Research opportunity · Concentration, Stack & Sovereignty
Do AI startups backed by a cloud giant end up running on that giant's cloud?
Earlier version (v1)
This is RO-LD11-04 as it was published at v1, kept unchanged. The current version is v2; its changes are listed there.
Question
Among AI startups that raised money from Amazon, Microsoft or Google (directly or through their venture arms), compared with similar startups funded by independent investors, how much more likely are they to build on the investing firm's cloud and models, and does that tie persist or loosen in later funding rounds?
Not yet measuredOne PhD chapter or paperLow to medium costNo study found
Why it matters
FTC staff documented that, in three of the largest cloud–AI partnerships, the AI developers must spend a large share of the investment on the investor's cloud, and investor analysts describe hyperscaler investment as a way to lock in cloud commitments. Whether this pattern extends across the wider startup ecosystem, and whether it amounts to steering, is not measured. This is one concrete, observable route to the "AI-specific foreclosure" effect that current evidence cannot quantify.
The debate lists this among what we still do not know: a quantified AI-specific bundling/default foreclosure effect
Evidence that motivates it
- FTC staff found that major cloud–AI-developer partnerships included equity, revenue-sharing and varying consultation, control and exclusivity rights, while not establishing antitrust liability.
FTC staff found equity, revenue-sharing and varying exclusivity and control rights in major cloud–AI-developer partnerships, without establishing foreclosure; this study measures one downstream outcome across many smaller deals. - The UK CMA found AWS and Microsoft hold significant market power in UK cloud and identified egress fees, interoperability barriers and Microsoft software licensing as constraints on switching and multi-cloud.
The CMA found switching constraints in cloud generally, but the claim explicitly does not license AI-specific foreclosure; that is the gap being measured.
A possible design
Build a sample of AI startups founded 2019–2025 from a funding database (Crunchbase or PitchBook through a university licence, or the free Crunchbase tier plus press releases). Treated firms received equity from Amazon, Microsoft, Google or their venture arms; comparison firms are matched on founding year, sector, location, stage and round size. Measure each firm's cloud and model stack from three public traces: job postings (skill mentions such as AWS, Azure, GCP, Bedrock, Vertex, Azure OpenAI), public technical documentation and marketplace listings (for example appearing in a cloud provider's marketplace), and press releases announcing cloud partnerships. Estimate the difference in the probability of using the investor's cloud as primary platform, and use an event-study around the investment round to separate selection (firms already on that cloud) from post-investment shifts. Track multi-cloud adoption in later rounds.
Data needed
Funding records for a few thousand AI startups; job-posting data (public postings collected over time, or an academic licence to a postings dataset); cloud marketplace listings; a hand-coded validation sample.
Feasibility
One PhD chapter or paper. Low to medium: free if the university licenses a funding database and job-posting data; otherwise a few thousand US dollars for data access. Selection is the main threat (investors back firms already on their cloud); the event-study and pre-investment stack measures address this only partly. Cloud credit terms are rarely public, so the study measures stack outcomes, not contract terms, and cannot by itself establish illegality or foreclosure.
Has this been done?
The search found no study that does the core design (first-pass screen of 12 October 2026; not yet checked in depth).
No study was found that systematically measures whether startups funded by cloud hyperscalers build on their investor's cloud compared with matched startups. The FTC staff report covers three large partnerships through compulsory orders; Widder and Kim map hyperscaler startup investments at scale from Crunchbase, including sole- or lead-investor roles and accelerators that offer cloud credits, but their published summary reports no measure of which cloud the investees actually use; investor and commentary pieces describe the motive and the "cloud credit circuit" with illustrative figures only. A matched, ecosystem-wide measurement of the stack outcome remains open.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 2 of the 4 nearest works below were judged from an abstract, summary or search listing only, and 1 has details that could not be confirmed; each is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search on the date shown; Google Scholar, SSRN, paywalled journals and subscription databases (PitchBook, CB Insights research) were not reachable beyond public pages. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- US Federal Trade Commission, "FTC Issues Staff Report on AI Partnerships & Investments Study," press release and staff report (17 Jan 2025) Judged from an abstract, summary or search listing only
Covers Microsoft–OpenAI, Amazon–Anthropic and Alphabet–Anthropic; reports that AI developers must spend a large share of investment on the partner's cloud and that partners hold varying exclusivity rights. Three deals only; no measurement across the startup ecosystem. Press release read; the full report was not read in full. - Gao, "Q3 2026 Fewer Deals, Bigger Bets: How Corporate Capital Is Concentrating US AI Venture Activity," PitchBook analyst note (21 Jul 2026)
Describes corporate capital concentrating AI venture activity and states hyperscalers invest partly to lock in cloud commitments; the public page gives no data on portfolio companies' actual cloud use. - Lee, "How a 2016 Accounting Rule Fueled Big Tech's Investments in AI Startups," ProMarket (19 May 2026)
Commentary arguing an accounting rule encourages Big Tech equity stakes and describing a cloud-credit loop, drawing on filings and the FTC report; illustrative figures only, no systematic measurement. - Widder & Kim, "How Big Cloud becomes Bigger: Scrutinizing Google, Microsoft, and Amazon's investments," SSRN working paper (Aug 2025), doi:10.2139/ssrn.5377426 Judged from an abstract, summary or search listing onlySome details not confirmed
Uses Crunchbase records to show that Google, Microsoft and Amazon take part in far more startup deals than other large tech firms, often as sole or lead investor and through accelerators that offer cloud credits (some Google programmes require use of Google Cloud). The authors' summary (Tech Policy Press, 12 Aug 2025) reports no data on which cloud investees use and no comparison with independently funded startups. Bibliographic record and summary read; the SSRN page itself could not be opened.
Search queries used
corporate venture capital hyperscaler investment AI startups cloud provider choice lock-in empiricalcloud credits startups steering AI startups to AWS Google Microsoft empirical study job postings tech stackFTC 6(b) report AI partnerships cloud service providers January 2025 staff reportBig Tech investments in AI startups paper compute-for-equity circular deals cloud credits antitrust empirical dataset 2025 2026empirical paper Big Tech corporate venture capital AI startups cloud provider adoption portfolio companies Amazon Microsoft Google investment effect
Provenance
RO-LD11-04 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-11 debate on 2026-10-12. Checked by an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.