AMD Helios Goes Into Production, and AMD Says the CUDA Moat Already Drained
AMD Helios, unveiled at the Advancing AI 2026 event, is entering production, shipping with major clients like Microsoft. Corporate VP Andrew Dieckman downplayed CUDA's…
AMD Helios, unveiled at the Advancing AI 2026 event, is entering production, shipping with major clients like Microsoft. Corporate VP Andrew Dieckman downplayed CUDA's…
As of July 2026, Samsung and SK hynix reconsider hybrid bonding for HBM4 due to JEDEC's potential increase in stack height limits. This change…
The real issue in AI isn't benchmarks, but sovereignty, focusing on control over compute, models, and access. Hardware access determines power, as seen in…
Keller welcomes the Cerebras IPO, then bets cheap DRAM and open RISC-V beat it on cost. The real prize is the CPU Qualcomm is…
As of June 2026, the RTX 5090's price has surged to $4,329, driven by high demand for its GDDR7 memory. While it boasts impressive…
OpenAI and Broadcom announce Jalapeño, a custom chip designed to run OpenAI's models. With a nine-month development timeline, the ASIC aims for efficiency and…
Qualcomm is acquiring Modular, the vendor-neutral stack that runs AI models on any chip. The phone NPU benefit is a side effect. The real…
Chiplets and UCIe are unbundling AI hardware the same way Modular unbundles CUDA. Why the moat doesn't vanish, it just relocates to TSMC's packaging…
Over 75 data center projects worth $130 billion were blocked in Q1 2026 alone, matching all of 2025 in three months. The bottleneck is…
Snapdragon Insider bias declared. Qualcomm's AI200 data center card plus the five-form-factor Hexagon NPU strategy is a bet that inference is where the real…
TSMC's A16 node brings backside power delivery (Super Power Rail) in 2026, answering Intel's 18A and opening the angstrom era for AI and HPC…
The Fable 5 shutdown highlighted the importance of owning AI models on personal hardware to avoid government restrictions. Although open-source AI is gaining traction,…
Reasoning models are measurably smarter. The real question is who controls the infrastructure they run on, and whether better reasoning creates value or creates…
128 GB of unified memory, a 40-core RDNA GPU, and local LLM inference nobody can revoke. The GMKtec EVO-X2 is the hardware argument for…
Jetson Thor owns the humanoid robot; Qualcomm Dragonwing is fighting for the high-volume middle. The 2026 edge robotics compute split, broken down.
One maker's tech digest, weekly. Chips, AI, gaming, EVs — no spam.