: [ai](), [ai-factory](), [ai-infrastructure](), [capacity](), [compute](), [dsx](), [efficiency](), [inference](), [nvidia](), [orchestration](), [production]() ### AI overview The article describes how Emerald AI's Conductor platform responds to utility demand signals by adjusting flexible data-center workloads while keeping high-priority AI inference running. It also reports that Lambda's validation found a fixed power budget could support 24% more token throughput when managed intelligently. ### Source excerpt On a sweltering August evening in Silicon Valley, as the sun dropped and air conditioning loads spiked, Silicon Valley Power sent a signal to an AI factory to adjust its power consumption. Varun Sivaram was watching on Zoom with about forty others -- his team at Emerald AI in their San Francisco conference room, engineers [...] ## How We Built Automated Capacity Testing for Kafka Consumers DevFeed: [How We Built Automated Capacity Testing for Kafka Consumers]() Original publisher: [Read original article]() Author: Kaan Karakaya Published: 2026-09-14T09:46:34Z Content type: tutorial Language: en Sources: [Booking.com Development - Medium]() Topics: [Kafka](), [Testing](), [HTTP]() Tags: [automated](), [capacity](), [health-checks](), [kafka](), [load](), [load-balancer](), [parallelism](), [partition](), [recovery](), [scale](), [site-reliability-engineer](), [sre](), [testing]() ### AI overview This article describes Booking.com's extension of an existing capacity-testing platform for Kafka consumers. It explains how changing partition assignment can provide a controlled, measurable way to test consumer throughput and whether remaining consumers can absorb reassigned work after an instance or failure domain disappears. ### Source excerpt Photo by GuerrillaBuzz on Unsplash Kafka makes it easy to distribute work across consumer instances. It is much harder to prove, safely and repeatedly, how those instances behave when the distribution changes and one of them has to carry more than its usual share. For teams that run Kafka at scale, this is a practical reliability question: how much load can a consumer instance actually handle? We had automated capacity testing for HTTP services, but Kafka consumers were still tested with manual drills. Those drills could tell us something, but they were disruptive, difficult to reproduce, and risky precisely when the system was close to its limit. We wanted a controlled way to answer three questions: What is the maximum sustainable throughput of a consumer instance? If an instance or failure domain disappears, can the remaining consumers absorb the reassigned work? Are we overprovisioning resources because we do not know the real limit? The result was an extension to our capacity-testing platform that turns Kafka partition assignment into a safe, measurable load-control mechanism. Why HTTP capacity testing did not translate Our existing platform was designed for request-response services behind a load balancer. A scheduled test selects one instance, routes an increasing share of traffic to it, runs health checks after each step, and records the highest ratio the instance can sustain. After the test, traffic returns to its normal distribution and the result is reported to the service owner. Kafka has no equivalent traffic knob. Consumers pull records, and the unit of parallelism is the partition. Within a consumer group, each partition is owned by one consumer at a time. If a topic has 12 partitions and four equally loaded instances, each instance owns about three. When one instance disappears, a rebalance gives the survivors more partitions -- and the extra work arrives as a step change, not as a smooth increase from a load balancer. The key translation: for an HTTP ## How we shipped 15 Tbps for OpenAI in 90 days (Session 2 of 3) DevFeed: [How we shipped 15 Tbps for OpenAI in 90 days (Session 2 of 3)]() Original publisher: [Read original article]() Author: Sridhar Rajarao Published: 2026-09-12T00:00:00Z Content type: article Language: en Sources: [Sridhar Rajarao]() Topics: [OpenAI](), [Network](), [GPU](), [Server](), [Cache](), [API](), [Oracle Database]() Tags: [api](), [architecture](), [build](), [cache](), [capacity](), [database](), [debugging](), [execution](), [gateway](), [gpu](), [network](), [object](), [openai](), [performance](), [platform-engineering](), [server](), [servers](), [sre](), [storage](), [testing](), [warp]() ### AI overview The second session describes turning an architecture for OpenAI's 15 Tbps system into a delivery plan. It covers coordinated capacity planning across network, gateway, server, storage, and database teams; caching object names through the Inventory API; delivery tracking; and performance validation. Early WARP testing found packet drops caused by an unsuitable MTU of 1500, which was changed to 9100. ### Source excerpt Architecture was only the first week. Session 2 is about the build: capacity, execution discipline, and the first signs that performance would be the real test. ## How Load Balancers Actually Distribute Traffic DevFeed: [How Load Balancers Actually Distribute Traffic]() Original publisher: [Read original article]() Author: Ashish Pratap Singh Published: 2026-09-09T12:17:03Z Content type: tutorial Language: en Sources: [AlgoMaster Newsletter]() Topics: [Load Balancing](), [load balancing algorithms](), [systems](), [Server]() Tags: [capacity](), [load-balancing](), [load-balancing-algorithms](), [nginx](), [server]() ### AI overview This tutorial explains how load balancers distribute requests across multiple servers using algorithms such as round robin and weighted round robin. It describes their simplicity, health-check behavior, and limitations when servers differ in capacity or handle long-running requests. ### Source excerpt When your application runs on multiple servers, you need a way to distribute incoming requests across them. ## Hot Chips 2026: XCENA and Samsung's Near-Memory Compute CXL Device DevFeed: [Hot Chips 2026: XCENA and Samsung's Near-Memory Compute CXL Device]() Original publisher: [Read original article]() Author: Chester Lam Published: 2026-08-30T07:25:37Z Content type: article Language: en Sources: [Chips and Cheese]() Topics: [samsung](), [ddr5](), [RISC-V](), [data](), [Cache](), [Arm](), [intel]() Tags: [2026](), [arm](), [cache](), [capacity](), [clusters](), [compute](), [core](), [ddr5](), [dram](), [intel](), [memory](), [parallel](), [pcie](), [performance](), [risc-v](), [samsung]() ### AI overview The article examines XCENA and Samsung's MX1, a CXL memory expansion device that can host up to 2 TB of DDR5 memory, connect SSDs, and provide onboard compute through 3,072 RISC-V cores. It describes the device's memory bandwidth, cache hierarchy, power use, and focus on data-parallel workloads. ### Source excerpt CXL memory expansion, with a side of compute ## AI data centers: the five hard problems money cannot buy away DevFeed: [AI data centers: the five hard problems money cannot buy away]() Original publisher: [Read original article]() Author: Sridhar Rajarao Published: 2026-08-30T00:00:00Z Content type: article Language: en Sources: [Sridhar Rajarao]() Topics: [Artificial Intelligence](), [data centers](), [Hardware](), [systems]() Tags: [ai](), [capacity](), [compute](), [data-centers](), [datacenters](), [hardware](), [heat](), [infrastructure](), [power](), [reliability]() ### AI overview The article argues that AI data-center expansion is constrained by the simultaneous need to secure power, grid connections, cooling and water strategies, equipment, skilled workers, permits, and productive compute. It discusses five industry challenges, with the supplied text covering power constraints and the physical and water implications of liquid cooling. ### Source excerpt The AI buildout is not mainly a real-estate problem. It is a race to integrate power, cooling, equipment, permits, and useful compute at the same time. ## What N, N+1, 2N, and 2(N+1) Mean in a Data Center DevFeed: [What N, N+1, 2N, and 2(N+1) Mean in a Data Center]() Original publisher: [Read original article]() Author: j2sw Published: 2026-08-27T13:59:08Z Content type: tutorial Language: en Sources: [Justin Wilson (j2sw)]() Topics: [datacenter](), [data]() Tags: [2n-redundancy](), [capacity](), [colocation](), [components](), [cooling](), [data-center](), [data-center-power](), [equipment](), [internet-architecture](), [maintenance](), [n-plus-1-redundancy](), [network-operations](), [power-distribution](), [system](), [ups]() ### AI overview This article explains what data center redundancy ratings N, N+1, N+2, 2N, and 2(N+1) indicate about component capacity, failures, maintenance, cooling, and alternate paths. ### Source excerpt Have you ever looked at a data center marketing slick and wondered what N, N+1, 2N, and 2(N+1) actually mean? The main thing they tell you is how many components can fail before the system drops below the capacity needed to support the load. When I evaluate a data center, I want to know what ... Read more The post What N, N+1, 2N, and 2(N+1) Mean in a Data Center appeared first on Justin Wilson (j2sw). ## Day to day PostgreSQL on Kubernetes: Resize, Scale, and Upgrade DevFeed: [Day to day PostgreSQL on Kubernetes: Resize, Scale, and Upgrade]() Original publisher: [Read original article]() Author: Hans-Jürgen Schönig Published: 2026-08-25T06:16:00Z Content type: tutorial Language: en Sources: [CYBERTEC PostgreSQL | Services & Support]() Topics: [Kubernetes](), [PostgreSQL](), [postgresql clusters]() Tags: [capacity](), [commands](), [declarative](), [github](), [how-to](), [kubectl](), [kubernetes](), [operations](), [patroni](), [pg-operator](), [postgresql](), [postgresql-on-cloud](), [pull-requests](), [rolling-update](), [scale](), [upgrade]() ### AI overview A tutorial demonstrates resizing, scaling, and minor version upgrades for PostgreSQL running on Kubernetes with the CYBERTEC PG Operator. Using end-to-end minikube tests, it shows declarative manifest changes, rolling updates, Patroni failover, and scaling while the cluster remains available. ### Source excerpt This blog is about the resizing, scaling PostgreSQL in Kubernetes. This blog helps you learn various operation related to operations of PostgreSQL in Kubernetes. The post Day to day PostgreSQL on Kubernetes: Resize, Scale, and Upgrade appeared first on CYBERTEC PostgreSQL | Services & Support. ## Hot Chips 2026: Samsung and HBM Base Die Opportunities DevFeed: [Hot Chips 2026: Samsung and HBM Base Die Opportunities]() Original publisher: [Read original article]() Author: Chester Lam Published: 2026-08-23T19:01:23Z Content type: news Language: en Sources: [Chips and Cheese]() Topics: [samsung]() Tags: [capacity](), [cpu](), [intel](), [machine-learning](), [memory](), [samsung]() ### AI overview The article discusses Samsung's HBM base-die design. Samsung moved HBM4 and HBM4E base dies to a 4nm logic process, creating additional usable area and enabling potential optimizations such as moving the memory controller onto the HBM base die. ### Source excerpt Machine learning applications demand ever more memory capacity and bandwidth. Samsung responded by fabricating their HBM base dies on a logic node, which opens up more opportunities ## Why GitHub feels less reliable lately DevFeed: [Why GitHub feels less reliable lately]() Original publisher: [Read original article]() Author: Sridhar Rajarao Published: 2026-08-23T00:00:00Z Content type: opinion Language: en Sources: [Sridhar Rajarao]() Topics: [GitHub](), [incident](), [migration](), [systems](), [Azure](), [GitHub Actions](), [pull-requests]() Tags: [autoscaling](), [capacity](), [concurrency](), [data-centers](), [deployment](), [github](), [incident](), [incident-management](), [istio](), [reliability](), [request](), [resilience](), [sre](), [transformation]() ### AI overview The article argues that GitHub's recent reliability problems reflect the difficult middle of a major infrastructure transformation. It connects incidents to migration complexity, unsafe automation, configuration mistakes, capacity and concurrency weaknesses, database migration errors, and autoscaling problems. ### Source excerpt GitHub is not having one outage problem. Its recent incident reports show the difficult middle of a platform transformation. ## Concurrent Servers: Part 8 - Go DevFeed: [Concurrent Servers: Part 8 - Go]() Original publisher: [Read original article]() Author: Eli Bendersky Published: 2026-08-22T14:52:00Z Content type: tutorial Language: en Sources: [Eli Bendersky]() Topics: [Go Language](), [Concurrency](), [Server](), [Concurrent Programming](), [Network]() Tags: [capacity](), [concurrency](), [concurrent](), [go](), [goroutines](), [misc](), [network-programming](), [servers](), [threads]() ### AI overview Part 8 of a series on concurrent network servers explains how Go implements sequential and concurrent servers. It demonstrates serving each client with a lightweight goroutine and discusses why concurrency may still need to be limited, including when tasks compete for finite CPU capacity. ### Source excerpt This is part 8 in a series of posts on writing concurrent network servers. In this part, we'll switch to Go and see how it tackles the challenges described earlier in the series. All posts in the series: Part 1 - Introduction Part 2 - Threads Part 3 - Event-driven Part 4 - libuv ... ## Solana Ecosystem Roundup: July 2026 DevFeed: [Solana Ecosystem Roundup: July 2026]() Original publisher: [Read original article]() Author: Solana Foundation Published: 2026-08-05T09:33:00Z Content type: news Language: en Sources: [Solana News Feed]() Topics: [Solana](), [Solana ecosystem](), [Finance](), [linux foundation](), [stripe]() Tags: [2026](), [blockchain](), [blockchain-technology](), [capacity](), [crypto-news](), [cryptocurrency](), [defi](), [ecosystem](), [finance](), [intel](), [linux-foundation](), [nfts](), [nvidia](), [payments](), [podcasts](), [recap](), [solana](), [solana-ecosystem](), [stripe](), [tsmc](), [web3]() ### AI overview Solana's July 2026 ecosystem roundup reports a 66% increase in maximum block capacity, growth in tokenized equities and real-world assets, expanded payments adoption, and $22.9 million in Solana-processed World Series of Poker buy-ins. ### Source excerpt Tokenized equities expanded, real-world assets hit $3.73B, payments reached 330,000+ merchants, and block capacity increased 66% in a milestone month. ## History of SpaceX: The Category-Dominating Commercial Spinoff and the Internalization of Anchor Demand DevFeed: [History of SpaceX: The Category-Dominating Commercial Spinoff and the Internalization of Anchor Demand]() Original publisher: [Read original article]() Author: Brendan Sechter Published: 2026-08-04T09:00:00Z Content type: article Language: en Sources: [Brendan A R Sechter's Development Blog]() Topics: [communications](), [Deployment](), [service]() Tags: [aerospace](), [beta](), [business](), [capacity](), [customer](), [deployment](), [history](), [international](), [launches](), [operational](), [partnership](), [regulatory](), [revenue](), [service](), [spacex](), [subscriber]() ### AI overview This article examines SpaceX's commercial spinoff as the internalization of an anchor customer. It explains how the spinoff consumes the parent's launch output at marginal cost, traces its deployment, service rollout, integration, subscriber and revenue development, direct-to-cell expansion, capital requirements, and regulation, and compares the configuration with other satellite-constellation businesses. ### Source excerpt This article is the eleventh in the History of SpaceX series and the third and last treating the capital-formation legs that the series opener introduced. The category-dominating commercial spinoff concerns the business the venture built on top of its own capability, and the article's organizing claim is that the spinoff is not a diversification into an adjacent market but the internalization of an anchor customer. Where the Anchor Demand article A283 treats a government customer buying launches, this article treats the venture becoming the customer it had previously needed someone else to be. The decisive economic property is not that the spinoff grew large. It is that the spinoff consumes the parent's output at marginal cost while every competitor attempting the same business must pay a market price the parent sets. The article walks the January 2015 announcement and the capacity argument that motivated it, the deployment sequence from the first operational batch of May 2019 through the service beta of 2020 and the commercial rollout of 2021, the vertical integration and the internal transfer price that the whole arrangement turns upon, the coupling between constellation deployment and launch cadence, the subscriber and revenue trajectory across the 2020 through drafting-date period, the direct-to-cell extension beginning with the carrier partnership announced in 2022, the capital intensity and the replenishment obligation that a short-lifetime constellation imposes, and the regulatory position across the Federal Communications Commission, the International Telecommunication Union, and the national regulators whose authorizations the service requires. The article contrasts the configuration against the Iridium and Globalstar precedents, in which comparable constellations were built without a captive launch capability, and against the OneWeb and Kuiper cases, in which competitors attempted the business while buying launch at market. The article closes with an expli ## Xiaomi Smart Storage NAS Revealed DevFeed: [Xiaomi Smart Storage NAS Revealed]() Original publisher: [Read original article]() Author: Rob Andrews Published: 2026-08-03T16:00:11Z Content type: article Language: en Sources: [NAS Compares]() Topics: [Hardware](), [Specifications](), [QNAP](), [Synology]() Tags: [2-bay-nas](), [capacity](), [china](), [consumer](), [crowdfunding](), [hardware](), [home-nas](), [hyperos](), [launch](), [leak](), [nas](), [nas-2026](), [nas-news](), [network-attached-storage-nas](), [news](), [news-2026](), [private-cloud](), [product](), [qnap](), [qnap-2026](), [qnap-alternative](), [qnap-news](), [realtek-rtd1619b](), [specifications](), [storage](), [synology](), [synology-2026](), [synology-alternative](), [synology-news](), [uncategorised](), [xiaomi](), [xiaomi-nas](), [xiaomi-smart-storage](), [xiaomi-youpin]() ### AI overview Xiaomi Smart Storage is a two-bay consumer NAS launched in China through Xiaomi Youpin crowdfunding. The article describes its family-focused storage use cases, bundled-drive pricing, and leaked engineering-sample hardware specifications, while distinguishing official details from unconfirmed specifications. ### Source excerpt Xiaomi has now officially moved into the home NAS market with Xiaomi Smart Storage, a two-bay consumer storage box that is being launched in China through Xiaomi Youpin crowdfunding. The product is being positioned less like a traditional enthusiast NAS and more like a family storage appliance for phone backup, photo management, files, and home [...] ## AMD Instinct MI455X GPU Detailed for Rack-Scale AI Deployments DevFeed: [AMD Instinct MI455X GPU Detailed for Rack-Scale AI Deployments]() Original publisher: [Read original article]() Author: George Cozma Published: 2026-07-23T17:37:10Z Content type: article Language: en Sources: [Chips and Cheese]() Topics: [GPU](), [Artificial Intelligence](), [networking](), [NVLink](), [systems](), [Ethernet](), [SOC](), [tsmc]() Tags: [2026](), [ai](), [architecture](), [article](), [capacity](), [communication](), [compute](), [ethernet](), [gpu](), [memory](), [networking](), [performance](), [processors](), [scale](), [soc](), [tsmc]() ### AI overview The article examines AMD's Instinct MI455X, a CDNA5 GPU designed for rack-scale AI deployments. It describes the Helios rack-scale system, UALink over Ethernet networking, compute and memory specifications, and changes to the WGP, register-file, and matrix-unit designs. ### Source excerpt Editor's Note (7/25/2026): The article has been edited with more information about the L2 behavior along with the bandwidth of the die to die interface. ## Upcoming GPU Pricing Updates DevFeed: [Upcoming GPU Pricing Updates]() Original publisher: [Read original article]() Author: Krishna Nallamothu Published: 2026-07-21T00:30:53Z Content type: release Language: en Sources: [DigitalOcean]() Topics: [Digital Ocean](), [GPU](), [model-deployment]() Tags: [2026](), [amd](), [billing](), [capacity](), [compute](), [cost](), [digitalocean](), [droplets](), [gpu](), [inference](), [infrastructure](), [nvidia](), [pricing](), [product-updates](), [training](), [updates]() ### AI overview DigitalOcean announces price increases for select NVIDIA and AMD GPU droplets effective August 1, 2026. Active workloads will be billed at the new rates, while existing reserved contracts retain their locked-in rates until renewal. ### Source excerpt Effective August 1st, 2026, we will be updating prices on select GPUs. This change reflects strong demand for advanced GPU capacity and helps us expand reliable access to high-performance compute for customers. Even with the updated rates, DigitalOcean continues to offer some of the most competitive GPU infrastructure pricing in the market. Below is a detailed breakdown of these upcoming changes and how they affect you. On-Demand GPU Price Adjustments Effective August 1, 2026, on-demand pricing for NVIDIA and AMD GPU droplets will be updated as follows: What this means for your bill: Any active workloads running on or after August 1, 2026 will be billed at the new rate. By continuing to access or use the services on or after August 1, 2026, you are agreeing to accept and pay the updated rates. These changes will be reflected in your total bill on September 1, 2026. If you do not wish to continue using the service at the updated rate, you will need to take action by August 1, 2026 to destroy your GPU Droplets. 12-Month Reserved GPU Price Adjustments For teams running predictable, continuous training or inference workloads, reserved plans remain the most cost-effective way to lock in lower rates. Effective August 1, 2026, we're also adjusting our 12-month reserved pricing: What this means for your bill: If you're currently in a contract with us that locks in your rate, there is no change to your rate. If you choose to renew after your terms expires, your rates will be adjusted to the new 12-month reserved rate outlined above. If you have questions about how these changes will impact your specific workloads, or if you want to explore reserving capacity, please contact our sales team--we're here to help you find the most cost-efficient path forward. Helpful Resources Visit DigitalOcean pricing for the most up-to-date pricing for GPU Droplets and all DigitalOcean products and services. Visit the billing dashboard for the latest on your account bill. Your use of the Digita ## Variance Reduction Below the Randomization Grain DevFeed: [Variance Reduction Below the Randomization Grain]() Original publisher: [Read original article]() Author: Tilman Drerup Published: 2026-07-01T16:28:36Z Content type: article Language: en Sources: [Instacart]() Topics: [experiments]() Tags: [capacity](), [causal-inference](), [economics](), [engineering](), [experiment](), [experimentation](), [experiments](), [machine-learning](), [marketplaces](), [metrics](), [reduce](), [science](), [techniques](), [variance]() ### AI overview This article explains how marketplace experiments can reduce metric variance below the level at which treatment is randomized. It describes cluster-level randomization for containing interference and shows how fine-grained outcome predictability can improve statistical power and reduce experimentation time. ### Source excerpt Sergio Camelo, Caitlin Kearns, Matias Cersosimo, and Tilman Drerup As artificial intelligence increases the velocity of engineering and science teams, experimental throughput is set to become a bottleneck for many product decisions. Many companies can now build faster than they can experiment, with queues of good ideas running the risk of not being tested because of lack of experimental capacity. This problem is particularly severe in marketplaces, where the presence of spillover and cannibalization effects between experimental units requires cluster-level randomization techniques. That randomization, in turn, has the unfortunate tendency to substantially reduce statistical power and slow down experimentation. In this post, we show that the predictability of outcomes at fine grains can be exploited to reduce the variance of aggregate metrics, even when experiments themselves are run at a coarse level. Since statistical power depends on metric variability, this yields considerable reductions in experimentation time. The Interference Problem In marketplace settings, behavior and outcomes for individual participants are inherently intertwined. In a delivery marketplace like Instacart, for example, the dispatch system solves a bipartite matching problem between shoppers and customer orders. Since assignments are global and interdependent, matching an order to one shopper means that the same order cannot be matched to another shopper. As a result, changing the handling for a single order creates ripples that affect the orders around it. If an experimenter were to assign a treatment intervention to one of these orders while leaving neighboring orders as controls, the latter would evidently be contaminated. A common response to this problem is to randomize treatments at the level of a cluster, chosen so that interference can stay within it. In food and grocery delivery, that cluster is typically a geographical region. Since every order within a region sees the same treatme ## RFC 10005: BGP Community for link capacity DevFeed: [RFC 10005: BGP Community for link capacity]() Original publisher: [Read original article]() Author: j2sw Published: 2026-07-01T13:31:00Z Content type: article Language: en Sources: [Justin Wilson (j2sw)]() Topics: [BGP](), [standard](), [Load Balancing](), [Network]() Tags: [bgp](), [capacity](), [internet](), [load-balancing](), [network](), [network-operations](), [rfc](), [rfc-10005](), [route](), [routers](), [routing](), [standard](), [traffic]() ### AI overview RFC 10005 defines a BGP extended community for carrying link bandwidth information in routes. When multiple BGP paths are eligible, routers can use the value to weight traffic according to link capacity, while normal best-path policy still determines route eligibility. The RFC is currently a draft. ### Source excerpt RFC 10005 defines a BGP extended community that lets a router attach link bandwidth information to a route. Another router can use that value when it spreads traffic across multiple BGP paths. RFC 10005 matters because links are not always of the same capacity. This RFC provides routers with a standard way to carry bandwidth ... Read more The post RFC 10005: BGP Community for link capacity appeared first on Justin Wilson (j2sw). ## How Rain Affects Wireless Microwave Links: 24 GHz, 60 GHz, and 80 GHz Compared DevFeed: [How Rain Affects Wireless Microwave Links: 24 GHz, 60 GHz, and 80 GHz Compared]() Original publisher: [Read original article]() Author: j2sw Published: 2026-06-15T09:47:43Z Content type: tutorial Language: en Sources: [Justin Wilson (j2sw)]() Topics: [Network](), [Network design](), [Internet](), [data centers]() Tags: [24ghz](), [60ghz](), [80ghz](), [adaptive](), [backhaul](), [capacity](), [changes](), [e-band](), [error-correction](), [fiber](), [microwave](), [network](), [network-design](), [perfomance](), [radio](), [reduction](), [rf](), [wireless](), [wireless-networking]() ### AI overview This article explains how rain fade affects wireless microwave links, with emphasis on 24 GHz, 60 GHz, and 80 GHz bands. It describes how precipitation reduces signal strength, causes adaptive modulation and throughput reductions, and can eventually cause a link to drop when fade margin is exhausted. ### Source excerpt Rain can have a major impact on wireless microwave links, especially as frequencies increase. Learn how rain fade affects 24 GHz, 60 GHz, and 80 GHz circuits, and why fade margin is critical for reliable network design. The post How Rain Affects Wireless Microwave Links: 24 GHz, 60 GHz, and 80 GHz Compared appeared first on Justin Wilson (j2sw). ## How to Choose a Database for an AI-Powered Product DevFeed: [How to Choose a Database for an AI-Powered Product]() Original publisher: [Read original article]() Author: David Weiss Published: 2026-06-01T00:00:00Z Content type: tutorial Language: en Sources: [Cockroach Labs]() Topics: [Database](), [AI Infrastructure](), [AI Platform](), [Data Infrastructure](), [autoscaling]() Tags: [ai](), [ai-infrastructure](), [capacity](), [concurrency](), [data](), [database](), [how-to](), [product](), [scale]() ### AI overview This article presents seven criteria for choosing a database for an AI-powered product in production. It emphasizes evaluating elastic scalability, concurrency, correctness, workload demands, global distribution, and access by autonomous agents rather than selecting a database solely for model support or vector search. ### Source excerpt Organizations across fintech, healthcare, retail, gaming, SaaS, and scores more verticals are embedding AI into business-critical offerings. ## Request-Based Autoscaling Is Now Generally Available on App Platform DevFeed: [Request-Based Autoscaling Is Now Generally Available on App Platform]() Original publisher: [Read original article]() Author: Greeshma Pillai Published: 2026-05-22T18:02:26Z Content type: release Language: en Sources: [DigitalOcean]() Topics: [autoscaling](), [Digital Ocean](), [App](), [Latency](), [Containers]() Tags: [autoscaling](), [capacity](), [container](), [containers](), [cpu](), [digitalocean](), [http](), [load](), [performance](), [product-launch](), [product-updates](), [production](), [real-time]() ### AI overview DigitalOcean App Platform now generally supports request-based autoscaling for shared and dedicated CPU instances. Apps can scale horizontally using live HTTP requests per second and P95 response latency, with containers scaling up when thresholds are exceeded and down when load falls. ### Source excerpt Traffic doesn't spike on a schedule. A product launch, a viral moment, or a flash sale can send request volume through the roof in seconds, long before your CPU metrics catch up. That gap is where performance suffers. Today, we're excited to announce that request-based autoscaling on DigitalOcean App Platform is now generally available. Your apps can now automatically scale based on live HTTP traffic signals (requests per second and P95 response latency) so your infrastructure reacts to what's actually happening, not what happened minutes ago. Now Available for Shared and Dedicated CPU Instances Until now, autoscaling on App Platform required a dedicated CPU plan. That meant a good portion of App Platform users (anyone running on shared CPU instances) had no path to automatic horizontal scaling at all. That changes today. Request-based autoscaling works on both shared and dedicated CPU instances. Whether you're running an early-stage project on a shared plan or a high-throughput production service on dedicated resources, you can now configure autoscaling to match your traffic--no plan upgrade required. Faster, More Responsive Scaling CPU-based autoscaling is reactive by nature. CPU is a lagging indicator: your containers have to be visibly struggling before the scaler knows there's a problem, and by then, your users are already waiting. Request-based autoscaling acts on the signals that actually reflect user experience: Requests per second per instance: how many requests each container is handling right now P95 request latency: the response time that 95% of your users are seeing When traffic rises and either threshold is exceeded, new containers spin up immediately. When load drops and all metrics fall back below their targets, the scaler brings containers back down. You get the capacity headroom you need, faster, and pay only for what you use. You can also combine request-based and CPU-based metrics on dedicated plans. The autoscaler scales up when any configured th [Next page]()