{"@context":"https://schema.org","@type":"WebPage","name":"API Versioning & Deprecation Policy","description":"Aguasto API compatibility, major version changes, migration notices and at least 90 days notice before retiring an operation. No current deprecations.","url":"https://www.aguasto.com/developers/versioning","dateModified":"2026-10-27","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/developers/versioning.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Developer Sandbox & Synthetic API Tests","description":"Test Aguasto search and calculator integrations against fixed synthetic fixtures with no account, production-data access or stored requests.","url":"https://www.aguasto.com/sandbox","dateModified":"2026-10-27","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/sandbox.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Aguasto | Modular AI Infrastructure, GPU Compute & Inference","description":"Aguasto is developing a distributed network of modular AI data centers designed for high-density GPU computing, dedicated enterprise infrastructure and high-performance AI inference.","url":"https://www.aguasto.com/","dateModified":"2026-10-27","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/index.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Infrastructure","description":"Aguasto's development approach to modular, liquid-cooled, high-density AI compute facilities with independent local infrastructure.","url":"https://www.aguasto.com/infrastructure","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/infrastructure.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"GPU Cloud","description":"Aguasto's planned dedicated GPU capacity, reserved enterprise compute, and workload-specific configurations. Capacity is not currently bookable.","url":"https://www.aguasto.com/gpu-cloud","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpu-cloud.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Inference","description":"Aguasto's planned enterprise inference layer for open-weight models, private endpoints, and API access. The service is in development.","url":"https://www.aguasto.com/ai-inference","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/ai-inference.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Development & Investment","description":"Aguasto's long-term strategy for infrastructure development, asset ownership, and project-specific financing with an independent parent platform.","url":"https://www.aguasto.com/development","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/development.md"}}
{"@context":"https://schema.org","@type":"AboutPage","name":"Company","description":"Aguasto is an emerging, U.S.-focused AI compute company developing modular facilities, dedicated GPU infrastructure and a planned model-serving business.","url":"https://www.aguasto.com/company","dateModified":"2026-10-17","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/company.md"}}
{"@context":"https://schema.org","@type":"ContactPage","name":"Contact","description":"Connect with Aguasto about future enterprise compute, model serving, infrastructure supply, and strategic relationships.","url":"https://www.aguasto.com/contact","dateModified":"2026-10-27"}
{"@context":"https://schema.org","@type":"WebPage","name":"Privacy","description":"Privacy information for the Aguasto corporate website and inquiry form.","url":"https://www.aguasto.com/privacy","dateModified":"2026-10-12"}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Compute Research Topics","description":"Four connected research paths through GPU systems, model serving, modular infrastructure and compute economics.","url":"https://www.aguasto.com/topics","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/topics.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"GPU systems & local clusters","description":"Build a configuration-specific shortlist without confusing an individual accelerator, a rack platform and a tested computing service.","url":"https://www.aguasto.com/topics/gpu-systems","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/topics/gpu-systems.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Model serving & inference","description":"Connect checkpoint identity, memory policy, scheduling and measured request behavior to a defined application boundary.","url":"https://www.aguasto.com/topics/model-serving","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/topics/model-serving.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Modular infrastructure & resources","description":"Follow the physical dependencies from authorized electricity through facility integration, cooling and accepted computing capacity.","url":"https://www.aguasto.com/topics/modular-infrastructure","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/topics/modular-infrastructure.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Compute economics & evidence","description":"Connect a declared cost and operating boundary to successful, quality-qualified output instead of an unsupported efficiency headline.","url":"https://www.aguasto.com/topics/compute-economics","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/topics/compute-economics.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Compute & Infrastructure Glossary","description":"25 source-linked definitions with practical measurement, licensing and operating boundaries.","url":"https://www.aguasto.com/glossary","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/glossary.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Research Sources & Methodologies","description":"How Aguasto classifies evidence, compares references, exposes calculations, reviews assertions and preserves publication records.","url":"https://www.aguasto.com/methodologies","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/methodologies.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Developer API, MCP & SDK Documentation","description":"Integrate Aguasto's public research search and educational calculators through HTTP, MCP, browser tools and downloadable agent skills. No GPU service credentials are issued.","url":"https://www.aguasto.com/developers","dateModified":"2026-10-27","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/developers.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Aguasto Research & Resources","description":"Explore Aguasto's sourced GPU and model references, inference software, engineering guides, original research and explainable tools.","url":"https://www.aguasto.com/resources","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/resources.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"GPU Library","description":"A curated manufacturer specification library for AI accelerators and rack systems. Compare like categories with source, unit and configuration boundaries.","url":"https://www.aguasto.com/gpus","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Model Index","description":"Curated open-weight checkpoint references, license distinctions, model memory considerations and serving documentation. Research references, not available Aguasto endpoints.","url":"https://www.aguasto.com/models","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Inference Stack Library","description":"Understand the software between GPUs and model APIs: serving engines, distributed inference, runtime ecosystems and GPU orchestration.","url":"https://www.aguasto.com/inference-stack","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Infrastructure Engineering Guides","description":"Practical explanations of model memory, serving architecture, inference reliability and the physical infrastructure behind AI compute.","url":"https://www.aguasto.com/learn","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Infrastructure Insights","description":"Aguasto's analysis of dedicated compute, managed inference and the operating economics of future AI infrastructure.","url":"https://www.aguasto.com/insights","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/insights.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Infrastructure Research","description":"Original sourced analysis, benchmark methodologies and dated industry evidence supporting Aguasto's modular and distributed infrastructure thesis.","url":"https://www.aguasto.com/research","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Public Infrastructure Case Analyses","description":"Evidence-based analysis of third-party infrastructure operators. Announced, contracted, energized and billed capacity remain distinct; these are not Aguasto customer cases.","url":"https://www.aguasto.com/research/cases","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/cases.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Why Modular AI Data Centers?","description":"Aguasto's approach to standardized engineering, phased development and independently operated AI facilities, with explicit power, permitting and capital trade-offs.","url":"https://www.aguasto.com/infrastructure/why-modular","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/infrastructure/why-modular.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Cooling & Resource Efficiency","description":"Aguasto’s evaluation framework for rack cooling, heat rejection, local and indirect water, electricity and quality-qualified compute. Design priorities, without operating environmental claims.","url":"https://www.aguasto.com/infrastructure/cooling-resource-efficiency","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/infrastructure/cooling-resource-efficiency.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"OEM & Strategic Collaboration","description":"Potential collaboration with GPU manufacturers, rack-system OEMs, infrastructure suppliers and enterprise AI organizations. No existing partnership is implied.","url":"https://www.aguasto.com/partners","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/partners.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Compare GPU & Rack References","description":"Compare compatible accelerator or rack-system references using manufacturer specifications. No fabricated token benchmark, available inventory or cross-unit performance ranking.","url":"https://www.aguasto.com/gpus/compare","dateModified":"2026-10-12"}
{"@context":"https://schema.org","@type":"WebPage","name":"Compare Open-Weight Model References","description":"Compare curated checkpoints by documented parameters, modality, license, context and theoretical weight memory. No Aguasto evaluated deployment is implied.","url":"https://www.aguasto.com/models/compare","dateModified":"2026-10-12"}
{"@context":"https://schema.org","@type":"WebPage","name":"AI Infrastructure Engineering Calculators","description":"Estimate weight memory, PUE-adjusted electricity and cost per million completed tokens with exposed formulas and explicit illustrative assumptions.","url":"https://www.aguasto.com/tools","dateModified":"2026-10-12"}
{"@context":"https://schema.org","@type":"WebPage","name":"The Growing Friction in Hyperscale Data Center Development","description":"A dated, jurisdiction-specific examination of data-center moratoria, a federal court proceeding and power constraints, balanced against continued construction.","url":"https://www.aguasto.com/research/hyperscale-development-friction","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/hyperscale-development-friction.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA H100 SXM","description":"Manufacturer reference for NVIDIA H100 SXM: Hopper, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/h100","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/h100.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA H200 SXM","description":"Manufacturer reference for NVIDIA H200 SXM: Hopper, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/h200","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/h200.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA B200 SXM","description":"Manufacturer reference for NVIDIA B200 SXM: Blackwell, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/b200","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/b200.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA B300","description":"Manufacturer reference for NVIDIA B300: Blackwell Ultra, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/b300","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/b300.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA GB200 NVL72","description":"Manufacturer reference for NVIDIA GB200 NVL72: Grace Blackwell, rack system, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/gb200-nvl72","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/gb200-nvl72.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA GB300 NVL72","description":"Manufacturer reference for NVIDIA GB300 NVL72: Grace Blackwell Ultra, rack system, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/gb300-nvl72","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/gb300-nvl72.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA Vera Rubin NVL72","description":"Manufacturer reference for NVIDIA Vera Rubin NVL72: Vera CPU / Rubin GPU, rack system, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/vera-rubin-nvl72","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/vera-rubin-nvl72.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD Instinct MI300X","description":"Manufacturer reference for AMD Instinct MI300X: CDNA 3, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/mi300x","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/mi300x.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD Instinct MI325X","description":"Manufacturer reference for AMD Instinct MI325X: CDNA 3, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/mi325x","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/mi325x.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD Instinct MI355X","description":"Manufacturer reference for AMD Instinct MI355X: CDNA 4, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/mi355x","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/mi355x.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD Instinct MI455X","description":"Manufacturer reference for AMD Instinct MI455X: CDNA 5, individual accelerator, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/mi455x","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/mi455x.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD Helios","description":"Manufacturer reference for AMD Helios: MI455X / EPYC Venice / Pensando, rack system, with sourced specifications, variant boundaries and deployment considerations.","url":"https://www.aguasto.com/gpus/helios","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/gpus/helios.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"GLM-4.5","description":"Research reference for GLM-4.5: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/glm-4-5","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/glm-4-5.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Qwen3-235B-A22B","description":"Research reference for Qwen3-235B-A22B: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/qwen3-235b-a22b","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/qwen3-235b-a22b.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"DeepSeek-R1","description":"Research reference for DeepSeek-R1: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/deepseek-r1","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/deepseek-r1.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Llama 3.3 70B Instruct","description":"Research reference for Llama 3.3 70B Instruct: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/llama-3-3-70b-instruct","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/llama-3-3-70b-instruct.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Gemma 3 27B IT","description":"Research reference for Gemma 3 27B IT: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/gemma-3-27b-it","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/gemma-3-27b-it.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"gpt-oss-120b","description":"Research reference for gpt-oss-120b: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/gpt-oss-120b","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/gpt-oss-120b.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Kimi-K2-Instruct","description":"Research reference for Kimi-K2-Instruct: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/kimi-k2-instruct","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/kimi-k2-instruct.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"GLM-5.3","description":"Research reference for GLM-5.3: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/glm-5-3","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/glm-5-3.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Qwen3.8-27B","description":"Research reference for Qwen3.8-27B: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/qwen3-8-27b","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/qwen3-8-27b.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"DeepSeek-V4.1-Flash","description":"Research reference for DeepSeek-V4.1-Flash: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/deepseek-v4-1-flash","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/deepseek-v4-1-flash.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Llama 4 Scout 17B-16E Instruct","description":"Research reference for Llama 4 Scout 17B-16E Instruct: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/llama-4-scout-17b-16e-instruct","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/llama-4-scout-17b-16e-instruct.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Gemma 4 31B IT","description":"Research reference for Gemma 4 31B IT: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/gemma-4-31b-it","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/gemma-4-31b-it.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Kimi K3","description":"Research reference for Kimi K3: official checkpoint, license and memory considerations. Documented in Aguasto Research; no Aguasto deployment or API availability is represented.","url":"https://www.aguasto.com/models/kimi-k3","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/models/kimi-k3.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"vLLM","description":"Technical reference for vLLM: Serve supported language and multimodal models using continuous batching, attention-cache management and distributed inference. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/vllm","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/vllm.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"SGLang","description":"Technical reference for SGLang: Serve supported language and multimodal models with prefix caching and multi-GPU inference. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/sglang","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/sglang.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"TensorRT-LLM","description":"Technical reference for TensorRT-LLM: Optimize and serve supported generative models using NVIDIA GPU kernels, cache management and parallel inference. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/tensorrt-llm","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/tensorrt-llm.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA Dynamo","description":"Technical reference for NVIDIA Dynamo: Coordinate inference engines through routing, separate prefill and decode workers, cache management and scaling. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/dynamo","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/dynamo.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"NVIDIA Triton Inference Server","description":"Technical reference for NVIDIA Triton Inference Server: Expose model inference through supported backends, scheduling, batching and model pipelines. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/triton","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/triton.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"AMD ROCm","description":"Technical reference for AMD ROCm: Provide GPU runtimes, compilers, math and communication libraries for supported AMD workloads. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/rocm","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/rocm.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Ray Serve","description":"Technical reference for Ray Serve: Build Python inference applications that compose models and business logic and scale replicas across Ray resources. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/ray-serve","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/ray-serve.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Kubernetes GPU orchestration — NVIDIA device plugin","description":"Technical reference for Kubernetes GPU orchestration — NVIDIA device plugin: Expose NVIDIA GPU devices as Kubernetes resources and allocate them to container workloads. Hardware assumptions and operating trade-offs are documented.","url":"https://www.aguasto.com/inference-stack/kubernetes","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/inference-stack/kubernetes.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Model weights are only the beginning of the memory budget","description":"Help engineers separate a transparent weight-storage estimate from the workload-dependent memory required to serve a model.","url":"https://www.aguasto.com/learn/model-weights-and-runtime-memory","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/model-weights-and-runtime-memory.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Why tokens per second is not enough to benchmark inference","description":"Define a reproducible future Aguasto test protocol that evaluates completed useful work, latency, errors, workload and measurement boundaries.","url":"https://www.aguasto.com/research/inference-benchmark-methodology","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/inference-benchmark-methodology.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Prefill, decode and the cost of separating them","description":"Explain the two serving phases and when to evaluate disaggregation without claiming automatic throughput gains.","url":"https://www.aguasto.com/learn/prefill-decode-and-disaggregated-serving","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/prefill-decode-and-disaggregated-serving.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Choosing between vLLM, SGLang and TensorRT-LLM","description":"Compare documented serving-engine roles and build a selection procedure rather than an unsupported performance ranking.","url":"https://www.aguasto.com/learn/choosing-an-inference-serving-framework","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/choosing-an-inference-serving-framework.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"From Power to Inference—Building a Distributed AI Compute Network","description":"Connect physical capacity, server/runtime assumptions and completed inference work through a transparent conceptual measurement methodology.","url":"https://www.aguasto.com/learn/from-power-to-inference","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/from-power-to-inference.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Dedicated compute and shared inference: compare the whole workload","description":"Provide a defensible cost-comparison method that includes idle reservation, completed tokens, operating scope and latency rather than a universal break-even claim.","url":"https://www.aguasto.com/insights/dedicated-and-shared-inference-economics","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/insights/dedicated-and-shared-inference-economics.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Distributed inference reliability begins with the failure boundary","description":"Distinguish replica recovery, control-plane recovery and regional failover, and specify the evidence needed before publishing availability claims.","url":"https://www.aguasto.com/learn/distributed-inference-reliability","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/distributed-inference-reliability.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Core Scientific: reading contracted, energized and billed capacity","description":"Analyze dated public infrastructure disclosures with strict measurement/status distinctions; this is an independent third-party case, not an Aguasto customer story.","url":"https://www.aguasto.com/research/cases/core-scientific-capacity-disclosures","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/cases/core-scientific-capacity-disclosures.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Why Modular AI Infrastructure Matters Now","description":"Explain modular deployment as an engineering and delivery strategy, with six distinct benefits/trade-offs and explicit limits around electrical service, permitting and tightly coupled compute.","url":"https://www.aguasto.com/research/why-modular-ai-infrastructure-matters-now","dateModified":"2026-10-12","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/why-modular-ai-infrastructure-matters-now.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"What Is a Modular AI Data Center?","description":"Distinguish a facility module, equipment assembly and GPU rack, then define the interfaces that connect an enclosure to usable AI compute.","url":"https://www.aguasto.com/learn/what-is-a-modular-ai-data-center","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/what-is-a-modular-ai-data-center.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"How Modular AI Data Centers Are Built and Commissioned","description":"Follow the handoffs between requirements, factory integration, transport, site acceptance and compute-service evaluation without treating any single milestone as completion.","url":"https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Liquid Cooling for High-Density GPU Systems: Follow the Interfaces","description":"Understand facility and technology cooling circuits, residual air paths and the evidence needed to evaluate a proposed GPU rack integration.","url":"https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Why Electricity Availability Shapes AI Infrastructure Deployment","description":"Separate electrical equipment, connection readiness and usable service, using government-laboratory research and dated operator delivery evidence.","url":"https://www.aguasto.com/research/electricity-availability-and-time-to-power","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/electricity-availability-and-time-to-power.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"What Is KV Cache? Capacity, Reuse and Isolation","description":"Connect attention-state shape, request length, allocation policy and exact-prefix reuse to a transparent inference memory and admission plan.","url":"https://www.aguasto.com/learn/kv-cache-capacity-and-policy","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/kv-cache-capacity-and-policy.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Continuous Batching: Throughput Without Hiding Latency","description":"Explain iteration scheduling, token budgets, cache admission and workload measurement without treating larger batches or utilization as proof of a better inference service.","url":"https://www.aguasto.com/learn/continuous-batching-and-latency","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/continuous-batching-and-latency.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Applied Digital: Reading the Polaris Forge 1 Delivery Milestones","description":"Analyze dated operator disclosures about phased critical IT load, contracted buildout and financing without turning facility capacity into AI throughput or an Aguasto operating claim.","url":"https://www.aguasto.com/research/cases/applied-digital-polaris-forge-capacity-milestones","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/cases/applied-digital-polaris-forge-capacity-milestones.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Crusoe Abilene: Separate Delivered Infrastructure from Expansion Plans","description":"Examine dated public evidence for an operating first phase, original-campus financing and a separately announced adjacent campus while preserving capacity and forecast boundaries.","url":"https://www.aguasto.com/research/cases/crusoe-abilene-delivery-and-expansion","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/cases/crusoe-abilene-delivery-and-expansion.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Data Center Cooling: Choosing Air, Liquid, Dry and Hybrid Architectures","description":"Separate equipment heat capture from final heat rejection, then evaluate an architecture against climate, hardware, water and operating constraints.","url":"https://www.aguasto.com/learn/data-center-cooling-air-liquid-dry-hybrid","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/data-center-cooling-air-liquid-dry-hybrid.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Can Data Centers Reduce Water Consumption? Define the Water Boundary First","description":"Explain freshwater substitution, peak withdrawal, direct consumption and electricity-related water use without turning a closed loop into a zero-water claim.","url":"https://www.aguasto.com/research/data-center-water-consumption-and-withdrawal","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/data-center-water-consumption-and-withdrawal.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Measuring PUE, WUE and AI Compute Efficiency at the Right Boundary","description":"Build a measurement framework that connects facility overhead and water intensity to useful compute, without confusing device telemetry, ratios or hypothetical arithmetic with operating results.","url":"https://www.aguasto.com/learn/measuring-pue-wue-and-ai-compute-efficiency","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/measuring-pue-wue-and-ai-compute-efficiency.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"The Environmental Trade-offs of High-Density AI Infrastructure","description":"Evaluate construction, operational resources, equipment refresh and reuse over a declared life cycle rather than assigning a green result to density or modular form.","url":"https://www.aguasto.com/research/environmental-tradeoffs-and-infrastructure-reuse","dateModified":"2026-10-13","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/research/environmental-tradeoffs-and-infrastructure-reuse.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Selecting GPU Systems for Enterprise Inference","description":"Turn a workload brief into a qualified system and acceptance record, separating checkpoint support, memory, topology, facility interfaces and operating evidence from accelerator headline specifications.","url":"https://www.aguasto.com/learn/selecting-enterprise-inference-gpu-systems","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/selecting-enterprise-inference-gpu-systems.md"}}
{"@context":"https://schema.org","@type":"WebPage","name":"Networking Rack-Scale AI Systems: Define the Communication Boundary","description":"Explain GPU scale-up, inter-node collectives, KV transfer, host/storage traffic and independent-replica routing as distinct qualification paths rather than one aggregate bandwidth number.","url":"https://www.aguasto.com/learn/rack-scale-ai-networking-boundaries","dateModified":"2026-10-14","encoding":{"@type":"MediaObject","encodingFormat":"text/markdown","contentUrl":"https://www.aguasto.com/learn/rack-scale-ai-networking-boundaries.md"}}