AMD · Advanced Micro Devices, Inc.

Post argues Qualcomm ($QCOM) has a newly confirmed hyperscaler custom-silicon engagement for data-center CPU with initial shipments later this calendar year, potentially driving an AI/data-center re-rating. It frames $QCOM as a “cheap legacy smartphone chipmaker” (low forward P/E cited) with hidden AI upside, while acknowledging handset demand/memory-shortage risks and secular mobile concerns. Mentions valuation comps ($ARM, $INTC, $AMD) and an analogy to Soitec (Soitec) as prior “hidden AI upsi

arXiv paper proposes UniMVU, an instruction-aware dynamic gating architecture for multimodal video understanding (video+audio+depth/temporal streams). It reduces “modality interference” from uniform fusion by reweighting salient regions within modalities and entire modality streams conditioned on the text instruction, showing sizable benchmark gains. Investable angle: improves accuracy/efficiency of multimodal video agents and sensor/stream fusion, reinforcing demand for GPU/cloud inference and

arXiv paper proposes GARD: diffusion-based denoising/restoration performed in the *feature space* of a feed-forward multi-view 3D reconstruction model, aiming to make 3D reconstruction robust to real-world image degradations; also adds an RGB decoder to recover improved imagery alongside geometry. This is early-stage research (no product/partner), but it reinforces a broader trend: more compute-heavy, diffusion-style enhancement pipelines migrating from pixels to learned representations, which c

Scientific paper proposes an exact decomposition explaining why neural-network curvature scaling differs by layer type, and derives an architecture-adaptive preconditioner (“Spectral Newton”) that reportedly beats AdamW on vision benchmarks where conv layers show curvature exponent ~2. If validated and productized, it is an optimizer/second-order training efficiency story (time-to-train, stability, fewer steps) that could modestly shift AI training cost curves—most plausibly affecting hyperscale

Paper claims visual graph-structured “mind map” scaffolds materially improve LLM multi-hop reasoning under “abstract guidance” (no direct answer hints), outperforming flattened text graph representations; benefits persist post SFT and KL distillation. Investable implication is incremental tailwind for multimodal/vision-language model stacks and tooling that enable structured visual reasoning and UI-level reasoning scaffolds, but it is early-stage and not yet a clear product catalyst on its own.

Scientific paper proposes fine-tuning an open VLM (LLaVA-1.5-7B via QLoRA) on a few thousand curated bridge-inspection image+text pairs to reduce inter-rater variability and automate damage description + rule-based repair priority scoring. Key investable implication: bridge/infrastructure owners can adopt AI triage workflows with modest data scale (2k–3k high-quality samples) and practical inference optimizations—supporting demand for (1) AEC/asset-management software that can embed vision AI, (

ABAW@CVPR 2026 highlights continued progress and benchmarking in multimodal affect/behavior understanding (emotion, action units, pose/motion, violence detection, fairness/robustness). While not directly commercial, it reinforces an investable theme: broader deployment of multimodal video+audio analytics in consumer devices, enterprise safety/security, and content moderation—driving incremental demand for AI compute (training + inference), edge AI SoCs, and select video-analytics platforms. Key

Post argues early-July selloff broadly marked down the AI buildout supply chain despite Morgan Stanley raising hyperscaler capex forecasts (2027/2028). The actionable catalyst window is Q2 earnings/capex commentary (roughly Jul 16–Aug 5; especially Jul 22–Jul 30), which could validate or refute elevated capex expectations and re-rate downstream AI buildout names (memory, foundry, semi equipment, photonics, power).

Post argues July 16–Aug 5 earnings/capex commentary will determine whether the AI buildout selloff was overdone. Notes sharp early-July drawdowns across semi equipment/test/implant and memory-related names, while Morgan Stanley raised 2027–2028 hyperscaler capex forecasts (and is “more bullish on Amazon capex than Amazon is”). Core implication: hyperscaler capex confirmation vs contradiction will flow through the entire AI supply chain (HBM/memory, foundry, photonics, power, semi equipment).

Post argues AI infrastructure bottleneck is shifting from GPUs toward CPUs as agentic/workflow-based AI increases branching, I/O, and decision-heavy tasks. Implies rising CPU demand intensity (CPU:GPU ratio moving toward 1:1) and underappreciated CPU supply/throughput constraints.

Paper claims a co-designed diffusion-transformer + kernel/quantization stack enabling real-time (24 FPS end-to-end) streaming video-to-video editing at ~720p on a single NVIDIA RTX 5090 (Blackwell), with DiT core at 58 FPS. The actionable market mechanism is: real-time generative video editing becomes feasible on consumer GPUs, pulling demand toward high-end NVIDIA GPUs and CUDA-optimized inference stacks; downstream, creator/live-streaming and game/UGC platforms could add real-time AI effects i

Paper proposes SURGE, a contrastive (InfoNCE) relational-geometry knowledge distillation method to make SAR ship-detection models much lighter while retaining/improving accuracy. If reproducible and productized, it is a practical catalyst for real-time/onboard SAR analytics (satellites, UAVs, maritime ISR), shifting value toward edge-deployable inference stacks and SAR data/analytics vendors. The investable mechanism is faster/cheaper ship-detection at the edge → more tasking, higher utilization