Amazon is turning ultra-fast delivery from a margin drain into a monetised infrastructure play by redesigning its network, inventory logic, and capacity stack.
In Brief
- Amazon has rebuilt its North American network around regionalised, same-day and 30‑minute delivery, while claiming a lower cost-to-serve.
- Everyday essentials and perishables now anchor demand, using ultra-fast services to lift frequency, basket size, and utilisation of fixed assets.
- Custom chips, power, AI, and robotics sit behind this model, treating speed as an infrastructure business rather than a promotional feature.
Where Amazon’s Cost Logic Breaks With The Past
The structural break in Amazon’s retail supply chain is not the fact of same‑day delivery; it is the claim that faster delivery is now cheaper to serve. In 2025, the company delivered more than 8 billion items to U.S. Prime members on a same‑day or next‑day basis, over 30 percent more than the prior year, while also reducing cost-to-serve for the third consecutive year. Groceries and everyday essentials accounted for roughly half of those rapid shipments.
Amazon has moved from a national fulfilment model, where inventory was pushed from a limited set of mega‑FCs, to a regionalised topology. The U.S. has been sliced into ten regions (up from eight), with inventory placed closer to demand and both inbound and outbound flows regionalised. The company reports that this redesign has improved local inventory placement and reduced distance travelled per package, supporting both higher service levels and lower unit cost.
The strategic shift is that same‑day and even 30‑minute promises are no longer opportunistic features layered onto an existing network. They are now the design point for the network itself, and the economics are being built upward from that premise.
How Ultra-fast Delivery Is Actually Executed
Operationally, ultra-fast delivery at Amazon rests on four intertwined elements: network topology, inventory policy, automation, and demand steering.
At network level, the move to ten regions is not simply a relabelling exercise. It implies regionalised inbound logistics, with vendors and internal sourcing flows directed into closer, smaller catchments. This reduces lead times and safety‑stock requirements for fast movers, and makes it viable to offer same‑day on thousands of SKUs without excessive overstock. Amazon has also indicated that inbound has been regionalised alongside outbound, tightening the loop between supplier, FC, and last mile.
Inventory logic is being tuned to the new demand mix. Everyday essentials grew nearly twice as fast as other categories in the U.S. in 2025 and now account for around one in three units. Perishables have been layered onto this base, with same‑day grocery and fresh delivery now available in more than 2,300 U.S. cities and towns using the existing network footprint. Where perishables are offered, nine of the ten top‑selling items are fresh products, and shoppers using the service buy three times more items per same‑day order and spend meaningfully more per month.
To make this work, Amazon must set stricter service thresholds and replenishment cadences for high‑frequency SKUs. In practice, that requires finer-grained demand forecasting, tighter slotting of fast movers near outbound docks, and routing logic that can guarantee perishables within short cut-off windows without unplanned expedites.
Automation and robotics are used to offset the labour and handling intensity that would otherwise accompany this model. Amazon reports having over one million robots in its fulfilment network, taking on tasks such as tote transport, sortation, and goods‑to‑person moves. This reduces touches per package and stabilises throughput under peak loads like Christmas Eve, when customers in roughly 4,000 U.S. cities were able to order up to midday and still receive same‑day delivery.
On the digital side, the company says it has more than 1,000 AI applications in production or development, including forecasting engines for its fulfilment network. These applications are used to predict demand at ever narrower time and location bands, allocate inventory across regions, and select the optimal node for each order based on availability, cost, and promised speed.
Demand steering closes the loop. Add to Delivery, launched in 2025, allows Prime members to attach items to an upcoming shipment. Within six months, it accounted for about 10 percent of Prime volume fulfilled through Amazon’s own network each week. This feature effectively consolidates demand into fewer, more efficient consignments, improving carton fill rates and reducing last‑mile trips. AI‑driven discovery tools like Lens and the Rufus shopping agent are also steering customers toward items that are in‑region and available within desired timeframes, tightening the match between virtual shelf and physical inventory.
In operational terms, this kind of model requires a very different planning cadence. SKU policies must distinguish between ultra-fast eligible inventory and slower‑promise stock, with separate parameters for safety stock, reorder points, and substitution rules. Allocation logic has to be dynamic, often re‑routing items between regions or reassigning stock from slower to faster service tiers as demand patterns shift. Master data on item dimensions, handling constraints, and freshness windows becomes a core input into routing and cartonisation decisions.
Turning Grocery and Quick Commerce Into Utilisation Engines
Ultra-fast delivery economics depend on density. Amazon is using grocery and everyday essentials to generate that density, particularly in urban and suburban catchments. The company reports more than 150 million Americans now use it as a grocery destination, with over 150 billion dollars in gross grocery sales. Customers who start buying perishables shop with Amazon roughly twice as often as before.
Same‑day services are structured to capture these repeats. Nearly 100 million U.S. customers used same‑day in 2025, making it the fastest‑growing delivery option. In markets such as India, where the Amazon Now quick commerce offer is established, customers who try 30‑minute delivery shop three times as often as they did previously.
At network level, this is implemented through a mix of central FCs, localised sorting and delivery stations, and, increasingly, stores. Perishables are being delivered same‑day in thousands of cities off the existing network, but Amazon also plans to open more than 100 new Whole Foods Market stores over the next few years. These stores are both retail assets and micro‑fulfilment nodes, particularly for fresh and chilled items that are difficult to ship over long distances.
By layering perishables onto the network, Amazon is increasing route density and stop productivity. Each same‑day route can carry more orders when baskets are three times larger due to grocery, and the fixed cost of assets such as delivery vans, sort centres, and local staff is amortised over a higher volume of units. That is the core of the monetisation logic: essentials and food are being used not only to drive revenue, but to improve the utilisation of a delivery network already built for general merchandise.
Peers are moving in the same direction, albeit from different starting points. Big‑box and speciality retailers in North America and Europe have spent the last 12–18 months redesigning last‑mile operations and adding direct fulfilment capability, especially for large or complex orders. Those efforts are typically constrained by a smaller store base or less flexible FC networks. Amazon’s position is different in scale and category mix, but the underlying dynamic is similar: same‑day is being treated as a core design variable rather than a bolt‑on.
The Cost Frontier: Where Pressure Shows Up
The claim that speed is cheaper is contingent on several fragile elements. Three stand out: capital intensity, unit cost discipline, and category mix.
First, the capital requirement is extreme. Amazon expects to invest around 200 billion dollars in capital expenditures over the coming years, predominantly in AWS infrastructure but with knock‑on effects for data, AI, and robotics capabilities that support the retail network. On the digital side alone, AWS added 3.9 gigawatts of power in the twelve months to Q4 2025, twice its 2022 base, and plans to double again by 2027. On the connectivity side, the Leo low‑earth‑orbit satellite programme will add around one billion dollars of cost to the North America segment in 2026 before later capitalisation.
Second, maintaining a 14 percent average price advantage versus major online retailers forces constant cost compression. Amazon has made explicit that robotics, cartonisation improvements, and inventory placement are the levers it is using to do this. The regionalisation of inbound, the increase in regions from eight to ten, and the focus on putting more units into each box are all classic supply chain levers translated into hyperscale.
Third, the shift in category mix doubles back into cost. Grocery and perishables are low‑margin, operationally demanding lines. They require cold chain capability, stricter traceability, and more complex returns and waste management. Amazon’s current posture suggests that the trade‑off is being managed by letting these categories dictate network density, then using other lines and marketplace revenue to carry the margin. Worldwide third‑party seller unit mix reached 61 percent in Q4 2025, and marketplace economics, plus retail media and seller services, sit on top of the same fulfilment backbone.
Capacity planning for this model is also exposed to upstream constraints. The cost and availability of AI chips, for example, is a material factor. Amazon has responded by vertically integrating into custom silicon with Graviton and Trainium, landing over 1.4 million Trainium2 chips and seeing most Trainium3 supply effectively committed by mid‑2026. This approach mirrors classic capacity reservation in physical supply chains and underlines how deeply infrastructure and logistics decisions are now intertwined in Amazon’s operating model.
What Amazon’s Model Now Enables and Constrains
Amazon’s next cost frontier is not about shaving a few cents off parcel rates; it is about treating ultra-fast delivery as an infrastructure business that pays for itself through frequency, density, and utilisation. Regionalisation, perishables, marketplace scale, custom chips, and robotics all support this construct. The constraint is that it ties the company even more tightly to high levels of capital deployment, precise demand steering, and a category mix that sustains dense, repeat flows. The operating model now enables service levels that are difficult to replicate at scale, but it obliges Amazon to keep turning operational productivity into the only margin that matters for its retail network: the spread between one‑hour promises and a structurally lower cost-to-serve.
