THOUGHT OF THE DAY
Agent autonomy is turning governance into a core infrastructure requirement. Menlo Ventures’ 2026 report says agents generate 80% of AI tokens, while Snowflake launched a gateway for access controls, auditing, and cost limits alongside a system for standardizing business definitions. If agent activity keeps expanding, control over tools, data, and spend becomes a prerequisite for scaling enterprise usage—not just a compliance add-on.
Inference workloads are opening room for specialized hardware alongside GPUs. SambaNova and Supermicro (SMCI) plan to deploy 10,000 SN50 chips, with the reconfigurable dataflow units targeting decode-heavy inference. The plan is a meaningful signal of demand for workload-specific systems, but its investment significance depends on execution and demonstrated throughput and power economics; it does not yet establish a broad shift away from GPUs.
COMPUTE & SEMICONDUCTORS
Purpose-built inference
SambaNova’s planned 10,000-chip SN50 deployment through Supermicro’s data-center systems is a notable test of specialized inference at scale. The chips target decode workloads, which can strain latency and throughput as models generate responses. SambaNova positions the system as complementary to GPUs, not a replacement; shipment progress and measured performance per watt will determine whether this creates a repeatable alternative for inference capacity.
ADOPTION & MONETIZATION
Enterprise agent governance
Menlo Ventures’ report puts year-to-date AI spending at $176.8 billion and says agents account for 80% of AI tokens—seven times the human share. Token volume signals rising workload intensity, but does not by itself prove corresponding revenue or return on investment.
Snowflake (SNOW) introduced Cortex AI Gateway for policy enforcement, auditing, model routing, and cost caps, alongside Business Ontology to standardize business terms across agents. The company says Glean, Kargo, dentsu, and Merkle are using the platform in production. The commercial opportunity is to make Snowflake’s data and governance layer central to agent deployments across models and clouds.
POSITIONING IDEAS
Bullish
- Snowflake (SNOW): Cortex AI Gateway and Business Ontology target the governance and business-context requirements that can constrain enterprise agent rollouts. Named production users offer an initial adoption signal, though the summaries provide no revenue contribution or customer-scale data.
- Supermicro (SMCI): The planned integration of 10,000 SambaNova chips supports demand for its AI systems. Treat this as an execution-dependent catalyst: deployment and performance evidence are still needed.