Why Enterprise AI Is Moving Slower Than You Think

2026-07-23 · Source: The Data Exchange · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, extended

Summary

Denise Teng, a partner at Gradient Ventures, discusses the slow reality of enterprise AI adoption, attributing it primarily to security, data privacy, and compliance hurdles. She highlights the increasing importance of open-source and open-weights models, noting their performance gap with frontier closed-source models has shrunk to 3-6 months, offering secure, private deployment options and exerting price pressure. Teng emphasizes the future role of smaller, custom models, often post-trained with reinforcement learning, to tackle complex tasks. However, she points out that RL remains challenging due to data bottlenecks, not just UX. The conversation also covers emerging AI agent infrastructure, including sandbox runtime environments, agent identity management, and AI memory layers for institutional knowledge. Vertical AI solutions in legal, healthcare, and finance are gaining traction. Inference infrastructure optimization is critical given skyrocketing demand and hidden costs, prompting a need for ROI tracking. Teng expresses moderate concern about the AI bubble (a "six" out of ten), citing company proliferation and security threats, while Ben Lorica is more worried (an "eight or nine") about the economics of AI data center buildouts and demanding fundraising metrics.

Key takeaway

For Directors of AI/ML evaluating enterprise AI strategies, recognize that security, privacy, and compliance are paramount, driving slower adoption but also the demand for open-weights models and robust infrastructure. Prioritize investments in agent-specific infrastructure like sandboxes and AI memory layers to manage identity, access, and institutional knowledge effectively. Focus on optimizing inference costs and developing granular ROI tracking for token consumption, as these hidden costs will increasingly impact budget decisions.

Key insights

Enterprise AI adoption is slow due to security and data privacy, but open-weights models and specialized infrastructure are accelerating progress.

Principles

• Open weights reduce supplier risk and price pressure.
• Smaller, custom models will proliferate for complex tasks.
• Data readiness is the bottleneck for RL adoption.

Method

Enterprises can bridge performance gaps by post-training open-source models using reinforcement learning, producing smaller custom models for specific tasks.

In practice

• Implement sandbox runtime environments for agents.
• Develop AI memory layers for institutional knowledge.
• Focus on vertical AI solutions for domain-specific problems.

Topics

• Enterprise AI Adoption
• Open-Weights Models
• AI Agent Infrastructure
• Reinforcement Learning
• Inference Optimization
• AI Data Centers
• Data Privacy & Security

Best for: CTO, VP of Engineering/Data, AI Architect, Director of AI/ML, Investor, Consultant

Related on AIssential

• Together AI positions open-weight AI models as the enterprise moat for cost, control and IP — Open-weight AI models offer enterprises superior cost, control, and IP protection for agentic AI workloads.
• RAISE Summit 2026: Goodbye AI Demos and Vibes, Hello Buildouts and Megawatts — AI's evolution demands massive infrastructure, low latency, and robust security for enterprise-wide autonomous agent adoption.
• The State of Enterprise AI Adoption — Enterprise AI adoption faces cost and vendor lock-in issues, driving a need for model agnosticism and diverse model access.
• How Companies Are Becoming AI Token Efficient — Enterprise AI success now hinges on token efficiency and "dollars per outcome," not just raw intelligence.
• [AINews] not much happened today — AI progress spans non-invasive BCIs, efficient inference, agent orchestration, and open-weight model commercialization, driven by infrastructure and evaluation.
• Why Data‑Driven Efforts Stall in Fragmented Environments - with Jason Loomis of Freshworks — Enterprise AI adoption requires a sequenced approach, prioritizing regulatory compliance, data trust, and AI-specific security frameworks to balance speed and safety.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Data Exchange.