Production AI Gets Budgets, Gates and a Paper Trail
Snowflake, H2O.ai and Dynatrace all shipped control layers on October 29: a gateway for AI spend, a hardened platform for regulated work, and observability across agent frameworks.
- Snowflake released Cortex AI Gateway into public preview as a centralized entry point for AI traffic with cost control. m:…5916
- H2O-3 Secure is built to FedRAMP High engineering standards and supports SOC 2, ISO 27001 and ISO 42001, according to H2O.ai. Self-reported m:…4871
- Dynatrace expanded AI Observability ingestion across coding agents and OpenInference-instrumented frameworks, covering more than 33 frameworks. Self-reported m:…5552
- Dynatrace will end support for OpenLLMetry on January 31, 2027. m:…5552
A gateway in front of the spend
Snowflake released Cortex AI Gateway into public preview as a centralized entry point for AI traffic, designed to give administrators cost control over inference requests. m:…5916
The gateway specification allows administrators to define an explicit list of models that the gateway will serve, and a usage history view provides a granular credit breakdown for requests that touch multiple models. m:…5916
That allowlist is the practical core of the gateway: instead of tracking model access team by team, an administrator sets which models the gateway will serve, and the usage history view then breaks down credits even when a single request touches multiple models. m:…5916
A MONITOR role allows users to query gateway usage data without the ability to send inference requests, separating oversight from spending in the design. m:…5916
The usage history view retains 365 days of request history and is available in both account-usage and organization-usage schemas, while shared resource budgets can trigger stored procedures at specific threshold percentages. m:…5916
Per-user quotas can block AI requests within minutes of reaching a limit, while budget evaluation for the gateway occurs periodically, taking up to 6.5 hours, or one hour with the low-latency option, so quotas are the faster brake and budgets are the slower alarm. m:…5916
A hardened edition for regulated work
H2O.ai launched H2O-3 Secure, a commercially licensed, production-grade edition of its open-source machine learning platform H2O-3, built to FedRAMP High engineering standards, the company says. m:…4871
The company says the edition supports production-grade deployment on Hadoop, Spark and Kubernetes, and supports compliance frameworks including SOC 2, ISO 27001 and ISO 42001. m:…4871
H2O.ai positions the edition for financial services, insurance, telecommunications, healthcare and the public sector, and its chief executive is quoted in the announcement as saying it provides security maintenance, governance and accountability for production models. m:…4871
The edition also includes automated model documentation through H2O AutoDoc and built-in drift monitoring, and the company says its MOJO artifacts score in a few hundred microseconds per prediction. m:…4871
H2O.ai notes that the open-source edition remains free under Apache 2.0, and says it is trusted by more than 20,000 organizations, including half of the Fortune 500, which frames the Secure edition as an upgrade path rather than a replacement. m:…4871
Seeing what agents actually did
Dynatrace is expanding AI Observability data ingestion across coding agents, Python-based language-model applications and frameworks instrumented with OpenInference, which the company says auto-instruments more than 33 frameworks. m:…5552
OneAgent now instruments Python-based language-model applications directly, capturing logs, traces and language-model-specific telemetry without code changes, according to Dynatrace. m:…5552
Dynatrace is also ending support for OpenLLMetry on January 31, 2027, a date teams using that route will need to plan around as they consolidate observability. m:…5552
Dynatrace has also released experimental Python sensors for providers including Anthropic, OpenAI, Google ADK, Agent SDK, Microsoft Foundry, Mistral AI, LangChain and others, and now supports Claude Agent and Claude Platform tracing as a listed observability partner on the Claude Platform. m:…5552
Beginning February 1, 2027, Dynatrace AI Observability will support three ingest paths: OpenTelemetry-based instrumentation, OpenInference-instrumented applications and OneAgent, giving teams a date to consolidate around. m:…5552
Taken together, the three launches describe the same buying question from three sides: who may spend, what may run, and what can be traced afterwards. m:…5916 m:…4871 m:…5552
What is still unresolved
- No pricing is given for H2O-3 Secure or for Dynatrace's expanded observability in the cited material.
- Whether Snowflake's gateway covers models outside its own platform is not detailed in the cited material.
- No customer results are given for the gateway or the hardened edition beyond the announcements themselves.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.