70% of logistics companies say they use AI. Most of them are applying it to infrastructure built before the internet existed.

70% of transportation and logistics companies now say they use AI. That number is up 17% from last year.

Most of them are applying AI to infrastructure that was built before the internet existed. The six largest publicly traded LTL carriers — FedEx Freight, Old Dominion, XPO, Saia, TForce, and ArcBest — collectively employ over 155,000 people, operate nearly 2,000 terminal facilities, and spend billions per year in capital expenditures maintaining their networks.

They are all investing in AI. None of them are using it to replace the terminal.

What "AI in Freight" Actually Means at the Incumbents

XPO: The Most Aggressive AI Play in Legacy LTL

XPO has deployed the most sophisticated AI stack in traditional LTL. Their proprietary platform:

• Reduces empty miles by 12%
• Forecasts demand up to 90 days out
• Drives 2.5-point productivity gains YoY
• Optimizes a $1.6 billion linehaul cost base

This is genuinely impressive work. XPO's CEO Mario Harik is a technologist and the company's AI investment is real.

But the AI is optimizing routes across 614 locations. The terminals still exist. The freight still dwells. The 5 touches still happen. XPO still employs ~38,000 people across 614 locations and carries a debt-to-equity ratio of ~2.1. Their operating ratio is 85% — roughly 1,000 basis points worse than Old Dominion — despite being the most technology-forward legacy carrier.

The AI makes the terminal model 10–15% more efficient. It does not change the terminal model. XPO's capex surged from 3.8% to 14.6% of revenue between 2018 and 2024 — the AI needs more infrastructure to run, not less.

Old Dominion: $75M/Year in IT, Still Terminal-First

Old Dominion invests $75 million per year in information technology. They are building AI-driven tools for workforce planning, route planning, billing, and equipment utilization.

Management's own characterization: these investments will "gradually reduce costs and improve service quality." The word "gradually" is doing a lot of work in that sentence.

ODFL's AI investment is real but incremental. The company's competitive advantage is not technology — it is operational discipline applied to 260 terminals and 22,522 employees. The AI is a layer on top of that operation, not a replacement for it.

TForce: "It's Unimaginable That in 2024 We Still Have Issues Billing Customers"

That is a direct quote from CEO Alain Bedard on the Q4 2024 earnings call. He also described IT costs as "way too high" and the billing system as "another excuse for not growing the business."

TForce Freight runs approximately 20,000 shipments per day through ~658 facilities with a Q4 operating ratio of 97.3%. The CEO called the quarter "a disaster."

This is not a company applying AI to freight. This is a company that cannot get basic invoicing to work three years after acquiring UPS Freight.

Saia: Playing Catch-Up on Dynamic Pricing

Saia is deploying dynamic pricing but larger competitors (XPO, FedEx Freight) have more developed yield management AI. Saia's focus has been on physical expansion — 21 new terminals in 12 months — not technology transformation. Those new terminals operate at a 95% operating ratio vs. 82.2% for mature locations.

Saia is spending $1 billion per year building terminals. The AI investment is a rounding error by comparison.

FedEx Freight: Being Separated, Not Transformed

FedEx announced in December 2024 that it would spin off FedEx Freight as a standalone public company. The rationale was not "we've built AI that makes this better" — it was that the LTL asset was "not fully appreciated" within FedEx and traded at 13x forward estimates vs. 20x for LTL peers.

FedEx Freight's path is financial engineering, not AI engineering. The 355 terminals and 39,000 employees will become a standalone company — still running the same infrastructure, just with a different stock ticker.

The Hype Layer vs. The Infrastructure Layer

Here is what "AI in freight" usually means when a logistics company says it on LinkedIn:

"AI-powered route optimization" — a slightly better algorithm for sequencing delivery stops. This is math that has existed since the 1950s. Calling it AI is like calling a spreadsheet "machine learning."

"AI chatbot for tracking" — a chatbot that reads your tracking number and returns a status. This is an API call with a conversational wrapper.

"AI-driven demand forecasting" — a regression model that predicts next quarter's volume will be roughly similar to this quarter's, adjusted for seasonality.

"AI freight matching" — a load board that sorts available trucks by proximity and price. This is a database query, not intelligence.

None of these are bad. But none of them address the structural problem. The physical infrastructure — terminals, trucks, carriers, facilities — is broken, and no amount of software optimization fixes broken infrastructure.

Where AI Creates Structural Change

AI in freight only matters when it is connected to the physical operation — not layered on top of it as a dashboard. Here is the difference:

AI as a Layer (What incumbents do)

Old Dominion builds AI tools for workforce planning → still employs 22,522 people at 260 terminals.

XPO deploys AI to reduce empty miles (targeting 12% reduction per investor presentations) → still runs 614 locations with 38,000 employees and ~$3.3B in long-term debt.

Saia builds dynamic pricing → still opens terminals at 95% operating ratio and burns through $296M in cash.

The AI improves the metrics by single-digit percentages. The underlying cost structure — labor, terminals, equipment — remains.

AI as Infrastructure (What a new model does)

When AI IS the operating system rather than a feature on top of one, it looks fundamentally different:

1. Pricing at density — 11 million quotes processed

Traditional LTL pricing: a sales rep quotes a rate based on a tariff and a discount schedule. The rate has little relationship to real-time supply and demand on a specific lane.

An AI pricing engine processes quote volume in real time. One network has processed over 11 million freight quotes — growing from 387,000/month to 1.27 million/month in 13 months. Every quote teaches the system about lane demand, price sensitivity, and carrier availability. Pricing gets cheaper as lanes densify — not because a sales rep offered a discount, but because the underlying cost structure improved with volume.

Compare this to Saia's CEO defending yield-over-volume pricing in a freight recession, or TForce losing profitable SMB customers and "replacing them with corporate accounts with sometimes negative margins."

2. Automated quality monitoring — every shipment, every carrier

In traditional freight, you find out something went wrong when the customer calls. TForce's CEO described billing failures — they cannot even invoice correctly, let alone monitor quality in real time.

An AI monitoring system connected to the physical operation — driver apps, ELD integrations, cross-dock scan events, GPS data — flags exceptions as they happen:

• Late pickups flagged when the driver has not departed within the window
• Missed scans flagged when freight enters a facility without a scan event
• Route deviations flagged when a truck deviates from the planned path
• Dwell anomalies flagged when freight sits longer than expected
• Temperature deviations flagged when a reefer unit falls outside target range

This requires physical instrumentation: a driver app on every carrier, ELD data from every line-haul truck, scan events at every dock door. AI without physical data inputs is just a dashboard — which is what most incumbents are building.

3. Carrier vetting at scale — 22,246 carriers, continuously evaluated

The U.S. freight market has hundreds of thousands of carriers. Most LTL carriers employ their own drivers and manage their own fleet. Old Dominion runs 11,284 tractors. XPO added 2,300 tractors and 4,400 trailers in 2024 alone. These are depreciating assets that need constant replacement.

An alternative: no owned fleet. Instead, a managed network of 22,246 active carriers continuously evaluated on authority, insurance, safety scores, equipment, on-time performance, and damage rates. Carriers with lapsed insurance are automatically suspended. Carriers with declining performance are removed from lane assignments.

Old Dominion spends $325 million per year on tractors and trailers. An AI-managed carrier network spends zero — and has 2x the carrier count of ODFL's entire driver workforce.

4. Cross-dock orchestration — 85.5% same-day throughput

A cross-dock facility is only as fast as its ability to match inbound freight to outbound capacity. In a traditional terminal, dock supervisors manage this manually. In an AI-orchestrated cross-dock, inbound scan data drives outbound dispatch timing in real time.

The result: the best facility processes 85.5% of freight same-day with an average dwell of 0.67 days. No traditional LTL terminal discloses dwell time — because no traditional terminal can match this.

The Cost Structure Comparison

This is the table that matters:

Cost Category	Legacy LTL (avg of top 6)	AI-Infrastructure Model
Terminal/facility capex	$300M–$1B/year	Fraction — cross-dock leases, no storage
Fleet capex	$300M+/year (tractors, trailers)	$0 — carrier partners own equipment
Employee count	14,000–39,000	Zero drivers, minimal facility staff
Annual labor cost inflation	5–8.7%	Carrier market rates (competitive)
Insurance cost trend	Rising 10–16.6% (Saia, ArcBest)	Carrier-borne
Debt	$0 (ODFL) to ~$3.3B (XPO long-term)	—
Operating ratio	73.4% (ODFL best) to 97.3% (TForce worst)	Improves with density
Volume trend	Declining 6–14% across carriers	Growing 7x in 21 months

Every major LTL carrier's cost structure is anchored to physical assets: terminals, trucks, and people. When volume declines — as it has for 2.5 years in the freight recession — those fixed costs crush operating ratios. ODFL's OR went from 71.9% to 75.9% in two quarters. ABF hit 91.2%. TForce hit 97.3%.

An AI-infrastructure model has a fundamentally different relationship to volume: costs are largely variable (carrier payments, facility throughput), and the system gets cheaper as density increases. Volume declines do not create the same deleveraging trap because there are no terminals to underutilize and no fleet to park.

What to Ask

When a freight company tells you they use AI, ask three questions:

1. How many terminals do you own? If the answer is 200+, the AI is optimizing a legacy network, not replacing it.
2. What is your operating ratio, and does it improve or deteriorate when volumes drop? If it deteriorates (as it did for every public LTL carrier in 2024), the cost structure is fixed, and AI is not changing that.
3. Does the system get cheaper as volume increases? If pricing, routing, and carrier selection do not improve with more shipments, the AI is decorative.

The $17.96 billion AI-in-logistics market is real. But the vast majority of that spend is going into optimization of terminal-based infrastructure that has not fundamentally changed since the 1990s. XPO will reduce empty miles by 12%. Old Dominion will "gradually" improve costs. TForce will try to fix its billing system.

The structural opportunity is not in making terminals slightly more efficient. It is in replacing them entirely — and building AI into the foundation of a network designed for it.