Plumbing the
information age

Progressive discovery solves context bloat by letting servers expose only relevant tools to specific agents or roles, preventing token waste and improving reasoning quality. Agent identity ensures audit trails show which system called which tool—critical for compliance in regulated industries. Stateless transport removes session pinning, enabling MCP servers to run cleanly on Kubernetes, Cloud Run, and other horizontally scaled infrastructure. The roadmap also formalizes authorization improvements including issuer validation, issuer-bound credentials, and Client ID Metadata Documents (CIMD) for enterprise registration. For AIOps teams, these changes make MCP suitable for large-scale orchestration where agents need to discover capabilities dynamically, hand off work safely, and maintain provenance across distributed systems.

CoCo automations bring agent execution into the warehouse as a first-class operational primitive—comparable to stored procedures but with agentic reasoning and planning. Each run executes as the user's default role plus all default secondary roles, which creates a privilege model different from traditional service accounts. Security teams should immediately audit EXECUTE AGENT TASK permissions, which are granted to PUBLIC by default. This means any warehouse user can schedule agents with combined role privileges. Operators should treat agent automations like scheduled jobs: enable audit logging, restrict role escalation, validate sandbox isolation for data residency, and implement approval workflows for high-impact tasks. The audit trail design (USER_ID, REQUEST_ID, role tracking) follows NIST NCCoE guidance on agent identity and attribution.

This disclosure validates that MCP adoption has reached production scale in enterprise observability and infrastructure stacks. Deserialization vulnerabilities are a known attack vector for RCE and lateral movement—especially critical in agentic systems where the MCP server mediates agent access to telemetry pipelines, incident management systems, and configuration data. The incident normalizes vulnerability disclosure for the MCP ecosystem, a maturity indicator. Organizations running Splunk MCP servers should immediately check versions and patch or restrict access. More broadly, this signals security teams should budget for regular updates and threat modeling around which agents can access which MCP servers, implement network segmentation between agent runtimes and MCP servers, and validate that authorization policies prevent agents from escalating privileges through server compromise. The timing also reinforces the roadmap priority around enterprise authorization and audit logging.

Rebellions, valued at $2.3B, has built inference-optimized silicon competing in the dense field of post-Nvidia accelerators. The reported talks—whether acquisition, partnership, or pure investment—reflect Nvidia's strategic approach to ecosystem control: when a startup threatens fragmentation or offers complementary capability that Nvidia's existing roadmap doesn't cover, Nvidia moves to acquire, integrate, or co-invest rather than compete head-to-head. For infrastructure teams, this signals continued consolidation in the inference-chip space. The practical question: if Rebellions is acquired or deeply integrated, will Nvidia open its APIs and driver stacks, or does Rebellions become Nvidia-internal IP? For teams evaluating non-Nvidia inference silicon, watch announcements about driver support, ROCm parity, and go-to-market independence.

The government will invest about 2.3 billion reais, dividing funding between a Chinese technology companies project and a separate competitive tender. Just over half the total, 1.3 billion reais ($251 million), will finance supercomputing infrastructure in Rio de Janeiro, with Huawei and iFlytek participating in a project intended primarily to support large language model development. The remaining 1 billion reais will fund a separate supercomputer through public tender, expected to rank among the world's 10 most powerful AI processing systems, installed in Rio Grande do Norte where energy potential influenced the government's choice. iFlytek's overseas revenue grew 160% year-over-year in the first half of 2026, and its technological breakthroughs with Huawei in domestic computing power and large model integration form core competitiveness behind this partnership. This signals structural geopolitical shift: developing economies now view compute infrastructure as critical strategic assets rather than commodity technology purchases, and Brazil's dual-vendor approach preserves flexibility while building domestic capacity.