This thesis examines competition in cloud infrastructure services, drawing on the recent work of competition authorities, in particular the UK Competition and Markets Authority and the European Commission, and on the industrial organisation literature on switching costs, contracts as barriers to entry and costly search. It first describes the market as a concentrated oligopoly, now reinforced by the ties between hyperscalers and AI laboratories, and identifies the technical and contractual sources of customer lock-in. It then examines three tariff instruments, egress fees, committed-spend discounts and granular consumption pricing, through a deterministic cost-scenario exercise that prices one standardised generative AI workload on the public list prices of AWS, Microsoft Azure and Google Cloud. The exercise is illustrative rather than inferential. It shows that, for a mid-market buyer without a negotiated agreement, the barrier works mainly through commitment contracts and search costs, while egress fees, which the CMA found most restrictive for high-volume and data-intensive customers, weigh little at the scale modelled. Finally, the thesis assesses FinOps. Native, provider-supplied tools reduce internal waste but can only recommend deeper commitment to the same platform. Independent price intelligence is inexpensive and most valuable to mid-market buyers, but today it is used mainly by the largest ones, because the cost of acting on it, rather than the cost of obtaining it, still weighs most on smaller organisations.
La tesi analizza la concorrenza nei servizi di infrastruttura cloud, alla luce del lavoro recente delle autorità di concorrenza, in particolare della Competition and Markets Authority britannica e della Commissione europea, e della letteratura di economia industriale sui costi di cambiamento, sui contratti come barriere all'ingresso e sui costi di ricerca. In primo luogo descrive il mercato come un oligopolio concentrato, oggi rafforzato dai legami tra i grandi provider cloud e i laboratori di intelligenza artificiale, e individua le fonti tecniche e contrattuali del lock-in dei clienti. Esamina poi tre strumenti tariffari, le egress fees, gli sconti legati a impegni di spesa e la granularità dei prezzi a consumo, attraverso un esercizio deterministico di scenari di costo che valuta un carico di lavoro standardizzato di intelligenza artificiale generativa ai prezzi di listino pubblici di AWS, Microsoft Azure e Google Cloud. L'esercizio ha finalità illustrative e non inferenziali. Mostra che, per un acquirente di medie dimensioni senza accordi negoziati, la barriera opera soprattutto attraverso i contratti di impegno di spesa e i costi di ricerca, mentre le egress fees, che la CMA ha ritenuto più restrittive per i clienti con grandi volumi di dati, pesano poco alla scala considerata. Infine, la tesi valuta il FinOps. Gli strumenti nativi dei provider riducono gli sprechi interni, ma possono raccomandare soltanto un impegno più profondo verso la stessa piattaforma. La price intelligence indipendente è poco costosa e ha il valore più alto per gli acquirenti di medie dimensioni, ma oggi è usata soprattutto da quelli più grandi, perché è il costo di agire sulle informazioni, non quello di ottenerle, a pesare ancora di più sulle organizzazioni minori.
Locked In by Design: Market Structure, Tariff Complexity and the FinOps Paradox in Cloud Computing
COSTANTINI, RICCARDO
2025/2026
Abstract
This thesis examines competition in cloud infrastructure services, drawing on the recent work of competition authorities, in particular the UK Competition and Markets Authority and the European Commission, and on the industrial organisation literature on switching costs, contracts as barriers to entry and costly search. It first describes the market as a concentrated oligopoly, now reinforced by the ties between hyperscalers and AI laboratories, and identifies the technical and contractual sources of customer lock-in. It then examines three tariff instruments, egress fees, committed-spend discounts and granular consumption pricing, through a deterministic cost-scenario exercise that prices one standardised generative AI workload on the public list prices of AWS, Microsoft Azure and Google Cloud. The exercise is illustrative rather than inferential. It shows that, for a mid-market buyer without a negotiated agreement, the barrier works mainly through commitment contracts and search costs, while egress fees, which the CMA found most restrictive for high-volume and data-intensive customers, weigh little at the scale modelled. Finally, the thesis assesses FinOps. Native, provider-supplied tools reduce internal waste but can only recommend deeper commitment to the same platform. Independent price intelligence is inexpensive and most valuable to mid-market buyers, but today it is used mainly by the largest ones, because the cost of acting on it, rather than the cost of obtaining it, still weighs most on smaller organisations.| File | Dimensione | Formato | |
|---|---|---|---|
|
Costantini_Riccardo.pdf
accesso aperto
Dimensione
1.39 MB
Formato
Adobe PDF
|
1.39 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/118933