Supply Chain Action Points
Read this first — the conclusion, and the moves to make:
- Within 30 days, split your European item list into ten velocity deciles and calculate handling cost per unit for each; flag any decile where that cost exceeds 20% of the item's landed value.
- Before your Q4 2026 tender, obtain two figures in writing from every fulfillment partner: dock-to-stock days across the last eight peak weeks, and appointment lead time in days; put both into the contract with a service credit attached.
- Commit no more than 60% of peak 2027 volume on contract and hold the rest for spot, with a decision date eight weeks before cut-off, so you are not paying for space that arrives after your season.
- Set a review for the second quarter of 2027, re-run the decile table against observed performance, and move the slowest decile to a cheaper node if dock-to-stock has not improved by 20%.
- Track one leading indicator every month, the share of your items a partner can receive without manual handling instructions; a flat line across two quarters is evidence the threshold has not moved on your lane.
Amazon announced a €10 billion ($10.8 billion) investment across its European fulfillment and delivery network, creating 25,000 jobs and rolling out next-generation robotics including an upgraded AI-powered Proteus warehouse robot capable of responding to conversational prompts. Announced at the Delivering the Future event in London on June 4, the investment represents one of Amazon’s largest-ever European supply chain commitments. The new Proteus robot enables natural language interaction between workers and robotic systems, signaling Amazon’s strategic bet that robotics-driven efficiency will define next-gen e-commerce fulfillment economics.
The Analysis
Every fulfillment argument eventually collapses into one number: the hours of labour available inside a shift. Not square metres, not shelving, not even the count of robots. A building can hold more inventory than it can move, and what decides how much it moves is how many pairs of hands are inside it during the window that matters.
Amazon said it will put 10 billion euros, about 10.8 billion dollars, into its European fulfillment and delivery network, add 25,000 jobs, and roll out a next generation of warehouse robotics including an upgraded Proteus that responds to conversational prompts. The announcement came at its Delivering the Future event in London on 4 June.
The figure I keep circling back to is not in the headline. Divide the ten billion by the twenty five thousand jobs and you get roughly 400,000 euros of capital per job. That is not a labour story. That is a capacity story wearing a labour story as a coat, and it tells you what kind of network you will be dealing with in 2027.
Start with where a pick operation actually binds. A picker's output is not linear in headcount, because a shift contains a fixed number of productive minutes and a new hire spends the first weeks climbing a learning curve. The real ceiling comes from variance: which item, which slot, which exception. A machine that only repeats a memorised path can justify itself only on the fastest movers, where the identical motion runs thousands of times. That is why warehouse automation has always clustered at the head of the velocity curve and left the tail to people.
Conversational instruction changes the cost of the first cycle, not the thousandth. If a worker can describe a task in ordinary language instead of waiting for someone to program a path, the setup time for each new item-and-task pair falls. Automation then becomes economic further down the curve, on goods that move a few times a week rather than a few hundred times a day. That is the line worth marking. The announcement is not that picking gets faster. It is that the threshold at which a slow-moving item can be handled by a machine has just moved down one notch.
The counterweight sits in the same release. Twenty five thousand jobs is not a by-product of the investment; in the near term it is the constraint on it. Hiring at that scale across European labour markets is measured in quarters, and a new building does not open at its rated throughput. It opens, then it ramps. The ramp curve is the least quoted number in any fulfillment announcement and the one your delivery dates actually depend on. Robotics lowers steady-state labour per unit. It does not lower the headcount needed on day one of a new site, and it does not shorten the training period for the exceptions that machines still hand back.
Who feels this first depends on whether you own the building or buy the service. Run your own European distribution centre and the benchmark against which you are judged has just shifted, because a competitor's cost per unit in a new building will be set on a different labour curve than yours. Buy fulfillment as a service and the first thing that reaches you is not speed. It is a repricing conversation with a fresh argument behind it.
Size splits the other way. High volume over a narrow range, and you are largely untouched, because you were already automated or already outsourced. Long tail, many items, thin volumes, and this is where the threshold shift lands, because your handling cost per unit is dominated by changeover rather than by the pick itself.
Trade direction decides which dock hurts. Importing into Europe from Asia, your constraint is receiving and putaway, and that is the part a new building adds slowly because it is staffed by people rather than robots. Selling inside Europe against a next-day promise, your constraint is the outbound cut-off, and that is the part robotics genuinely moves. Selling into European buyers from outside, you will meet it as a change in what your customer asks for, because the delivery promise they expect gets set by somebody else's network.
Nothing you can buy changes before the fourth quarter of 2026. Construction and hiring lead, robots follow, service levels follow last. My working assumption, and it is an assumption rather than a figure from the announcement, is four to eight quarters between a single-site commitment and a change you can write into a routing guide. What arrives sooner is the contract conversation, because the operators competing with this network will reprice and prune their peak commitments for 2027 while the news is still a press release.
Peak is the other timing signal. A network carrying 400,000 euros of capital per job has to run close to full to earn a return, and an operator in that position has every reason to sell the last cubic metre and the last appointment slot. Flexible peak capacity is the first thing to disappear in that model, not the last. If your peak plan assumes you can buy overflow space in October, that assumption gets more expensive every year.
Run the counterfactual in both directions. Treat this as background noise and you keep a contract priced on the old labour curve while treating every item as if it costs the same to handle, so twelve months later you are paying head-of-curve prices for tail-of-curve goods. Overcorrect and pre-book peak space eighteen months out, and you pay for capacity that lands after your season, in a building that may not be the one that received the investment.
There is a cheap version of the arithmetic, and it needs none of the ten billion. Sort your European item list by annual units and cut it into ten deciles. For each decile, divide annual handling cost by annual units. In most catalogues the bottom three deciles carry a handling cost per unit several times that of the top decile, and almost nobody prices storage that way. That table alone tells you which goods are candidates to move to a slower, cheaper node the moment a partner can receive them without special instructions. You need an afternoon and your own invoice data.
The misreading I see most often is treating an announcement as capacity. It is not capacity until a building's ramp curve has flattened, and that is a date you can ask for. The second misreading is measuring the wrong thing: teams track on-time shipment, which is an output, and never track dock-to-stock days, which is the input they can actually move with an appointment and a clean advance ship notice.
↑ Back to the key points