Power, Scale, and the Price of AI

What happened
A new study in Nanoenergy Advances revealed that atom-thin carbon films (like graphene) can act as memristors—tiny memory devices that remember electrical flow and can be programmed with light. These “photomemristors” could integrate memory and computation, reducing the need to shuttle data between separate chips.

Why it matters
AI’s energy appetite is a growing problem. This breakthrough could enable AI chips that are dramatically more efficient, especially for edge devices and vision applications. By merging memory and processing, the technology could help scale AI without unsustainable power demands.

What’s next
Still in the research phase, the next steps are proving reliability and integrating with standard chipmaking. If successful, this could usher in a new era of low-power AI hardware for both cloud and edge computing.

What happened
A new industry analysis found enterprise adoption of agentic AI is exploding—Gartner projects an 800% jump in apps using AI agents this year. But the report also flagged rising security incidents, including prompt-injection attacks and credential misuse, and highlighted the lack of robust AI governance in most organizations.

Why it matters
Agentic AI is reshaping enterprise IT, but it’s also creating new vulnerabilities. With machine identities now outnumbering humans 82 to 1, insider threats and data breaches are a growing concern. The need for strong AI security and governance has never been more urgent.

What’s next
Look for a wave of investment in AI security tools, new governance frameworks, and increased regulatory scrutiny as organizations race to balance productivity gains with risk management.

What happened 
SiliconANGLE reported on RingCentral’s 2026 Agentic AI Trends study, which found that 97 % of surveyed organizations are already using some form of AI. Generative AI is the most common use case (77 %), followed by predictive analytics (54 %) and process automation (53 %). Early deployments move quickly: 69 % of decision‑makers got their first agentic AI initiative live within a year, with 77 % seeing a return on investment and 92 % satisfied with their results. However, as companies attempt to scale, they encounter friction. Roughly 61 % report productivity gains and 58 % see faster workflows, yet 40 % of organizations have paused or canceled at least one AI project due to integration complexity, internal resistance and unclear ROI.

Why it matters 
The data suggest that AI agents are transitioning from isolated experiments to system‑level architecture. High adoption and tangible ROI show enterprises view agentic AI as essential, but the high cancellation rate underscores the difficulty of stitching agents across fragmented workflows. Success depends not on bigger models but on orchestration — connecting agents, people and tools so context flows seamlessly.

What’s next 
Expect a focus on integration layers, observability and governance. Companies will demand platforms that coordinate multiple agents, capture “dark data” from voice and video, and handle exceptions without losing context. Vendors that can demonstrate reliable orchestration and clear business outcomes will gain an edge, while organizations with fragmented systems will struggle until they invest in integration.

What happened 
Manufacturing Dive described an inflection point in physical AI. At CES 2026, Nvidia CEO Jensen Huang called the moment the “ChatGPT moment for physical AI”. Hyundai Motor Group unveiled its Atlas humanoid robot for production settings and plans to deploy it across operations. A Deloitte survey of 3 200 global business leaders found that 58 % are already using physical AI (e.g., robotic arms, cobots and smart monitoring) and 80 % expect to use it within two years; 15 % plan extensive use and 3 % will fully integrate it. Executives like Path Robotics’ CEO anticipate fast adoption but warn that moving from demo to fully functional systems will take time. The article also notes that manufacturers are turning to AI agents and IoT sensors to monitor equipment and supply chains, but increased data exposure raises cybersecurity risks.

Why it matters 
Physical AI is moving from pilot projects to scalable deployments. High adoption intent signals that robots and AI‑driven sensors will become core to manufacturing, logistics and other industries. Yet reliability and security remain critical barriers; autonomous systems must achieve near‑perfect uptime to justify investment. As operations digitize, more data flows expose companies to ransomware and other cyber threats, prompting a need for AI‑enhanced cybersecurity.

What’s next 
Expect increased testing of humanoid robots like Atlas in industrial settings and more integration of physical AI with IoT networks. Firms will focus on improving dexterity, pressure control and perception so robots can handle unstructured tasks. Adoption will accelerate once reliability approaches 99 % uptime and companies are confident that AI agents and sensors can enhance productivity without introducing unacceptable cyber risk.

What happened 
CX Dive reported that a new Gartner prediction challenges the assumption that generative AI will cut customer service costs. The consultancy forecasts that by 2030 the cost per resolution for generative‑AI systems will exceed $3 — more than many offshore human agents. Rising data‑center expenses, AI vendors shifting toward profitability and more complex use cases are expected to drive costs upward. Gartner analysts caution that full automation will be “prohibitively expensive” for most organizations; instead, AI should enhance customer engagement rather than serve purely as a cost‑cutting tool. Other researchers note that AI adoption comes with hidden costs, such as managing access credentials and acquiring specialized data sets.

Why it matters 
Businesses often justify AI investments by assuming they’ll replace more expensive human labour. Gartner’s analysis suggests the economics may not be so favourable: generative AI requires expensive infrastructure and ongoing data stewardship. If costs per resolution surpass offshore labour rates, companies may need to rethink their ROI assumptions and focus on non‑financial benefits like customer satisfaction and loyalty.

What’s next 
Expect a more nuanced conversation about where generative AI delivers value. Rather than aiming for full automation, leading organizations will deploy AI to augment human agents — handling routine queries, providing context and freeing staff for higher‑value interactions. Vendors will need to offer transparent cost structures and address issues like data‑center efficiency and model governance to make AI economically viable over the long term.

AI is scaling fast — but power, security, integration, and cost are becoming the real constraints. The next winners won’t just build smarter systems; they’ll build ones that are efficient, secure, and economically sustainable.