ModelGap — Geopolitical AI Procurement & Compliance Marketplace

Source: Priya research — “The Great Fracture — US govt cracks down on Anthropic, Amazon ratted on its own investment”

One-Liner

A procurement intelligence and compliance marketplace that helps enterprises navigate AI model export restrictions, flag supply-chain geopolitical risk, and automatically route to compliant alternatives when a model gets locked.

Target Customer

Primary: Enterprise AI procurement teams at multinational companies operating across US, EU, and Asia — needing a single source of truth for which frontier models are accessible in which jurisdictions.

Secondary: AI startups and VC firms evaluating model suppliers, hyperscaler investments, or building products atop restricted models — needing geopolitical risk scoring before committing stack or capital.

Problem

Three interlocking crises triggered by one event:

The model vacuum. US government just restricted Fable 5 and Mythos 5 to US nationals only. International subsidiaries of US companies, foreign enterprises, and global developers who built on Anthropic now have an unsupported, un-upgradable, legally grey model dependency in their stack. No one knows which model gets restricted next, or where.
The investor–competitor trap. Amazon is an Anthropic investor, yet Amazon CEO Andy Jassy privately flagged Anthropic security risks to the same administration that imposed restrictions. Every hyperscaler-AI startup deal (Microsoft/OpenAI, Google/Gemini frontier, Oracle/Codex) now carries the same latent risk: the “investor” might be your future regulator. How do procurement teams model this?
Procurement can’t keep up. Enterprise procurement workflows are designed for SaaS contracts and cloud credits, not for a world where a model you signed a 3-year deal on can become illegal to use in your Tokyo or London office overnight. There’s no real-time compliance layer for model access rights.

Solution

ModelGap — three layers:

Layer 1: Model Access Registry (The Source of Truth)

Real-time database of which frontier models (Anthropic, OpenAI, Google, Mistral, DeepSeek, etc.) are legally accessible in which jurisdictions
Tracked per model version, not just per provider — because Fable 5 is restricted but Fable 4 isn’t. Mythos 5 is out, Mythos 4 might still be available.
Automated alerts: when a regulation, executive order, or export control change affects a model in your stack, you get a notification with the compliance deadline, remediation options, and estimated migration cost
Includes “investor relationship” metadata — flag models where the investor (Amazon↔Anthropic, Microsoft↔OpenAI) also has a competing or regulatory relationship with the model provider, creating latent risk

Layer 2: Geopolitical Risk Scoring (The Rating)

Every model provider gets a ModelGap Risk Score (0–100) based on:
- Regulatory exposure (which governments could restrict their models?)
- Investor entanglement (is a Big Tech investor also a regulator or competitor?)
- Data sovereignty compliance (does the model meet EU AI Act, China’s CAC rules, India’s DPDP?)
- Supply-chain dependency (does the model rely on a chip design or cloud infra that could be sanctioned?)
Procurement teams embed the risk score into their RFI/RFP process — “Only approve models with Risk Score < 40”
Quarterly report: “The Geopolitical AI Risk Index” — which models are safe, which are one executive order away from breaking your stack

Layer 3: Model Routing Engine (The Marketplace)

When a model in your stack gets restricted or downgraded beyond your risk threshold, ModelGap automatically finds and benchmarks the best alternative in your jurisdiction
Example: Your London office was on Mythos 5 → restricted to US nationals → ModelGap routes to Mistral Large 3 or DeepSeek V4, benchmarks performance on your workload types, estimates migration cost and timeline
Multi-provider fallback architecture: procurement teams define “if X is restricted, route to Y or Z” — the engine handles the compliance mapping
Marketplace layer: alternative model providers (Mistral, DeepSeek, GLM, AI21, Cohere) pay listing fees or referral commissions to be surfaced as “compliant alternatives” in restricted markets

Why Now

It already happened. This isn’t a hypothetical — Fable 5 and Mythos 5 went from “available globally” to “US nationals only” overnight. Every enterprise using Anthropic internationally is in limbo. The market needs an answer today, not next quarter.
The Amazon–Anthropic leak exposes a structural risk in every hyperscaler-AI startup deal. Procurement teams who ignored investor relationships as a risk factor now have a case study proving it matters. ModelGap gives them a framework to quantify and price that risk.
GPT-4.5 and o3 are being retired (June 27 and August 26). OpenAI is consolidating around GPT-5.x before IPO. More models will disappear. More restrictions will follow. The window to build the registry is now, before the next domino falls.
The IPO pipeline is loading. OpenAI S-1 filed, EngineAI humanoid robot filed for HK IPO. Every investor prospectus now needs a “geopolitical AI model dependency” risk disclosure. ModelGap provides the data to draft that disclosure.
The Great American AI Act (mandatory audits) + US Executive Order (voluntary 30-day review) + EU AI Act (mandatory compliance) = a regulatory trifecta that demands a single, continuously updated compliance intelligence layer.

Pricing Model

Access Registry API: $0.005 per model-check query (enterprise flat: $15K–$50K/yr for unlimited queries + alerts)
Risk Scoring Reports: $5K per provider assessment (annual renewal: $2K)
Marketplace commissions: 5–15% referral fee for alternative model placements
Geopolitical Risk Index subscription: $25K/yr (quarterly reports + ad-hoc briefings)
Integration fee: $10K one-time for procurement toolchain integration (Coupa, SAP Ariba, Workday procurement modules)

Estimated TAM: $800M by 2028 (2,000 multinational enterprises × $40K avg subscription + marketplace commissions on the estimated $5B in alternative model procurement it would influence annually)

Wedge

The first-mover advantage is credibility and speed. A registry like this isn’t technically hard — it’s hard because someone needs to track every executive order, every export control revision, every hyperscaler investment filing across every jurisdiction, in real time. The company that builds the registry first becomes the default source of truth for every procurement team in the world. By the time competitors start, ModelGap has the data moat.

Competition

Player	Gap
S&P Global / Moody’s	Geopolitical risk for sovereign debt, not AI models. Wrong dataset, wrong distribution.
Standard procurement tools (Coupa, SAP Ariba)	Handle pricing and contracts, not geopolitical compliance or model access rights.
In-house legal/compliance teams	Doing this manually, jurisdiction by jurisdiction, reacting after the fact. Reactive, not preventive.
Law firms (international trade practices)	Billable-hour advisory for each event. No product, no real-time API, no automated routing.
Cloud provider lock-in (AWS Bedrock, Azure OpenAI)	Each surfaces only their own models. No cross-provider risk comparison. Incentive-aligned to recommend their own stack.

Exit / Outcome

Most likely acquirer: S&P Global or Moody’s — they already sell risk intelligence to enterprise procurement teams. ModelGap gives them the AI model vertical before their competitor buys it. Alternatively, Accenture (already in AI governance consulting) packages it as part of their Great American AI Act compliance offering. Could be standalone at $30M ARR within 2 years given the regulatory velocity.

Filed by

Rohan (Idea Generator) — 2026-06-14