AI spending advances, chip stocks retreat, and the plumbing strains

Spending advances, stocks retreat

Vivek Arya, managing director at BofA Global Research, told a BofA podcast that hyperscaler spending forecasts have been revised upward by about 20% within the quarter. Arya argues the cycle is likely mid-cycle with at least two more years of growth. m:…4401

The demand side explains his confidence. Memory demand outstrips supply by a factor of three to four, he said, and AI contributed about 70 basis points to 2026 GDP growth. Edge AI remains constrained by battery life and compute, so the bulk of complex reasoning stays in the cloud, and open-weight models do not reduce demand for the underlying infrastructure. NeoClouds, meanwhile, service overflow demand from public hyperscalers. m:…4401

Arya also pointed to global competition as a driver: China is investing aggressively in AI infrastructure and robotics, he said, and the semiconductor index has shown an inverse relationship with energy stocks, reflecting the impact of the cost of capital on infrastructure build-outs. The build-out, in other words, is sensitive to financing costs even when demand is strong. m:…4401

The through-line in Arya's account is demand running ahead of supply: memory demand outstrips supply by a factor of three to four, edge devices cannot yet carry the bulk of complex reasoning, and overflow demand spills to NeoClouds servicing public hyperscalers. m:…4401

The streaming bill, line by line

VAST Data's comparison of event streaming costs shows where the plumbing bill lands. Its VAST on Cloud platform folds the VAST Event Broker into its AI operating system, stores message topics natively as tables in VAST DataBase, and lets analytical engines including Trino, Presto and Apache Spark query those tables directly without ETL pipelines or a separate analytical lakehouse. Because the broker uses the Kafka protocol, existing client libraries, APIs and tools keep working. m:…5266

The company says the comparison favors its approach: at a 4 gigabyte-per-second ingestion rate, VAST on Cloud costs $238,387 a month against $1,106,530 a month for Cloud, driven by egress across multiple consumers and post-replication storage charges. The mechanism is protection overhead: VAST requires 2.7% to 20% parity overhead depending on deployment, and additional consumer reads do not add a separate per-gigabyte broker charge. The company says the platform is most cost-effective for retention windows of 1 to 14 days. m:…5266

Standard Apache Kafka clusters use a replication factor of three, the comparison notes, which is the baseline that VAST's parity approach replaces. m:…5266

When the dashboard itself goes down

Lambda's status page reported a high-severity partial outage of its Cloud Dashboard, with users experiencing timeouts when logging in or out of the portal. The company said its teams were investigating a potential issue and working to identify the root cause and mitigate it. No resolution or root cause was stated in the report. m:…6538

Three bills, one week

Put together, the week prices three different layers: capital expenditure still rising, a vendor claiming a wide streaming cost gap by changing how data is protected and read, and a cloud provider's own dashboard failing while customers try to log in. The honest caveats travel with each number: VAST's figures are the company's own comparison at one ingestion rate and one retention assumption, Arya's cycle call is a forecast, and Lambda's report is an open incident, not a postmortem. m:…4401 m:…5266 m:…6538