Rail Freight Slumps: How AI Stabilizes Volume

U.S. rail freight just logged its eighth straight week of year-over-year declines, with total volume down 1.4% for the week ending Dec. 13, 2025. That’s not a rounding error—especially in December, when shippers are trying to clear inventory, hit service targets, and close the year without ugly surprises.

Here’s the part most teams miss: a bad month in rail doesn’t automatically mean “rail is down.” For the first 50 weeks of 2025, overall U.S. rail volume is still up 1.7% year over year. Weekly softness and year-to-date growth can coexist—and when they do, it usually signals a planning problem more than a capacity problem.

This post is part of our AI in Transportation & Logistics series, and I’m going to take a firm stance: rail volatility is a forecasting and network-orchestration problem first. AI can help—but only if you aim it at the right operational decisions: demand sensing, intermodal conversion, lane-level capacity protection, and exception management.

What the latest rail numbers really say (and what they don’t)

The direct read is simple: rail volumes dipped again, and both carloads and intermodal participated in the decline.

For the week ending Dec. 13, 2025:

• Total U.S. rail volume: 518,904 carloads + intermodal units (-1.4% y/y)
• Carloads: 224,620 (-1.7% y/y)
• Intermodal containers/trailers: 294,284 (-1.2% y/y)

Year-to-date through 50 weeks:

• Total: 24,685,267 units (+1.7% y/y)
• Carloads: 11,113,752 (+1.8% y/y)
• Intermodal: 13,571,515 (+1.7% y/y)

North America broadly mirrors this pattern:

• Weekly North America: 705,800 units (-1.1% y/y)
• YTD North America: 33,973,610 units (+1.6% y/y)

Mexico stands out (and not in a good way):

• Weekly Mexico carloads: 12,944 (-11.6% y/y)
• Weekly Mexico intermodal: 15,375 (+15% y/y)
• YTD Mexico total: -5.6%

The operational takeaway

Eight down weeks in a row doesn’t mean demand is gone. It often means demand is shifting faster than your planning cycle can absorb—between modes, corridors, ports, suppliers, SKUs, and even customer service expectations.

And December amplifies every weakness:

• Retail and CPG are correcting forecasts after peak season
• Manufacturing schedules get choppy around holiday shutdowns
• Weather disruptions create ripple effects across dwell and interchange
• Carriers and railroads tighten operating plans to protect service

If your logistics team is still doing “monthly planning + weekly firefighting,” you’re structurally late.

Why rail volatility is an AI problem (not just a market problem)

The best use of AI in freight isn’t predicting the future perfectly. It’s updating decisions faster than humans can—using imperfect signals—so you stop getting surprised.

Rail and intermodal are especially sensitive because:

• Lead times are longer (you commit earlier)
• Service variability compounds across terminals and interchanges
• Small forecast errors turn into missed cutoffs, rolled containers, and stockouts

A blunt truth: most shippers don’t lose money because they chose rail. They lose money because they chose rail with the wrong assumptions.

AI-driven logistics optimization helps you manage those assumptions in real time:

• What demand is actually doing (not what last month’s S&OP deck said)
• Where capacity is tightening (before tenders start failing)
• Which lanes are likely to miss service (before customer OTIF gets hit)

Demand forecasting vs. demand sensing

Traditional forecasting works on history and planned inputs. Demand sensing works on fresh, high-frequency signals, such as:

• Order intake and cancellations
• Customer POS or consumption proxies
• Inventory positions and days-of-supply
• Port throughput and appointment data
• Spot/contract spread signals in adjacent modes

Rail’s eight-week slide is exactly the kind of pattern demand sensing is built to catch early—and to translate into action (mode shifts, pre-build inventory, reposition equipment, protect key lanes).

Four AI plays that stabilize rail performance (with real decisions attached)

AI only matters when it changes what you do on Monday morning. These are the four plays I’d prioritize for rail-heavy or intermodal-heavy networks.

1) Predictive lane planning that protects the lanes that matter

Answer first: Use AI to forecast volume at the lane-and-customer level, then allocate rail/intermodal capacity where service risk is lowest and business value is highest.

Many networks plan capacity at a coarse level (“Midwest to Southeast”) and wonder why one customer gets rolled while another glides through. AI forecasting can get granular enough to support capacity protection rules, such as:

• Reserve premium intermodal on lanes with historically tight cutoff windows
• Shift specific SKUs (high margin, high penalty) to the most reliable services
• Identify lanes where rail dwell variability is rising and pre-emptively hedge

What changes operationally:

• You stop spreading capacity evenly
• You start assigning capacity intentionally

Snippet-worthy line: “Stability comes from choosing where you’re allowed to be unstable.”

2) Load balancing across rail, intermodal, and truck—before disruptions force you

Answer first: AI can recommend mode and route mixes that minimize total landed cost while meeting service targets, updating as constraints change.

Intermodal down 1.2% in the latest week is not just “intermodal demand.” It can reflect shippers quietly opting back to truck on critical lanes, or failing to execute intermodal plans due to timing and terminal constraints.

A practical AI optimization setup uses:

• Service-time distributions (not single-point transit times)
• Cost curves (including detention, storage, accessorials)
• Emissions constraints (for sustainability reporting)
• Capacity constraints (equipment, dray, rail schedules)

Outputs that matter:

• Lane-by-lane recommended rail share vs truck share
• A “hedge plan” when probability of delay crosses a threshold
• Targeted pre-booking for peak weeks

This matters because mode shifts are cheapest when they’re planned and brutal when they’re reactive.

3) Exception prediction: stop treating delays like surprises

Answer first: AI can flag shipments likely to miss appointments days earlier, giving you time to reroute, expedite, or reset customer expectations.

Rail service often fails in predictable ways:

• Congestion at specific terminals
• Weather corridors that repeatedly create dwell spikes
• Interchange points where handoffs add variance

With the right telemetry (waybill events, EDI milestones, terminal activity signals, dray status), models can produce an ETA confidence score and a “reason code” style explanation:

• “High miss risk due to terminal dwell trend + late gate-in risk”

Then your team can act:

• Proactively reschedule receiving appointments
• Flip a subset of loads to team truck or expedited intermodal
• Reprioritize dray capacity to protect the most at-risk containers

If you’re measuring OTIF, this is where you win it back.

4) Network resilience planning for cross-border and Mexico variability

Answer first: When a region shows structural softness (like Mexico’s -5.6% YTD), AI helps you separate demand change from execution change and respond with the right playbook.

Mexico’s weekly carloads were down 11.6%, while intermodal units were up 15%. That divergence can happen when:

• Shippers consolidate into containers
• Certain commodities slow while consumer goods remain steady
• Border and terminal dynamics shift the optimal mode

A practical AI approach:

• Model cross-border flows with border-specific lead-time variance
• Simulate inventory impacts of 1–5 day delays
• Identify which SKUs can tolerate rail variability vs which need buffer stock

The goal isn’t to “predict Mexico.” The goal is to design a network that doesn’t panic when Mexico moves.

“People also ask” answers your team can use internally

Is rail freight declining overall in 2025?

Not overall. Recent weeks are down year over year, but 2025 volume is still up 1.7% through 50 weeks. The story is volatility, not collapse.

Why would weekly rail volumes drop while YTD is up?

Because demand and routing choices shift during the year. Peak periods earlier can offset late-year softness. Weekly declines often reflect short-term inventory corrections, mode switching, and service-risk hedging.

Where does AI help most in rail and intermodal logistics?

AI delivers the most value in demand sensing, lane-level capacity allocation, exception prediction (ETA risk), and mode optimization—the decisions that reduce costly surprises.

A practical 30-day rollout plan (no science project)

If you’re trying to turn “AI in logistics” into measurable outcomes quickly, here’s what works.

Week 1: Pick the decisions, not the tools

Choose one of these decision points:

• Weekly intermodal plan by lane
• Dray capacity allocation by terminal
• Customer appointment risk management

Define success in numbers (examples):

• Reduce rolled loads by 15%
• Improve OTIF by 2 points
• Cut expedite spend by 10%

Week 2: Assemble the minimum data set

You usually need:

• Shipment history by lane/customer
• Planned vs actual transit milestones
• Costs (linehaul + accessorials)
• Basic operational constraints (cutoffs, appointment windows)

Week 3: Build a pilot model and force it into workflow

The mistake is treating AI as a dashboard. Make it a recommendation engine:

• “Shift 12 loads from lane A intermodal to truck this week”
• “Pre-book intermodal on lane B; service risk is low and savings are high”

Week 4: Measure, tune, and expand to the next decision

Keep it honest:

• Track adoption rate (did planners follow recommendations?)
• Track outcomes (service + cost + variability)
• Tune thresholds and constraints with operations feedback

Where this goes next

Rail’s eight-week losing streak is a warning light, not a verdict. It’s telling shippers and 3PLs that weekly variability is now a permanent feature of the operating environment, especially across intermodal networks where small disruptions compound fast.

If you’re serious about stabilizing rail freight performance, treat AI as an operations layer: predict what breaks, decide what to protect, and rebalance the network before the week gets away from you. That’s how you keep year-to-date gains from getting erased in the messy part of the calendar.

If you want a concrete next step, start with one lane family, one terminal cluster, or one customer segment—and ask: Which decision, if we improved it by 10%, would remove the most chaos from our rail plan in Q1?