The Futurum Group research enterprise IT leaders actually read.
Volume 341, Issue 3 · 9 articles from 2 issues
Futurum argues Taiwan's confirmed near-autonomous AI attack moves agentic threats from theoretical to operational.
Futurum Group analyzes what Taiwan's Ministry of Digital Affairs confirmed in July 2026: a near-autonomous AI cyber attack in which autonomous agents mapped 21 connected government systems.
The analysis argues agentic AI threats are no longer theoretical now that a state has publicly validated one. The confirmation reshapes how organizations should assess autonomous AI risk in their own environments.
Futurum examines what the incident implies for critical cybersecurity strategy going forward. The piece is aimed at security leaders whose planning still treats autonomous offensive AI as a future scenario.
Futurum's Tom Hollingsworth argues 1.6 Terabit Ethernet forces simultaneous rethinking of topology, power and cooling.
Futurum Group analyst Tom Hollingsworth examines the architectural consequences of moving to 1.6 Terabit Ethernet. The transition is not a simple speed upgrade because it forces infrastructure teams to rethink physical topology, power delivery and cooling systems simultaneously.
Each of those changes would normally be handled on its own cycle. The report argues most facilities are not ready for a change that touches all three at once.
It examines what preparation is required for AI deployments that depend on the higher speeds.
Enterprise AI systems are shifting from providing assistance to autonomously performing business operations.
The agentic shift represents a fundamental transformation in how enterprises deploy AI, moving beyond informational tools toward systems that actively execute business processes.
Key developments include Microsoft's Project Perception for autonomous action, TeamViewer and ServiceNow's autonomous IT operations, and Atlassian's agent-enabled Jira strategy.
This transition demands enhanced data security, governance, and accountability as autonomous agents operate at machine speed. Infrastructure investments from HPE, Intel, and Cadence reflect growing compute demands, while organizations across IT operations, customer engagement, and public services redesign workflows around systems capable of independent action.
Enterprise AI governance emerges as a critical priority as agent proliferation creates coordination challenges.
The Futurum Group's weekly briefing examines how agentic AI deployment is forcing enterprises to address governance, accountability, and data access control.
Organizations face a narrow 12 to 18 month window to align platforms, data, AI systems, and teams around shared outcomes before fragmentation becomes permanent.
The infrastructure stack is evolving to handle agentic workloads, with companies such as Marvell introducing memory expansion using CXL technology. Key themes include agent sprawl management, orchestration maturity models, and the shift from data consolidation to governed, selective data access for AI applications.
Enterprise AI shifts focus from raw capability to infrastructure costs, pricing models, and measurable outcomes.
The AI market is entering a pragmatic phase where infrastructure capital, model access, and operational discipline matter as much as technical capability.
Market data shows 57.7% of enterprise buyers prefer consumption-based or outcomes-based pricing for generative AI over per-user-per-month models, signaling demand for measurable value.
Key developments include Intel's $20 billion stock offering for 14A fab tooling, NVIDIA's DSX reference architecture, and expanding AI deployment in Malaysia and Indonesia.
Applications are evolving across Oracle's agentic HCM capabilities, Intuit's AI-native mid-market suite, and consumer hardware from Meta, Google, and Samsung. Success increasingly depends on controlling infrastructure and delivering reliable business outcomes.
Futurum argues enterprises are shifting from costly cloud AI APIs to localized, quantized open-weights models.
This Futurum Group Utilizing AI episode, featuring Nick Patience and Brad Shimmin, examines enterprises moving away from frontier cloud AI APIs. Runaway token costs are the immediate driver of the shift.
Organizations are turning to localized, quantized open-weights models that run on infrastructure they control. Data sovereignty is a second driver, since local inference keeps data inside defined boundaries.
Operational control is the third, giving teams predictability that metered API access does not provide.
Futurum argues fragmented observability inflates MTTR and hides cloud spend until an outage or budget shock exposes it.
Futurum Group argues that fragmented observability across multi-cloud environments is a financial exposure rather than only an operational inconvenience.
Fragmentation silently inflates mean time to resolution because engineers must correlate across disconnected tools during an incident. It also obscures cloud spend, which stays invisible until an outage or a budget overrun forces attention. The analysis uses Stratpoint's unified observability stack as a worked example, citing 99.8% uptime reliability.
The framing is aimed at finance and operations leaders who treat observability as a monitoring line item.
Futurum analyzes AWS Context and Context Ontology as a governed knowledge graph that ends the semantically naive agent.
Futurum Group analyst Brad Shimmin examines AWS Context, announced in preview in June, alongside the more recently introduced Context Ontology.
he analysis frames these as AWS collapsing the semantic divide between raw enterprise data and what an agent can actually reason over.
A naive agent, in this framing, is one that retrieves data without understanding what any of it means.
A governed knowledge graph supplies that meaning consistently across systems. Shimmin assesses what the combination delivers for AI systems and where the approach still has gaps.
Futurum finds buyers reject per-seat pricing for core software but demand it for AI, forcing vendors to split architectures.
Futurum Group analyst Keith Kirkpatrick examines enterprise software pricing in the AI era. Buyer data cuts against the prevailing consensus that per-seat pricing is dead: most current commentary treats seat-based models as obsolete, but the research shows a more complicated picture.
Buyers reject per-seat pricing for core software while demanding it for AI capabilities, where predictability matters more than efficiency.
The report argues vendors therefore cannot run one commercial model across both. It concludes vendors must split their pricing architectures to stay competitive rather than converging on a single approach.
More from Futurum Group · All IT section newsletters · This week