Alphabet

Waymo remains paid-scale anchor; China Apollo/Pony/WeRide grow fleets & overseas faster. Sleeve: GOOGL/BIDU/TSLA options + UBER/BYD shovels + PONY/WRD satellites. See freight.

SemiAnalysis: scaling / reasoning / agentic eras; open weights close gap in ~half the time each era (Kimi K2.6 vs Opus 4.5 ~4.8 mo), but benchmarks ≠ product. See AgentX v3, CUDA moat, NVDA.

Apache 2.0 AgentX: $3M dataset, 1M+ context, 393 Claude Code traces, 95%+ KV reuse; ~2MW/1000+ chips full-stack test. DCP/PCP composability favors NVDA; post Aug 21 B200 perf/$ passed MI355X. See TileRT, open models note, AMD.

SA sees SpaceX ~10GW cumulative by YE2027 as credible: speed-first builds + pricey short contracts. MSFT signed 10GW+ YTD with a near-term gap; ~$100B/GW/yr inference economics → likely largest offtaker. See MSFT, NeoCloud.

2026-08-05 DeepMind leadership overhaul; SA sees frontier odds collapsing as compute tilts to GCP/external TPU sales. Gemini ARR ~$12B in 2Q26 vs ~$200B YE2027 GCP external AI+TPU systems narrative. See GOOGL.

From ~2026-08-14, markets cited SemiAnalysis: possible Google–AMD on future TPU and CoWoS-driven cuts to Broadcom-related TPU/packaging (e.g. 2026 TPU v7 ~3.2M→2.7M units). **Unconfirmed by GOOGL/AMD.** See AMD, AVGO, GCP notes.

GPUs buy silicon; returns hinge on utilization. Collectives, HBM, schedulers, power/thermal, and inference batching decide effective output from hundred-billion AI CapEx. Hub NVIDIA; complements AVGO/ANET; constraints TSM/MU; physical layer VRT. See CPU industry, AgentX v3 (agentic KV/routing).

Alphabet Q2 2026 revenue $119.8B (+24%), Google Cloud $24.8B (+82%), op income $40.8B. ~$205B CapEx ceiling is the near-term focus; price ~$334. Separate GAAP EPS (SpaceX mark-to-market) from operating EPS ~$2.85.