THOUGHT OF THE DAY
AI infrastructure value is broadening from accelerators to data movement. Credo reported 115% year-over-year revenue growth and forecast more than $600 million in optical revenue, while Astera Labs reported 98.5% growth. The signal is that high-speed connectivity is becoming a material revenue pool in AI clusters, not just a supporting component alongside GPUs.
A long-term contract puts a concrete price on sovereign GPU capacity. Alpha Compute announced an $81.7 million, four-year agreement for a U.S.-based cluster with 592 NVIDIA B300 GPUs, including a 25% upfront payment. That commitment supports demand for domestically controlled compute, though reliance on one marketplace leaves the provider exposed to customer concentration.
Robot training is targeting the cost of moving from simulation to real-world deployment. Hesai’s Kosmo platform converts scanned environments into simulation-ready 3D worlds; a Sharpa robot reportedly achieved 99% task success after 15 minutes of scanning. If results generalize beyond the proof of concept, faster environment capture could reduce the time and expense of training robots for specific worksites.
COMPUTE & SEMICONDUCTORS
- Micron (MU) forecast 351% revenue growth and 700% earnings growth, citing long-term contracts, an HBM4 ramp, and fab expansion. These are forecasts, but they reinforce expectations for strong demand across AI memory.
- Credo (CRDO) reported 115% year-over-year revenue growth and forecast more than $600 million in optical revenue. The outlook points to rising demand for optical links as AI clusters scale.
- Astera Labs (ALAB) reported 98.5% revenue growth and a 75.1% gross margin, adding evidence of demand for components that move data between accelerators and memory.
- Alpha Compute’s announced contract covers 592 liquid-cooled NVIDIA (NVDA) B300 GPUs, with 576 designated for production and 16 as spares. The $20.2 million upfront payment improves initial contract visibility, while the single-customer exposure remains a meaningful risk.
DATA CENTERS & INFRASTRUCTURE
- Pegatron and Compal are showcasing liquid-cooled, rack-scale systems designed to accelerate deployment of AI infrastructure. The competitive focus is shifting toward integrated, ready-to-deploy systems rather than standalone server components.
- LITEON’s 800VDC architecture and 2.6MW enterprise cooling distribution unit illustrate the engineering work required to support higher-power AI racks. These are infrastructure design signals, not evidence that deployment bottlenecks have been resolved.
- CoreWeave’s reported 13% share decline and 14.5% bond yields highlight the financing risk in capital-intensive compute expansion, despite the company’s reported $8.5 billion in pre-sold capacity.
ROBOTICS & PHYSICAL AI
- Hesai’s Kosmo platform uses scans of real environments to create simulation-ready 3D settings. A Sharpa robot reportedly reached 99% task success after 15 minutes of scanning. The result is promising, but a single proof of concept does not establish repeatable performance across sites or tasks.
- NEURA Robotics demonstrated a 4NE1 humanoid completing a laundry sequence autonomously, while its Neuron AI system combines sensing, memory, and real-time reasoning. The commercial question is whether these capabilities can deliver reliable performance in sustained, varied workflows.
- ABB and NVIDIA are developing physics-accurate digital twins for robotics, with reported 99% sim-to-real fidelity. If validated in production settings, better simulation could reduce the cost of robot development and deployment.
ADOPTION & MONETIZATION
- Palantir (PLTR) was reported to have delivered 149% year-over-year revenue growth, a strong commercial signal for AI software. The summary does not specify the revenue base or contribution from AI products, so the figure alone does not establish how much growth is directly attributable to AI.
- Oracle (ORCL) NetSuite launched a free AI-powered finance platform for startups through Quick Start. The move lowers the entry barrier for small businesses, while leaving future conversion to paid products unproven.
- Voya (VOYA) and UnqorkAI reported 10% year-over-year growth in fee-based revenue tied to AI. That points to potential revenue expansion, not just cost savings, though the summary provides limited detail on attribution.
POSITIONING IDEAS
Bullish
- Micron (MU): Long-term contracts and an HBM4 ramp underpin a sharply higher revenue and earnings forecast. The key validation point is whether those forecasts translate into delivered growth.
- Credo (CRDO) and Astera Labs (ALAB): Reported growth and Credo’s optical revenue forecast support exposure to AI-cluster connectivity beyond accelerators.
Bearish
- CoreWeave (CRWV): Reported bond yields of 14.5% and a 13% share decline signal high financing costs for a capital-intensive business. Pre-sold capacity offers demand visibility, but it does not eliminate the risk that debt costs erode returns.