The procurement consortium model: how buy-side firms are pooling MCP connector contracts across expert networks

Buy-side research operations teams are quietly rewriting how their firms buy access to expert networks. The shift is not about which vendor wins a logo war. It is about who signs the contract, on what terms, and on behalf of how many funds. Through late 2025 and into 2026, a recognizable pattern has emerged: connector budgets carved out as separate line items, Model Context Protocol access bundled across multiple expert networks under a shared compliance schedule, and procurement conversations that increasingly look less like vendor selection and more like consortium formation.

Our read is that this is the early structural shape of how AI-era primary research gets wired into LLMs at scale. The unit of consumption is changing from a seat to a call to a token, and the contracting layer is adapting faster than the pricing models. When the underlying economics are unstable, buyers coordinate. They always have.

What changed in the contracting layer

The traditional expert-network procurement conversation was bilateral and simple in structure. A fund signed a seat-based or call-pack contract with one or more networks, the compliance team layered MNPI controls on top, and the research desk consumed calls and transcripts through a portal. Pricing was predictable enough that procurement could be handled at the fund level, and contract terms varied across networks but the variance did not matter operationally because the consumption pattern was human, slow, and auditable through call-by-call workflows.

MCP broke that equilibrium. Once an expert network exposes its transcript library to a model like Claude through a connector, the consumption pattern changes shape. A research analyst no longer reads one transcript at a time, they query a corpus. An agent no longer schedules one call, it triages a backlog. The unit of consumption stops being the call and starts being the query, the token, or the agent session, depending on the vendor. And the compliance perimeter stops being the analyst's screen and starts being wherever the model context lives.

None of this is a problem at one vendor. It is a problem at three or four, when each connector has its own audit log format, its own citation provenance scheme, its own retention default, and its own opinion about what an agent is allowed to do with material the network surfaced. The fragmentation is not malicious. It is the natural state of a category where every vendor is shipping connector functionality on its own product clock. But for a multi-strategy hedge fund running ten portfolio managers against three expert networks through two model providers, the fragmentation is the operational problem of the year.

Why a consortium, not a procurement RFP

The instinctive response to vendor fragmentation is a heavier procurement function. Standardize the RFP, dictate the security schedule, push back on retention defaults, win on volume. Large allocators have been doing this for two decades on market data, custody, and prime brokerage. The reason it does not fully work for MCP connector access is that the buyer is not negotiating against one vendor's pricing power, the buyer is negotiating against the absence of an industry standard.

A procurement team can demand a 30-day audit log retention from one network. It cannot demand that three networks adopt the same audit log schema, because none of them owe the buyer that, and none of them are coordinating with each other. The asymmetry of a single-fund RFP against a category-wide standardization problem is exactly the asymmetry that historically produces collective buy-side action. The MFA coordinated hedge fund positions on swap execution venue rules. AIMA coordinated international manager positions on MiFID II research unbundling. The FIA ran reference-data consortia in derivatives. The common pattern: when the contracting question is structural rather than commercial, individual buyers underdeliver and collective buyers set the floor.

The Aiera consortium-backed content layer is the most visible early example of this pattern in AI-era primary research. The structure is notable not because of which content sources it routes, but because it is jointly funded by buy-side and sell-side participants rather than bilaterally licensed. That funding model is the consortium tell. Bilateral licensing produces winners and losers among vendors. Consortium funding produces a shared infrastructure layer that no single vendor controls and no single buyer underwrites alone.

We read the Aiera structure as a category signal, not a vendor verdict. It says the market has decided that the contracting layer for AI-era primary research access is a candidate for shared infrastructure, the same way market data consortia, reference data utilities, and KYC utilities became shared infrastructure in earlier cycles. The question for the rest of the category is not whether consortium contracting will exist. It is whether each vendor will participate in a consortium structure or insist on bilateral licensing while the market routes around them.

The three forces pushing buy-side coordination

Connector terms vary widely and create compliance fragmentation

The first force is the one research operations teams feel daily. Each expert network's MCP connector ships with its own contractual approach to the operational primitives that matter under a model-mediated workflow: which fields appear in the audit log, how citation provenance is preserved when a model summarizes a transcript, how long context is retained on the network's side after a query, what the MNPI handling protocol looks like when an agent rather than a human is consuming the content, and whether the connector exposes the same metadata to every downstream model or differs by provider.

For a single fund running one network, these terms are negotiable as a one-off addendum. For a multi-strategy firm running three networks, the addenda do not compose. A compliance officer cannot write a unified policy that says how transcript material will be handled by the firm's models when each network's connector treats handling differently. The compliance function either has to write three policies, one per network, or has to push the networks toward a common floor. The first path scales badly. The second is the consortium path.

Pricing is in flux as MCP shifts the unit of consumption

The second force is that nobody yet knows what the right unit of consumption is for expert content delivered through a model. Per-seat pricing assumes the consumer is a human. Per-call pricing assumes the unit is a scheduled conversation. Neither maps cleanly to an agent that ingests a hundred transcripts in a research sprint, surfaces three, and prompts a human analyst to schedule one follow-up call.

Vendors are experimenting. Some are pricing connector access as an enterprise overlay on top of existing seat contracts. Some are introducing query-based pricing tied to model calls. Some are bundling connector access into a higher-tier subscription and accepting that the early-mover usage data is more valuable than the incremental revenue. The pricing volatility is rational on the vendor side, and disorienting on the buyer side. When a category's pricing model is in active flux, buyers protect themselves through coordination. They cannot lock in a five-year price on connector access individually, because the unit of consumption may not exist in five years. They can lock in shared contractual terms collectively, which is what consortium contracting is for.

The parallel data-room category gives a useful read. Hebbia's integration with SS&C Intralinks is structurally a deal-room consortium pattern in miniature: the AI layer sits on top of a shared content custodian rather than ingesting deal documents bilaterally from each sponsor. The economic logic is the same. When the AI consumption pattern is uncertain, the buyer pushes toward shared infrastructure that abstracts away the per-vendor contract.

LPs are starting to ask GPs how they govern AI access to expert content

The third force is upstream of the research desk entirely. Large allocators have been refining their operational due diligence questionnaires through 2024 and 2025 to cover AI governance, and the questions are moving past the generic toward the specific. How does the GP control which models can access expert call transcripts. What is the retention default on connector-mediated content. How is MNPI handled when an agent rather than an analyst consumes the source. Who at the firm signs off on a new MCP integration.

These are not theoretical questions. They are the operational questions an allocator asks before underwriting a multi-year mandate. A fund that can answer them with a unified policy across its expert-network relationships answers them faster and cleaner than a fund that has to compose an answer out of three different vendor schedules. The allocator pressure pushes the buy-side toward unified contracting terms even when individual research desks would tolerate fragmentation, because the cost of fragmentation is no longer paid in operational drag, it is paid in ODD friction.

What the vendor landscape looks like under consortium pressure

The expert network category is unusually well-positioned to absorb consortium contracting. The networks themselves are not commoditized, each carries differentiated expert rosters, geographic coverage, and compliance reputation. What is being standardized is not the content, it is the contracting layer that sits between the content and the model. Vendors can comply with a consortium-set floor on audit logs, citation provenance, and MNPI handling without giving up content differentiation.

The public signals across the category suggest the larger networks are leaning into MCP rather than resisting it. Guidepoint, Third Bridge, and AlphaSense each exposed transcript libraries to Claude via MCP through the late 2025 and 2026 window. AlphaSense's USD 7.5B raise and Accenture Ventures partnership signaled that the capital markets and the systems integrator channel both see the AI-distribution layer as the strategic axis, not the seat-license business. None of these moves are a verdict on consortium contracting, but they are consistent with a category where the vendors expect the contracting layer to consolidate and are positioning to ship connector access at scale rather than to defend a per-seat moat.

The vendors that struggle under consortium pressure are typically the ones whose contracting terms encode product fragility. If a network's audit log format is idiosyncratic because the underlying logging system was built before MCP existed, a consortium-set floor forces an engineering investment. If a network's retention default is permissive because the legal team has not yet engaged with model-mediated consumption, a consortium-set floor forces a policy update. These are tractable problems for the larger networks. They are existential for any vendor whose connector was a marketing exercise rather than a product investment.

The counterargument

The honest counter to the consortium thesis is that buy-side coordination is hard to sustain. The MFA and AIMA precedents are real but they took years of regulatory pressure to organize, and even then the coordination was uneven across managers. A multi-strategy hedge fund that signs a consortium agreement on MCP connector terms gives up some negotiating flexibility relative to a fund that goes bilateral, and not every fund is willing to make that trade.

There is also a vendor counter. Networks that own differentiated content can plausibly argue that consortium contracting strips them of the ability to charge for the differentiation. If every connector has to ship with the same audit log schema, the same retention default, and the same MNPI protocol, the floor of the contract becomes uniform and the only remaining lever is content quality. That is fine for the networks with the strongest rosters, and harder for the networks competing on price or convenience.

Our view is that both counters are real and neither defeats the thesis. Buy-side coordination is hard, and it is happening anyway, because the alternative is each fund writing its own three-vendor compliance policy under LP scrutiny. Vendor differentiation gets squeezed at the contracting layer, and the networks with the strongest content are already moving to position around that squeeze rather than against it. The consortium thesis does not require every fund to join or every vendor to agree. It requires enough of the buy-side and enough of the sell-side to set a floor, and the Aiera structure plus the procurement reset INFLXD has been tracking suggests the floor is already being set.

What to watch over the next 18 months

The operational signals worth tracking are concrete. Whether the larger multi-strategy hedge funds publicly disclose participation in connector-access consortia, or whether the structure stays one layer deeper, organized through prime brokers and outsourced trading partners. Whether the existing buy-side trade associations, the MFA and AIMA, formally engage on MCP contracting standards the way they engaged on research unbundling. Whether the expert networks coordinate among themselves on a common MCP schedule, the way market data vendors eventually coordinated on FIX and FpML, or whether the coordination is buyer-driven and vendor-reactive.

The pricing signal is also worth tracking. If query-based pricing settles as the dominant model for connector access, consortium contracting becomes structurally easier because the unit of consumption is observable and auditable in a standard way. If the category fragments across per-seat, per-call, and per-query models, consortium contracting gets harder because the buy-side coordination has to span pricing structures that do not map onto each other.

And the LP signal is the most leading. When the operational due diligence questionnaires from the top ten allocators converge on a shared set of questions about AI access to primary research, the consortium pressure becomes self-reinforcing. Funds will not coordinate against their own vendors voluntarily. They will coordinate when their LPs ask the same question of all of them in the same quarter, and the cheapest answer is a shared contracting layer.