Gordie Howe bridge set to open July 27 after delay; 400 trucks per hour projected

The Analysis

The Gordie Howe International Bridge opens on July 27, weeks later than planned, with roughly 400 commercial vehicles projected per hour between Detroit and Windsor. It is the number everyone will quote and the least useful one in the release, because it describes the road and says nothing about the buildings at either end of it.

A border crossing is two servers in series, sometimes three. The span decides how many trucks are in motion. The booth decides how many may continue. Add capacity to the first and the queue does not vanish, it relocates to wherever waiting costs more, since a truck idling on a span is merely slow while a truck idling at a booth is burning detention and losing its slot.

Where the constraint sits after the ribbon, how long the corridor takes to settle, and which of your lanes feels it first are all measurable. The 400 is not.

The release establishes a new crossing, a date that already slipped once, a nameplate of about 400 commercial vehicles an hour, and relief for one of the busiest freight corridors on the continent. It does not establish the booth count by shift, the lane split by direction, the empty-versus-loaded split, how much traffic is expected to leave the Ambassador Bridge, or whether 400 is a design peak or a rate meant to hold for a full day. Those parameters sit between the nameplate and your transit time.

Border arrivals are not smooth. There is a morning wave, a midday plateau and an evening wave, and the shape differs by direction because loaded southbound freight and empty northbound returns do not run on the same clock. Service varies too, a couple of minutes for a pre-cleared load and twenty for a referral. Push a variable arrival process through a variable service process and the wait is governed by how close the mean sits to capacity, not by the mean itself. In the simplest single-server model the expected wait is roughly four times service time at 80 percent utilization, nine at 90, nineteen at 95. That is a cliff, and published figures usually describe the flat part.

Which is why the span is neither the transformation nor the non-event that the two loud readings suggest. At low-80s peak utilization, extra approach capacity barely moves the peak hour and helps only the shoulders. At mid-90s, it buys real time by spreading arrivals, but the ceiling is booth-minutes. You cannot tell the regime from the release. Four weeks of your own timestamps can.

The split by trade profile is stark. Carriers with dedicated lanes and trusted-trader status gain first, because short service times and low variance turn approach capacity almost directly into schedule reliability, which is what they sell. Smaller fleets gain secondhand as larger ones vacate the old crossing, and they absorb the first month of oscillation. Time-critical freight gains most in absolute terms, since an hour costs a plant feeding an assembly line far more than it costs the average load. Low-value bulk barely notices, and that is not a failure, because a river crossing is a small slice of landed cost and those routes are chosen on rate. Direction slices it again, loaded southbound and empty northbound being different products at the booth. Inside a company the planner gains options and wants them all at once, customs inherits new lanes and a second transit assumption to maintain, and finance inherits detention accruals that read as variance for two quarters before they read as a trend.

You feel this in stages. Day one is ceremonial. The first two weeks are a learning curve on both banks, and transit times run worse than the eventual steady state even when the facility works as designed. Between week four and week eight the corridor finds a new routing equilibrium and carriers start committing capacity. The first reliable signal is not time across the span but appointment availability at the plaza and the gap between your 50th and 90th percentile crossings. A mean that improves while the spread widens is not stability, and a buffer sized on the mean will fail.

Here is the missing piece. The 400 is a supply-side figure taken from road geometry. What governs your day is booth-minutes per truck, and nobody published it. Take the nameplate at face value and invert it. At 400 trucks an hour, if primary inspection averages two minutes, you need 400 times two divided by 60, or 13.3 booths staffed continuously, with no breaks, no shift change, no referrals. That two-minute average is my assumption, not a published figure, and it is likely optimistic at peak, so the real requirement is higher. Nobody has said how many booths open at seven on a Tuesday in October. The second omission is subtler: your transit distribution just became non-stationary. Buffers are sized against variance, and a corridor in transition can lower its mean while raising its variance, which is the state that breaks promises quietly.

A small piece of arithmetic that is yours. Suppose twenty loads a week save forty-five minutes each once the crossing settles. That is fifteen truck-hours a week, not a rounding error for a driver pool. I will not convert it into money, because the loaded cost of a truck-hour depends on your contract, your pay model and how much of the wait is already paid as detention, and any hourly figure I invented would be doing the job your invoice should do. Apply your own cost to the fifteen hours, then weigh it against the detention you eat while the corridor oscillates.

Run the counterfactual both ways. Understaffed booths leave headline transit time roughly flat while detention and missed windows rise, and the bridge gets judged a disappointment by people who never measured the second server. Rebaselining on week-one data bakes in the worst of the learning curve and has you under-promising for a year against numbers that were never real. Hold the old distribution, collect a new one, and swap only when enough weeks separate trend from noise.

Three traps deserve naming. Treating the date as certain, when it has already moved once. Comparing the crossings without matching hour, weekday and direction, because the peak shapes differ and a naive average flatters whichever one you sampled on a quiet Tuesday. And forgetting that your broker, your appointment system and your drayage provider are servers too, each with a queue. A faster bridge into a saturated drayage market does not deliver faster. It delivers trucks waiting at a different address.

Model two servers, instrument both, and let four weeks of data choose the route. The span will be fine. Whether your routing numbers still are, is the open question.

↑ Back to the key points

— By Ravi Chandran