SHREVEPORT, 19 AUG 2026 — Amazon has raised its planned investment in northwest Louisiana from US$12 billion to US$18 billion, adding a third data centre campus in Shreveport. The company says the additional US$6 billion will create 210 direct jobs.
The ratio between those two numbers, sitting next to each other in the same announcement, is the most useful figure for data centre economics published this month.
The commitment
The announcement was made on 18 August. The third campus sits at Resilient Technology Park in Shreveport, giving three planned campuses across Caddo and Bossier Parishes. The original US$12 billion commitment was announced in February.
Amazon says it pays for its own energy and utility infrastructure and will invest up to US$400 million in water infrastructure upgrades at no cost to residents.
Twenty-eight million dollars per job
Divide US$6 billion by 210 direct jobs and the figure is roughly US$28.6 million of capital per position. Across the whole US$18 billion against 750 eventual full-time roles it is about US$24 million each.
This is not a criticism of Amazon or evidence of a problem. It is the nature of the asset. A modern data centre is a building full of machines that require power, cooling and a small number of very skilled people. The jobs are real, well paid and permanent. There are just not many of them per dollar.
The reason to state it plainly is that the number is almost never presented this way. Investment announcements lead with the capital figure and the job figure in the same sentence, and readers reasonably infer a relationship between them that does not hold. A US$6 billion factory would employ thousands. A US$6 billion data centre employs a couple of hundred.
That does not make the investment bad for a region. The construction jobs, tax base, utility upgrades, and anchor effect on other infrastructure are all significant. It just means that direct job creation is the wrong yardstick. Any government negotiating incentives on a per-job basis is using a metric this asset class cannot really meet.
The water number is the one to keep
The pledge of up to US$400 million for water infrastructure upgrades, paid by the company, is the most significant part of the announcement. It is a template for future projects.
Large data centres need enough water for cooling to strain a municipal system. The political failure mode is a familiar story: a new facility connects to the existing grid, demand rises, the utility gets stuck with the upgrade costs, and everyone's rates go up. The company has done nothing wrong at any step, and households pay for capacity built to serve it.
Committing to fund the upgrade directly prevents that outcome. It is a much better arrangement, and it should be a standard question for every project, not an occasional detail.
We reported in June that ASEAN's data centre boom was hitting a wall of power and water, with Johor already turning projects away. Water is the constraint that gets discussed least and binds soonest, because unlike electricity it cannot be imported from the next state.
This is the FERC question, answered voluntarily
We wrote this morning about the Federal Energy Regulatory Commission giving all six American grid operators sixty days to justify or replace their large-load interconnection rules, with the deadline falling on 17 August. Every technical question in those proceedings resolves into who pays for network reinforcement.
This announcement lands the day after that deadline and answers a version of the same question without waiting for a ruling: the company pays. Amazon states it funds its own energy and utility infrastructure, which is the position FERC may or may not end up requiring of everyone.
There is a strategic angle here. An operator that already funds its own infrastructure is well-positioned for a rule mandating it — and better positioned than competitors who assume those costs will be socialised. Voluntary compliance ahead of an expected rule is a familiar move, and it is usually cheaper than the rule.
It also raises the bar for everyone else. Once one hyperscaler funds its own water upgrades in public, the next developer arriving in a county with a smaller offer has a harder conversation.
What this means for site selection here
For Southeast Asia, the lesson is not about Louisiana. It is about what a competitive offer for a data centre now looks like.
Regional authorities have generally competed for data centre investment on tax incentives, land and speed of approval. The US$400 million water commitment suggests the frontier has moved to who absorbs infrastructure cost, and that is a term a host government can negotiate rather than a subsidy it must fund.
Johor, Batam and the Philippine economic zones are all pursuing projects whose power and water requirements exceed what local systems currently supply. The useful question in those negotiations is not about the headline investment number. It is about who pays for the upgrade, on what schedule, and what happens if the project scales back after the infrastructure is built.
Thailand has moved partway there by requiring a grid-capacity letter before investment approval, which tests whether capacity exists. The remaining question is who funds it when it does not.
What we could not establish
The power arrangement. Amazon states it pays for its own energy infrastructure, and nothing published says where the electricity comes from, at what price, under what contract, or what generation is being built to serve three campuses in two parishes. That is the largest unknown by a wide margin.
Also unestablished: the megawatt capacity of the campuses; the construction timeline and how the 750 permanent roles phase in; what tax incentives or abatements Louisiana provided, without which the net fiscal position cannot be assessed; the water volume the facilities will consume as against the US$400 million upgrade; whether the commitment is contractually binding or a stated intention; and what the 2,500 supported positions consist of and how they were counted.
What to watch
First, see if this water commitment appears in the next major data centre announcement, from Amazon or a competitor. A one-off is just a local deal. If it is repeated, it becomes the new standard — one set by industry, which is usually cheaper than waiting for regulators to impose one.
Second, keep an eye on FERC's cost allocation ruling. If the commission requires large loads to fund their own grid reinforcement, this announcement looks like smart early compliance. If it opts to socialise more of the cost, this was an expensive move for Amazon.
Finally, track the capital-per-job ratio in other announcements. The number is trivially computable from figures companies already publish, but it is almost never calculated. It is the key metric for any government trying to value one of these projects against other uses for the same incentive budget.
View author profile → · Editorial policy
