According to logisticsviewpoints.com, Citadel founder Ken Griffin told the Future of Global Markets Conference in New York on October 20 that companies are questioning whether AI spending is producing enough productivity to justify its cost — a test that transportation, warehousing and supply chain operators now face as they evaluate a wave of AI-enabled software.
Spending is real, returns are uncertain
The scale of the capital flowing into AI infrastructure is striking. Reuters, citing Anthropic’s draft IPO prospectus, reported an $80.7 billion net loss for the first half of 2026, including a $72 billion noncash accounting charge tied to convertible financing instruments.
The same filing disclosed commitments of at least $518 billion over a decade for cloud services, computing and infrastructure. The headline loss should not be read as an equivalent cash operating loss, but the infrastructure obligations are substantial, and they matter well beyond Silicon Valley.
For buyers of transportation management systems, warehouse management systems and supply chain planning tools, the cost and availability of AI models directly influence enterprise software economics. Suppliers are rolling out conversational interfaces, predictive tools, intelligent exception management and autonomous agents — some valuable today, others still demonstrations of what may eventually be possible.
Productivity is measured in operations, not demos
A transportation planner who uses AI to compare carriers in two minutes rather than twenty has achieved a genuine gain in task efficiency. But if freight expense, equipment utilization, staffing and on-time performance do not change, the business impact may be modest.
The more consequential application identifies an impending service failure, evaluates alternatives and helps prevent an expensive expedite or missed delivery. Warehousing presents a similar distinction: AI can recommend slotting changes, anticipate labor requirements or detect congestion, but the measures that matter are throughput, pick productivity, order accuracy, labor cost and equipment utilization.
Supply chain planning adds another layer. A more accurate forecast is useful, but its real value depends on whether it changes purchasing, production, capacity or inventory decisions. If planning cycles and departmental incentives remain unchanged, improved analytics may produce relatively little financial benefit.
The source draws a distinction between faster individual tasks, more productive operating workflows and measurable enterprise returns — related, but not interchangeable. Supply chain technology should ultimately be judged on total implementation cost, exception handling and the human work still required.
Advantage shifts beyond the model
ARC Advisory Group‘s 2026 Warehouse Management Systems MarketMap examines a market where WMS is extending beyond transactional inventory management toward coordination of labor, robotics, automation and operational decisions.
The important differences among suppliers concern execution depth, interoperability, architecture and fit with the operating environment — not simply the number of AI features on a demonstration slide. ARC’s Supply Chain Decision Intelligence MarketMap addresses a related development across planning and execution, focusing on the connection between data, analytics, optimization, workflow orchestration and enterprise systems.
Visibility alone does not resolve a disruption. A decision intelligence capability must help determine what to do and connect that recommendation to an authorized action. When an inbound shipment is delayed, a visibility application identifies the problem, a planning system estimates inventory exposure, a TMS considers alternatives and a warehouse adjusts receiving capacity. The greatest value comes from coordinating those steps rather than placing a separate AI assistant in each application.
As foundation models become more accessible, conversational interfaces and document summaries may stop commanding a premium. Durable differentiation will increasingly come from operational expertise, trusted data, integrated workflows, reliable execution and measurable customer outcomes.
What happens when intelligence gets cheaper?
Griffin offered an analogy: computing chips resemble airline seats because idle capacity has relatively low marginal operating cost. Once infrastructure exists, suppliers have an incentive to sell unused capacity, even below earlier price expectations.
If AI computing supply outpaces demand, prices could fall sharply. The telecommunications buildout around 2000 offers a precedent: excess fiber capacity contributed to severe investor losses while eventually making communications cheaper for users. The same distinction could emerge in AI — a disappointing return for infrastructure investors does not necessarily imply disappointing economics for customers buying computing services.
For logistics, falling model costs could expand the range of viable applications. A workflow economical for a few hundred high-value planning decisions might become practical for thousands of routine shipment, warehouse or procurement decisions, and smaller operators could gain access to capabilities that currently require substantial technology budgets.
That outcome is not guaranteed. Integration, data preparation, security, testing, governance and human oversight still cost money, and more complex AI reasoning can consume more computing resources. A financing correction could also delay data center construction, strain AI providers and disrupt the roadmaps of dependent software suppliers — which is why buyers should evaluate interoperability, vendor resilience and the ability to change underlying models without rebuilding an entire workflow.
The test is the operating statement
Supply chain executives do not need to predict the top of the AI market. They need to establish baselines, measure the full cost of implementation and identify the operating outcomes that must improve.
A transportation application should lower freight or exception costs. A warehouse application should improve throughput and labor performance. A planning application should deliver better inventory, capacity or service results. Griffin may be right about the bubble, and AI may still become a durable part of the operating stack — but the verdict will be written in the operating statement, not the demo.
Source: logisticsviewpoints.com
Compiled from international media by the SCI.AI editorial team.
