Quick Note on Subway Operating Costs
A London Underground benchmarking report using CoMET data from 2013 compared operating cost breakdowns and revenues between the Underground and its international peers. CoMET data is in principle anonymized, but it’s not hard to find which city is which, and in particular, whereas London, Paris, and Berlin all spent around $6/car-km on operations in 2013 costs, New York is the city labeled “Am” that spent $10/car-km.
More recently, I followed up on these costs by looking at 2020s data, finding little for Berlin and even less for Paris, but finding exact costs per car-km for New York and per train-km for London. London has seven cars per train, from which we can impute, in 2024 PPP prices, $6.2/car-km in London; New York’s exact costs are $11.58/car-km. BVG’s costs are bundled across modes, but the total costs for 2024 reported in the Lagebericht und Jahresabschluss 2025 are 1.6B€; if U-Bahn costs per car-km and bus costs per bus-km were as in New York and tram costs per tram-km were as in Boston, the two rail modes compared on a per train length basis (thus, a Berlin U-Bahn train of 100 meters is deemed to be six New York City Subway cars), the total would be exactly twice on a PPP basis. If the same comparison were made with an adjustment for bus speed (17.9 km/h here, 11.3 km/h in New York), make it 1.6B€ vs. 2.6B€. Either way, it’s very likely Berlin’s U-Bahn operating costs are in the $6-7/car-km range in 2024 prices.
All of this is remarkable, because prices between 2013 and 2024 rose, by a factor of 1.34 in the US. And yet, despite this inflation, London more or less kept its operating costs unchanged, New York had an increase of slightly less than half the inflation rate, and Berlin likely had a small increase like New York or even smaller.
Moreover, none of the three systems engaged in massive automation over this period, not even the incremental automation of Paris. Furthermore, New York’s subway costs in the short and medium runs tend to rise when ridership decreases and fall when it increases, as fixed costs are spread across more service; the number of employees per unit of service provided rose when ridership fell after WW2 and fell when it recovered. However, the period 2013-24 was not one of major service increases in any of the cities: all three opened new lines, but only short ones, and none of the three embarked on a scheme to massively increase service – London had some increases but New York if anything provides less service now than it did in the early 2010s. Thus, no short-term shock can explain the over-the-decade fall in real operating costs in all three cities.
This contrasts with buses, which are dominated by variable labor costs. In the United States, the cost of running a bus is the wage of the operator plus various overheads, of which the largest is the wage of the maintenance crew. In Europe, bus driver wages are lower, but buses are also cheaper to procure and more fuel-efficient, and overall the system is dominated by wages rising faster than inflation and by variable and not fixed costs.
Pedestrian Observations 
I get different results for London. From the link you shared for LU operating costs, I see £28.52m for FY 2023/24. During that year, LU operated 78.3m train-km https://tfl.gov.uk/cdn/static/cms/documents/kilometres-operated.csv, or 548.3 car-km (using actual numbers of cars per line, but it averages to almost exactly 7 anyway). That would give £0.052 per km, or $0.078 using the World Bank PPP of 0.67 for 2024. That’s a nominal increase of 29%, though a slight reduction in real terms as both UK and US inflation were around 36% over this period. Interestingly, London ran 6% fewer train-km over this period
You are mistaken.
The overall cost of LU operations is probably around 6000 million Pounds. In the Excel table, 28.52 is the British Pound cost per km of London Underground trains. You can then divide it by 7 to arrive at the cost per LU car-km.
One can compare LU car/km costs to the ones of Bordeaux 44 meter trams.
The Bordeaux cost was 8.26 Euros/km in 2010 which, given inflation, is equivalent to 10.55 Euros in 2025. While LU could keep running costs stable, Bordeaux trams costs increased slightly faster than consumer prices, up to 11.08 E/km. 2025 was specially difficult with 9 months of works on a core section of the network. It will be interesting to see whether the December 2025 restructuring of the network was successful. Many think that the trams hit a ceiling years ago and that, like Rennes and Toulouse, Bordeaux should have invested in an automated subway.
As per page 74 of the annual report it’s £2.7bn plus whatever share of the central costs is reasonable and not included.
Thanks. Page 73, TfL states that total recurring operating cost savings delivered since 2016 equal 1.5 billion. Although they don’t specify how much of these savings occurred in the Underground, this is consistent with Alon’s observation.
Enough to pay for 2km of HS2 or 25km of Bordeaux-Toulouse!
Sorry 40km of HS2.
I suppose you meant 40 km of Bordeaux-Toulouse. Anyway, we are not talking of leftover cash accumulated in a bank account. The 1.5 billion aren’t there but it is nevertheless a significant achievement by TfL.
While TfL prefers talking about Elizabeth Line success than about cost per car-km, others often talk about savings. Over the years, multiple HS2 cost-cutting opportunities were found by pruning links or branches, while the final bill and the painful construction period kept increasing. The latest saving of up to 2.5 billion by reducing top speed from 360 to 320 km/h is the most egregious since the much larger damage caused by the absurd request for record speed is already done and no-one is taking responsibility.
Not to be undone, CAHSR is saving billions by serving LA and San Francisco with connecting buses, going single track, and moving Central Valley stations out of town. First, they spent billions they did not have by refusing to build HSR built in cheap and abundant agricultural land and insisting that Palmdale and Central Valley towns had to be served, often by existing corridors. Two, these towns which often grew around these corridors will now be served by parkway stations and, three, they will be cut in two by the drastic FRA requirements for physical separation between freight and high speed rail.
CAHSR’s trifecta is impressive but Brightline West is a contender for the gold medal in savings. Its single track follows the median of a curvy and steep highway. Its main stations are further away from downtowns than the airports they should compete with, and I wonder what the FRA is going to say about HS trains barreling down a 6% grade without clear track ahead of them.
Guess what, the HS2 “Reset” is done by the team who completed Crossrail, which was itself over budget, delayed, and mismanaged.
https://learninglegacy.crossrail.co.uk/wp-content/uploads/2023/03/Project-Leadership-Getting-Crossrail-back-on-track.pdf
The new CEO’s great takeaways from the Crossrail experience is “Systems thinking is essential” “Deadlines can be deadly” and “Complexity kills”.
Perhaps that’s why they ditched the 360km/h certification to do 320km/h because it’s less complex 🤣.
@dralaindumas
Criticizing CAHSR for taking the eastern valley route (through Fresno, Bakersfield, etc./Hwy 99) vs the western valley (through farmland/I-5) is valid, there is a successful HSR model (France) that takes the ‘bypass intermediate cities for direct service approach.’ However, “served by parkway stations” and “cut in two by drastic FRA requirements” are mutually exclusive; if the lines are not stopping in downtown then they are not cutting the city in two. Also, grade separation requirements have to do with HSR speed, not separation of freight and HSR; there would be the same grade separations if the HSR line was totally independent of a freight track. Also, the most important station, Fresno, is right in the heart of downtown at Mariposa St, as is Merced. Visalia/Hanford/Tulare/Lemoore is certainly parkway/beetfield station, along with Madera, but the first four are third tier cities in a cluster, not a single urban area, and Madera is very small so the out of center stations are appropriate. Bakersfield is the big miss in not serving the downtown, driven in part by the dogleg to Palmdale, which is the real problem with CAHSR. Also, if the route had gone through the cheap farmland to the west, then the stations would be served by far worse than parkway stations, they effectively wouldn’t be served at all with stations up to 75km away by road.
For the record, I think serving the Central Valley cities is the right choice versus the I-5 route. The CV combines for ~4M people, with Fresno/Bakersfield each having about a million. For comparison Shizukoa Prefecture (3.5M) has six stops on the Tokaido Shinkansen, and Gifu (2M) gets one, while cities like Hamamatsu with several hundred thousand people get Hikari service not just Kodama. I would have placed almost all of the stations in city centers (except for the Vislalia/Hanford compromise, and accepted the slight increase in time from the non-stops having to do less than full speed for a few km right through town. Losing that time to serve millions is a far better deal than the time lost to serve a few hundred thousand (at most) in Palmdale, which I would not have done. If it gets to be a point where that slight time loss is a major inhibition to growth, then your ridership is probably so high you can think of doing the I-5 route as an express-only bypass route, even cheaper if it has no stations.
As for Brightline, the highway median route is actually very flat and mostly straight across the Mojave Desert. The only curvy and steep part is the climb up Cajon Pass, but with LA at 300 ft elevation and Las Vegas at 2000 ft (more relevantly, San Bernardino at 1000 ft and Victorville at 2700 ft) there is going to be a short stretch of steep and curvy track no matter how much HSR would prefer there is not. The Vegas station is crazy, especially when the existing rail line is 700-1000m from the Strip from Flamingo Rd all the way to downtown/Front Street. As for Rancho Cucamonga being the “LA terminus”, however, it is right at the Metrolink station on the San Bernardino line running to LA Union Station, Metrolink’s busiest, on a mostly straight right of way. If Brightline West got HSR trains to Rancho Cucamonga, I can’t imagine that even California would not electrify those remaining km on existing tracks to get the trains to LA.
@Sia, same politician in charge of both too, Heidi Alexander.
@Onux and Sia.
The only downtown station left is Fresno. CAHSR is now planning to stop south-east of Merced and near Bakersfield airport in order to save a few billions. “If the lines are not stopping in downtown then they are not cutting the city in two” is also true in Corcoran where CAHSR appears to be aiming for the town and then bends around it, but not in Shafter. Shafter, a city of 23000, is too close to Bakersfield to warrant a station, but big enough to create trouble.
Trying to follow old Central Valley right of ways and trying to beat speed records in the Chilterns was asking for trouble but what about Crossrail ? The difficult integration of TPWS, ETCS and CBTC signaling systems caused costly delays. Opening Crossrail with track side signals would have been anachronistic. Maybe ETCS, which allowed 24 trains per hour on Thameslink, would have been good enough.
Good enuff?!! Some railfans insist it has to be 30 to 35tph or it’s amateur hour …
I never understood why they didn’t upgrade the lines that were sharing to the same (newest?) system, which had to be cheaper and certainly simpler and less troublesome than a custom made one to cope with three different systems. It is remarkable it works at all, and suppose it was the cause of those problems the system experienced in its first year.
At the foot of the Grapevine (California Interstate 5, on the “path not taken” by CAHSR, in the Central Valley) the new Hard Rock Casino Tejon has just opened — 2000 slot machines, 50 table games, and lots of parking.
Indian reserve casinos and parkway stations have in common large catchments areas and a potential for addiction. I am familiar with the popular Avignon and Aix-TGV stations. Only a single digit share of their users heads downtown with the buses and TER trains respectively available in Aix and Avignon. The others use private cars and prefer the less congested access and the large parking and drop-off areas available in these green field stations.
CAHSR faces independent obstacles to fast services. The Grapevine and/or Altamont alternatives do not preclude HSR service to the Central Valley towns starting with Bakersfield located at 7 km from I-95. There are 385 km on an almost straight line between the feet of the Tejon and Altamont passes. Bending the HS line in order to pass next to Hanford, Fresno, Merced, Turlock (73000 inhabitants) and Modesto (220 000) would only add 15 km i.e. 3 minutes to a non-stop trip. However, telling Californians that CAHSR is going to restart from scratch is unthinkable. What has been built between the north of Shafter, Fresno and Merced will have to be used. Lengthwise, the penalty is insignificant. Time wise, it depends on the speed allowed in inhabited areas like the 20 km Fresno one and next to active freight lines. CAHSR is not telling us and I won’t try to guess.
They share with freight west of London. But east of London yes.
Really? I’m shocked by that. Do you mean the line is used by freight trains at night? It’s just a bit hard to see a line set up for 30-35tph sharing with freight during normal pax operations. Do any RER lines share with freight?
@Michael
The Crossrail services have a lot of short turns and some branches to the West. The effective frequency up to Reading is much lower than the set up 30-35tph. Crossrail essentially took over the local lines of the WCML, but it still is a mixed traffic railway in general.
@dralaindumas and Onux
The downtown Bakersfield station is allegedly still in the cards for the full “Phase 1”, when the Bakersfield-Palmdale section opens. The peripheral Bakersfield North as a temporary station will be opened with the initial operating system.
The downtown Fresno and Bakersfield station locations themselves aren’t a problem in my opinion. In a country with functional government there would be expertise to coordinate all the local stakeholders in a productive fashion, and the existing alignments are generally good enough. (Bakersfield is a bit harder than Fresno, but this is easy mode compared to Japan). Unfortunately the inability for government to coordinate makes it so that peripheral stations might make sense, but not for any actually good physical/alignment/ridership reasons.
Given the IOS will only operate as an effectively a short intercity line, the downtown stations are more important as access time is a larger proportion of the trip time. I’d argue for full system buildout more peripheral stations could be accepted, but not preferred.
Under the current legislative paradigm I can see why SNCF wanted the I-5 route because they knew coordinating with utilities and different stakeholders would be a process mired in dysfunction.
Sia, CAHSR follows the German, Belgian and Dutch model of high speed tracks admixed with a large amount of conventional ones. On the 832 km LA-SF route chosen by CAHSR, the 163 km/h average speed of the fastest Berlin-Munich ICEs results in a 5h06 trip duration, well above the 2h40 mandated by the 2008 proposition 1A.
According to leeks reported on Alon’s July 2012 “CAHSR-SNCF Bombshell” blog, banks were ready to finance a shorter $38 billion route along I-5 while SNCF wanted to fulfill the 2008 Proposition 1A requirements of serving Central Valley towns on a route presumably similar to the 645 km one I described a couple of days ago. At the average speeds achieved on French LGVs, these routes would allow a 2h40 to 3h L.A.-SF trip.
I believe “downtown” stations located on an undistinguished stretch of SR 204 in Bakersfield and along the SP/UP corridor in Fresno would have been advantageously replaced by greenfield stations offering much faster trips to L.A. and the Bay Area. I admit I have never been to Bakersfield or Fresno. Am I missing something? Can someone tell me why the stations suggested by CAHSR are preferable ? Located at kms from the Amtrak ones they don’t even fulfill proposition 1A requirement of seamless connectivity to commuter rail.
So short of money that it will only lay a single track on the segment under construction, CAHSR added some 200 km of HSR double tracks and more than two hours on the L.A.-SF trip. It doesn’t make sense. CAHSR and California politicians don’t provide an estimated timetable along their chosen route through Palmdale and Gilroy and hide SNCF’s recommendations. These recommendations appear to be California’s Epstein files. Governmental dysfunction is a weak excuse for the lack of transparency.
They didn’t see your omniscience and therefore it must be a conspiracy to hide your omniscience. If they didn’t see it how do they know to hide it?
This was because I made a inaccurate comment on Bluesky thread with Alon.
It suggests clear that the reason LUL has so many strikes precisely because TfL drives hard bargains. (Not there isn’t room to improve particularly on station staff on the sub-surface lines).
Operating costs: New Jersey signed a contract with FIFA to provide transportation “at cost”. New Jersey proclaimed it would not have its citizens subsidize a rich sports league, so the price to ride the train from Penn Station to Meadowlands for the World Cup became $150 round trip (a one-way distance of about 8 miles/12 km). Some corporate sponsors have since stepped-in to get the cost to around $105.
This is total madness, absolute malaise, 150 $ can not be real cost.
For comparison, it costs 25 euro to go 450 km from Bratislava to Praha at peak hour, on Regiojet. This is market price, there is NO subsidy. So for that price you could travel about 3000 kilometers on train.
Boston’s MBTA has just announced that the $80 train fare (56 miles round-trip distance) from downtown to Gillette Stadium for the World Cup will not cover the cost to provide the service.
That’s presumably with overtime and various FIFA requirements. The average operating costs of the MBTA at that roundtrip distance are $1,265.6 per car, and they’re expecting a lot more than 16 passengers per train.