Category / T-CAT / HARDWARE
Hardware
Hardware / T-2026-9272
The $3 Trillion Bet: Is the AI Data Center Boom a Bubble or a Supercycle?
AI data center investment could reach $3 trillion by 2030. Hyperscalers report capacity shortages, but debt risks and tech obsolescence loom. A structural shift or a bubble?
SourceWill There Be an AI Data Center Bubble in 2026? A Deep Dive into ...
Hardware / T-2026-5608
190 GW of announced data center capacity is reshaping the hardware investment stack
Bessemer Venture Partners maps six layers of AI data center infrastructure opportunity, from permitting software to fusion power.
SourceRoadmap: The AI data center stack - Bessemer Venture Partners
Hardware / T-2026-7624
Megawatt Racks and Campus-Scale Systems: Data Center World 2026 Shows AI Is Rewriting the Rules
At Data Center World 2026, Oracle, Nvidia, and Google engineers described a systemic redesign as facilities shift from general-purpose IT to integrated AI compute systems.
SourceData Center World 2026: AI Pushes Infrastructure to New Limits
Hardware / T-2026-0875
Qualcomm enters the AI data center with inference chips, stock jumps 11%
Qualcomm announced the AI200 and AI250 AI chips for data centers, focusing on inference. The stock rose 11%. The move challenges Nvidia's dominance with a different architectural…
SourceQualcomm announces AI chips to compete with AMD and Nvidia
Hardware / T-2026-8658
AMD's 'No Asterisk' Bet: MI450 vs. Nvidia Rubin
AMD promises its MI450 GPU will outperform any Nvidia product, including Rubin Ultra. The claim is a bet on silicon, software, and timing.
SourceAMD says MI450 GPUs will outperform every Nvidia product including ...
Hardware / T-2026-8037
AMD's MI350 to MI500 Roadmap Puts a Clock on Nvidia's AI Hardware Lead
AMD's Instinct MI350 series, built on CDNA4 and 3nm chiplets, ships this year. The roadmap through 2027 signals a direct challenge to Nvidia's dominance in AI training and…
SourceFrom MI350 to MI500: AMD's Bold AI Accelerator Roadmap Through 2027
Hardware / T-2026-7646
With Maia 200, Microsoft builds an inference-first accelerator for the GPT era
Microsoft's Maia 200 inference accelerator, built on TSMC 3nm, delivers 10 petaFLOPS at FP4 with 30% better performance per dollar. The chip is already serving GPT-5.2 and…
SourceMaia 200: The AI accelerator built for inference - The Official Microsoft Blog
Hardware / T-2026-5073
VAST Data's 2026 Inference Predictions: Agents, Cost Fluidity, and the End of GPU-Only Thinking
VAST Data argues 2026 is the year AI inference goes mainstream. Its five trends reveal a maturing industry wrestling with scale, cost, and architectural diversity.
Source2026: The Year of AI Inference - VAST Data
Hardware / T-2026-8346
Google splits its next TPU into two chips, one for training and one for reasoning
Google splits its next TPU into two chips — 8t for training, 8i for reasoning — and unveils a data center fabric that stitches a million accelerators together.
SourceAI infrastructure at Next '26 | Google Cloud Blog
Hardware / T-2026-2588
Custom AI ASICs hit an inflection point: 27.8% of server shipments in 2026
Custom AI ASIC shipments will hit 27.8% of the market in 2026, with Broadcom alone carrying a $73 billion backlog. An inflection point for inference compute.
SourceThe custom AI ASIC state of play (May 2026) - Tom's Hardware
Hardware / T-2026-7107
AMD's Vivado bait-and-switch betrays Linux users
AMD changed Vivado licensing to require a Windows tool, breaking Linux workflows and eroding trust in its open-source commitments.
SourceAMD pulls a bait-and-switch on Linux users with Vivado licensing changes
Hardware / T-2026-2473
Rust on a Jailbroken Kindle: The Hardware Tinkerer's Playground
A developer ports Rust and Slint to a jailbroken Kindle, revealing the state of embedded software tooling and the appeal of repurposing old hardware.
SourceRust (and Slint) on a Jailbroken Kindle