ByteDance
ByteDance is reportedly weighing up to $70B in capital expenditure for AI infrastructure in 2026, doubling down on the compute needed to power its growing portfolio of video generation and synthetic media models.
LLM Inference
KV caching is the unsung optimization that makes modern LLMs feel real-time. Here's how it transforms transformer inference from quadratic drudgery into a fast, token-by-token stream.
AI Infrastructure
IT capital expenditure has surged to a record share of S&P 500 spending as hyperscalers pour unprecedented sums into AI infrastructure, reshaping the economics of the entire tech sector and the compute foundation behind generative AI.
Alibaba
Alibaba's new Zhenwu M890 chip roadmap is designed specifically around AI agent workloads, signaling a shift from raw training power to inference-heavy, agentic compute that could reshape the AI hardware race.
MoE
ZAYA1-8B, a Mixture-of-Experts model with just 760M active parameters trained entirely on AMD GPUs, reportedly outperforms GPT-5-High on math benchmarks—challenging NVIDIA's dominance in frontier AI training.
AI Market
JPMorgan analysts report the AI trade now accounts for more than half of the S&P 500's total weight, raising concentration risks and questions about the sustainability of the AI-driven market rally.
LLM
Speculative decoding lets large language models generate text faster by using a smaller draft model to predict tokens ahead, then verifying them in parallel. Here's how this inference optimization technique works under the hood.
AI Infrastructure
UK-based AI infrastructure firm Nscale has secured $790M in financing to expand its Norway data center operations, strengthening European GPU capacity for training and inference workloads driving the generative AI boom.
Alphabet
Alphabet is preparing a yen-denominated bond issuance to help finance its surging AI infrastructure costs, signaling how hyperscalers are diversifying debt markets to fund the global GPU and data center buildout.
ByteDance
ByteDance is boosting 2026 capital expenditure by 25% as AI infrastructure and memory chip costs surge, signaling intensified compute investment from the TikTok parent and a major force in AI video generation.
AI Infrastructure
A four-person Miami startup called SubQ claims a new attention mechanism that runs 52x faster than standard transformers at one-fifth the cost of Claude Opus, hinting at cheaper long-context inference for video and multimodal AI.
Anthropic
Anthropic is raising Claude Code usage caps after striking a new compute deal with SpaceX, easing throttling complaints from developers and signaling an unusual entry by Elon Musk's space company into the AI infrastructure market.