The Daily Chart

Why it matters: Artificial Analysis scored NVIDIA's new Nemotron 3.5 Lightning at 24 on the Intelligence Index—9 points above Nemotron 3 Nano (15) and level with OpenAI's gpt-oss-120b (24)—despite using roughly a quarter of the total parameters (31.6B vs ~120B).

Why it matters: Artificial Analysis scored Meta’s Muse Glimmer (high) at 35 on the Intelligence Index—21 points above Llama 4 Maverick (14), Meta’s last open-weights release—placing the 30B model alongside Kimi K2.5 (36) and just behind Qwen3.6 27B (38).

Why it matters: SK hynix’s board approved about ₩54.3 trillion ($38.3B) for two new fabs—₩35.2T for Yongin Y2 to make HBM and next-gen DRAM, and ₩19.1T for Cheongju M17 NAND—locking in mid-to-long-term AI memory capacity through 2031.

Why it matters: Nikkei Asia reports the four U.S. tech giants now hold $1.46 trillion in property, plant and equipment—up 140% in three years—as AI data-center buildouts turn former asset-light firms into infrastructure giants. Amazon alone accounts for $538.7 billion, nearly matching the group’s entire PP&E from three years ago.

Why it matters: Artificial Analysis’s Aug 6 Intelligence Index v4.1.1 grader upgrade lifted Muse Spark 1.2 (xhigh) by 2.7 points—the largest move of any model—from 54 to 57, while Claude Opus 5 still leads at 63.

Why it matters: Artificial Analysis scored Meta’s Muse Spark 1.2 at 1631 Elo on GDPval-AA v2—260 points above Muse Spark 1.1 (1371) and ahead of Claude Opus 4.8 (1588)—while Claude Opus 5 still leads at 1852.

Why it matters: Artificial Analysis found a 31-point spread serving the same gpt-oss-120b model: Amazon Bedrock matched 101% of the reference endpoint while Cloudflare preserved 70%, showing that provider configuration can materially change model quality.

Why it matters: TrendForce forecasts top-nine cloud-provider CapEx to grow 90% in 2026, versus 31% growth in AI server shipments—a sign that the buildout extends beyond server counts into chips, data centers, networking, power, and cooling.

Why it matters: Claude Fable 5 solved 64% of Epoch AI’s MirrorCode tasks, more than triple GPT-5.6 Sol’s 20% solve rate on a benchmark that requires agents to rebuild complete software projects.

Why it matters: TrendForce expects spending to rise by nearly 50% in 2027 as cloud providers keep building AI servers, data centers, networking, cooling, and power infrastructure.