Hyperscaler-Exclusive-Then-Multi-Home
Initial deal grants hyperscaler exclusivity in exchange for investment + capacity; AI lab later restructures to multi-home as compute demand outgrows one provider
An AI lab and a hyperscaler enter into a formation deal that grants the hyperscaler exclusivity or strong preference (API exclusivity, capacity exclusivity, distribution exclusivity, model rights). Over time — typically 18-36 months at frontier scale — the AI lab''s compute demand or strategic posture outgrows what one hyperscaler can supply, and the relationship restructures to allow multi-homing. The original hyperscaler typically loses preferential terms, retains some preferred status, and absorbs the friction cost of the restructure. Microsoft-OpenAI 2019-2025 is the canonical case (Azure API exclusivity, ROFR on compute, $13B+ investment; restructured 2025 with Microsoft losing ROFR and OpenAI committing $250B to Azure but free to multi-home). Anthropic-AWS-then-Google follows the same arc compressed (AWS as primary then Anthropic adding Google TPUs at scale).
Fits when: AI lab is in early-frontier stage and needs single-hyperscaler-level capital + capacity to scale, hyperscaler is willing to underwrite the bet and wants strategic AI workload + distribution, both parties accept that exclusivity will eventually constrain growth.
Does NOT fit when: AI lab can negotiate multi-cloud terms at formation (rare for early-stage; possible for later-stage labs with credible alternative offers), hyperscaler accepts non-exclusive deal terms from the start, or the AI lab''s compute demand profile is small enough that one hyperscaler will indefinitely suffice.
Structural shape (formation phase): exclusivity (API, capacity, or distribution; sometimes all three), large investment (often $1B+ at frontier scale), multi-year compute commitment, ROFR on subsequent compute or capability, board observer rights, distribution rights through hyperscaler marketplace, sometimes co-developed product surfaces (Copilot was the Microsoft-OpenAI product layer).
Structural shape (restructure phase): exclusivity lifted (replaced by preferred-status or non-exclusivity), ROFR replaced by negotiated rights, compute commitment renegotiated at scale, distribution rights preserved or expanded, governance terms refreshed, sometimes valuation revaluation (Microsoft-OpenAI restructure included revaluation of Microsoft''s position to ~$135B / 27% on as-converted diluted basis).
Five kill-list moves: under-investing in restructure mechanics at formation (no defined trigger or process for moving off exclusivity, so restructure becomes ad-hoc and contentious); exclusivity scope creep (the formation deal''s exclusivity expands via interpretation to cover scenarios the parties didn''t intend); ROFR-as-anchor (right-of-first-refusal becomes the effective veto, blocking the lab from credibly negotiating with alternatives); valuation timing trap (restructure happens when both sides have leverage at very different times, so neither side gets the valuation it expects); leadership-relationship-only mitigation (the relationship runs on the founder-and-CEO friendship, which doesn''t survive restructure stress).
AI-specific considerations: compute demand at frontier scale is growing faster than any single hyperscaler can build capacity, so multi-homing is increasingly the default end-state for top labs; chip diversification (Trainium, TPU, NVIDIA GPU) is now a competitive moat for labs that can multi-platform their training; AI lab corporate restructuring (OpenAI''s PBC, Anthropic''s expansion) interacts with the partnership restructure and is often the forcing function; regulatory scrutiny on hyperscaler-AI-lab tie-ups (FTC 2024, UK CMA, EU competition authorities) creates external pressure to multi-home.
Hypothetical, illustrative — not actual deal terms. Practitioners should not use these clauses verbatim; they illustrate structure and what to negotiate.
- Sunset exclusivity with defined triggerscope_limits
Hyperscaler''s exclusivity rights under Section [X] shall sunset on the earlier of: (i) the [N]th anniversary of the Effective Date; (ii) the date on which Lab''s aggregate compute consumption exceeds [committed threshold] across all infrastructure providers; (iii) the date on which Lab raises external capital of $[threshold] from one or more strategic investors at a pre-money valuation of $[threshold] or greater; or (iv) the occurrence of a Change of Control of Hyperscaler. Upon sunset, the exclusivity rights shall be replaced with the Preferred Partner Rights set forth in Schedule [Y], which grant Hyperscaler [most-favored-pricing / priority capacity reservation / first-look on new model releases] but no exclusivity.
Why it matters. The single most-protective formation clause in this pattern. Without defined sunset triggers, restructure becomes a renegotiation under stress where both sides have built up positions and grievances. With defined triggers, both parties enter the formation deal with shared understanding of when and how exclusivity ends — restructure becomes a scheduled transition, not a crisis. The Preferred Partner Rights structure (replacing exclusivity with negotiated preference) is the off-ramp that preserves the relationship through the transition.
- ROFR with practical-feasibility limitsrights_of_refusal
Hyperscaler''s right of first refusal under Section [X] applies only to (i) compute capacity additions exceeding $[threshold] in any twelve-month period, (ii) where Hyperscaler can deliver such capacity within Lab''s required timeline (defined as no later than [N] months after Lab''s request), and (iii) at pricing no higher than [median of three competing third-party bids]. If Hyperscaler declines or fails to match within [N] business days, Hyperscaler''s ROFR shall be deemed waived as to that capacity addition without prejudice to ROFR on future additions. Lab shall not be required to delay deployment to await Hyperscaler''s response beyond the foregoing window.
Why it matters. Unconstrained ROFR is the practical equivalent of veto — the lab cannot credibly negotiate with alternatives because every offer must be tendered back to the hyperscaler with right-of-first-refusal. The capacity-delivery feasibility constraint, the pricing benchmark, and the deemed-waiver mechanic together convert ROFR from a veto into a useful first-look right. The waiver-without-prejudice language prevents the hyperscaler from arguing that one decline forecloses future ROFR — preserving the relationship through individual capacity decisions.
- Restructure mechanics at formationrestructure
The Parties acknowledge that this Agreement may require material restructuring in connection with (i) Lab''s growth beyond [defined scale milestones], (ii) Lab''s corporate restructuring (including conversion to or recapitalization as a public-benefit corporation), (iii) material changes in the AI infrastructure market, or (iv) regulatory developments materially affecting the relationship. Upon written notice from either Party identifying a Restructuring Event, the Parties shall enter into good-faith negotiations to amend this Agreement to address the changed circumstances, with the following baseline commitments: (a) economic value transferred under this Agreement shall be preserved on a no-economic-loss basis for both Parties to the extent commercially feasible; (b) Hyperscaler''s preferred-status rights shall be preserved in substance even where their form must change; (c) Lab''s strategic independence and ability to address market needs shall be preserved; (d) the Parties shall use an agreed independent process (described in Schedule [Z]) to resolve any inability to reach agreement within [N] months. Failure to reach agreement under (d) shall not constitute breach of this Agreement but shall trigger the Restructure Arbitration Process.
Why it matters. Restructure mechanics drafted at formation are the structural alternative to ad-hoc restructure-under-stress. The no-economic-loss baseline, preserved-in-substance commitment, and independent process (typically a structured arbitration with industry-expert panel) together create a path forward that doesn''t depend on the personal relationship at the time of restructure. Microsoft-OpenAI did not have this structure at formation; the 2025 restructure took 18+ months and was contentious. This clause is what its successors should include.
- Governance refresh on material changegovernance
On the occurrence of any Restructuring Event, the governance terms of this Agreement (including Hyperscaler''s Board Observer Rights, Information Rights, and any consent rights) shall be refreshed through good-faith negotiation. The default position in such refresh shall be that governance rights expand only to the extent the underlying economic relationship has materially expanded, and contract proportionally where the relationship has narrowed. No expansion of governance rights shall occur without corresponding expansion of economic commitment or material increase in deal-specific investment by Hyperscaler.
Why it matters. Governance refresh on material change is the practical defense against governance creep across the formation-to-restructure arc. Without refresh, the original governance terms are anchored to the original economic relationship; as the relationship grows or changes, governance rights become disproportionate. The default-to-proportional-refresh rule keeps governance and economics aligned through the relationship''s evolution, preventing the asymmetry that can stall future restructures.
The intuitive moves that alliance research has documented as predictably failing for this pattern. Each one comes with a mitigation that addresses the underlying mechanism, not just the symptom.
- 1.Under-investing in restructure mechanics at formation
Formation deal has no defined trigger or process for moving off exclusivity, so restructure becomes ad-hoc and contentious.
Why it fails. Without pre-agreed restructure triggers and mechanics, both parties enter restructure with maximally opposed positions, no shared framework, and personal-relationship dependencies that may have decayed. Restructure timelines stretch to 18+ months (Microsoft-OpenAI 2024-2025) and become news cycles instead of business decisions.
Mitigation. Sunset exclusivity clauses with defined triggers. Restructure mechanics at formation with no-economic-loss baseline and independent process. Governance refresh on material change.
- 2.Exclusivity scope creep
Formation deal''s exclusivity expands via interpretation to cover scenarios the parties didn''t intend (model release exclusivity creeping to all model types, capacity exclusivity creeping to all infrastructure).
Why it fails. Exclusivity is interpreted broadly by the party benefiting from it; narrowly by the party constrained by it. Disputes over scope are slow to resolve and politically costly. The lab discovers it has less freedom than it thought.
Mitigation. Exclusivity drafted with explicit positive and negative scope language. Examples and carve-outs in the agreement. Periodic written affirmations from both parties on scope interpretation.
- 3.ROFR-as-anchor
Right of first refusal becomes the effective veto blocking the lab from credibly negotiating with alternatives.
Why it fails. Unconstrained ROFR means every third-party offer must be tendered back to the hyperscaler. Third parties realize they''re running stalking-horse bids and stop participating. The lab loses the credible alternative that makes the ROFR market-disciplined.
Mitigation. ROFR with practical-feasibility limits (delivery timeline, pricing benchmark, deemed-waiver). Carve-outs for capacity below threshold. Explicit acknowledgment that ROFR exercise is finite, not perpetual.
- 4.Valuation timing trap
Restructure happens when both sides have leverage at very different times (lab in mid-fundraise, hyperscaler in mid-recapitalization, etc.) so neither side gets the valuation it expects.
Why it fails. Valuation is inherently temporal — the same business is worth different amounts at different points in a market cycle. Restructure that combines economic restructure with valuation restructure puts both into the same negotiation, and one side''s leverage on one dimension typically corrupts the other.
Mitigation. Restructure mechanics separated from valuation events. Independent valuation mechanism (mutually-agreed appraiser) for any equity component of restructure. Phased restructure with valuation locked at trigger date, not negotiation date.
- 5.Leadership-relationship-only mitigation
The relationship runs on the founder-and-CEO friendship; structural mechanics are minimal because trust substitutes for governance.
Why it fails. Leadership turns over; personal relationships don''t transfer through transitions. When restructure stress arrives, the people who structured the relationship are no longer the people negotiating the restructure. Trust-based assumptions get re-litigated under new leadership.
Mitigation. Institutionalize the restructure mechanics regardless of relationship quality. Restructure clauses drafted as if the original principals will not be at the table during restructure. Quarterly business reviews documenting evolving terms.
The primary-source research this pattern is grounded in.
- Doz (1996) — Alliance evolution as developmental processes; conditions for productive restructure[theory]
- Reuer & Zollo (2005) — Termination outcomes in strategic alliances; restructure dynamics[empirical]
- Khanna, Gulati & Nohria (1998) — Learning races and incentive structure within strategic alliances[empirical_theory]
- Greve, Mitsuhashi & Baum (2013) — Partnership endings and the role of pre-defined termination provisions[empirical]
- Lavie (2007) — Alliance portfolios and firm performance; multi-homing implications[empirical]
- Hamel (1991) — Learning races and asymmetric capability transfer[theory]
- Gulati (1998) — Equity vs contractual governance selection[empirical_theory]
- Bamford, Gomes-Casseres & Robinson (2004) — 30-70% alliance failure rate[empirical_review]