AI Infrastructure
GPUs, silicon, data centres and the data platforms underneath.
Anthropic’s Reported Volta Deal and AI Cloud Infrastructure
TechCrunch reports a $10 billion compute agreement between Anthropic and Volta, while leaving key details about delivery and terms unclear.
Read Article
Evolution Strategies for Long-Horizon LLM Agents: A Technical Review of Agentic ESOpt
A technical review of Agentic ESOpt, examining evolution strategies for full-parameter optimization of long-horizon LLM agents, memory reduction, and WebArena-Lite benchmark results.
QVAC Fabric and AI Edge Infrastructure: BitNet Fine-Tuning
QVAC Fabric brings BitNet LoRA fine-tuning to consumer GPUs and phones, showing how compact models can support local AI workloads.
AI Hardware Architecture, Huawei-Style: Why China's Champion Thinks Scale Trumps Efficiency
Huawei's HUAWEI CONNECT 2025 keynote laid out a billion-dollar bet that million-processor clusters can outflank Nvidia's roadmap — if the ecosystem holds together.
Quantize Where It Counts: Custom Tensor Formats and Mixed-Precision Layers to Optimize Inference Costs on the GPU Cloud
How packed binary formats like GGUF and per-layer mixed-precision quantization cut memory and bandwidth, the real levers behind how much an LLM costs to serve on shared GPU cloud clusters.
AI Model Local Deployment: Benchmarking MiniCPM-V 4.7 35B 131K Context on Apple Silicon M4 Max
Performance profiling of MiniCPM-V-4.7-35B-A3B 131K context on Apple Silicon M4 Max with measured TTFT, memory footprint, and prefix caching metrics.
Evaluating Nemotron-3.5 Bolt 74B-A9B 131K Context Performance for AI Model Local Deployment
Measured performance of NVIDIA Nemotron-3.5 Bolt 74B-A9B 4-bit experimental conversion on Apple Silicon M4 Max across a four-turn growing-prefix probe up to 131K context capacity.
Choosing MLX weight precision by tensor role and storage budget for ai model local deployment
An in-depth explainer on MLX affine group quantization, 4-bit, 8-bit, and mixed 4/8-bit precision choices for local model deployment on Apple Silicon. Covers tensor-role preservation for norms, scales, biases, and router gates using an audited LFM mixed pack example.
AI Model Local Deployment: Qwen3.8-Flash Throughput Gains Between mlx-serve 26.8.11 and 26.10.2 — What Is Real?
Comparing two benchmark runs of Qwen3.8-Flash-Next mixed-quantization on Apple Silicon M4 Max across mlx-serve releases, separating genuine build improvements from environmental contention artifacts, with quality scores and deployment guidance.
Where Local LLM Serving Hits Its Concurrency Ceiling
Comprehensive October 2026 benchmark analysis of cross-model concurrency scaling, batching efficiency, early saturation, and local deployment strategies on Apple Silicon.
AI Model Local Deployment: Analyzing the MiniCPM-V Text-Only Conversion Scorecard
A detailed technical analysis of the MiniCPM-V 35B 4-bit text-only benchmark scorecard running on Apple Silicon, exploring local deployment practices, execution throughput, and agentic AI capabilities.
AI Model Local Deployment: Deep Dive into Qwen3.8 Flash Next Scorecard
Comprehensive single-model deep dive into Qwen3.8 Flash Next local deployment on Apple Silicon, covering plain vs MTP decode, prefill metrics, context ladder, latency, and capability scorecard.