THOUGHT OF THE DAY
AI compute economics are shifting toward useful throughput per watt
Cerebras’ CS-3 is reportedly available through AWS Bedrock, with a claimed 15x speedup, while Karman says its power-orchestration platform can raise compute capacity by 60% within the same power budget. Together, these developments point to two routes beyond simply adding GPUs: specialized accelerators and better utilization of installed power. The investment test is whether those gains hold on production workloads and improve cost per unit of inference.
AI chip complexity is lifting the test-and-validation layer
Teradyne reports that AI-driven revenue now exceeds 60% of its business, alongside 128% year-over-year growth in semiconductor test revenue. That suggests AI investment is supporting suppliers that qualify and validate chips, not only those that design or manufacture them. If chip complexity and production volumes keep rising, test equipment could capture a larger share of semiconductor capital spending.
COMPUTE & SEMICONDUCTORS
- AWS has integrated Cerebras’ wafer-scale CS-3 into Bedrock. The reported 15x speedup could make alternative architectures more accessible to cloud customers, but the performance claim needs workload-level and cost comparisons before it can be treated as a broad substitute for NVIDIA GPUs.
- Teradyne is benefiting from AI-related demand in semiconductor testing: the company says AI-driven revenue exceeds 60% of its business and test revenue rose 128% year over year. This is a useful signal that AI demand is reaching equipment suppliers beyond the accelerator vendors.
DATA CENTERS & INFRASTRUCTURE
- Karman says its power-orchestration platform can increase AI compute capacity by 60% within an unchanged power budget. If validated in operating facilities, software-led utilization gains could defer some demand for new power and data-center capacity.
- Ciena reportedly has a backlog above $10 billion, while Corning has multiyear agreements with Amazon, Meta, and Verizon. These signals support continued demand for the optical and fiber infrastructure needed to connect AI systems, though backlog and contracts still need to convert into revenue.
- Celestica expects strong growth in its AI-driven cloud and connectivity systems business. That supports the case for contract manufacturers supplying AI infrastructure, but execution and customer concentration remain important risks.
ROBOTICS & PHYSICAL AI
- Microbot Medical’s LIBERTY robotic system has received FDA clearance, and the company is engaging with hospitals. Clearance removes an important regulatory barrier, but commercial adoption will depend on hospital purchasing, procedure volume, and demonstrated clinical utility.
ADOPTION & MONETIZATION
- Paychex has launched its WISE AI platform, Spotify is using AI-powered taste profiles, and ServiceNow is applying AI in defense operations. These launches show AI reaching HR, consumer personalization, and mission-critical workflows, but the summaries provide no evidence yet of material incremental revenue or productivity gains.
POSITIONING IDEAS
Bullish
- Teradyne (TER): AI-related revenue and semiconductor test growth support exposure to the expanding chip qualification and validation cycle.
- Ciena (CIEN) and Corning (GLW): Reported backlog and multiyear agreements support demand for AI networking and fiber infrastructure.
Bearish
- Leveraged AI data-center developers, including Applied Digital (APLD): Reported debt and losses raise the risk that buildout costs outrun cash generation. This is a financing and execution concern, not evidence that demand for AI compute has weakened.