2025 Annual Forecast: GCC Edition

This is the third document we have shipped that takes its accuracy seriously enough to publish a grading methodology in advance. Below are ten specific calls for 2025. We will grade each in our year-end retrospective on the criteria we name here, not on criteria we invent later.

The ten calls

1. Anthropic raises at a $60B post-money valuation in 2025. Inside two raises if necessary. Anchor leads include a sovereign-Gulf check. We grade verified if a public round closes between $50B and $75B with a confirmed Gulf participant.

2. A Chinese frontier-class reasoning model triggers a public Nvidia re-pricing inside the first six weeks of the year. Single-day drawdown in the high-single-digit to low-double-digit range. We grade verified if the drawdown occurs within the window. This is the continuation of our 2024-W44 call.

3. The Trump administration ships a US AI Action Plan inside the first one hundred and twenty days. Three components anchor the plan. Federal procurement standardization. Compute-export policy. A formal posture toward sovereign-Gulf compute partnership. We grade verified if all three are present in the published plan.

4. Saudi Arabia ships a sovereign-AI commercial entity that exits PIF preparatory mode and enters operational mode. We expect Humain or a similarly named vehicle to be publicly announced inside Q1, with a CEO, a board, and a stated capex commitment of $10B or more in the first year. We grade verified on entity launch plus capex announcement.

5. MBZUAI publishes a reasoning model in the R1 class. Arabic-native training, distilled variants for on-device use, open release or permissive license. We grade verified on a release with a published benchmark inside ten percent of frontier reasoning performance, by end of Q4 2025. Partial if the release lands in 2026.

6. MCP crosses 10,000 public server implementations by year-end. This is the continuation of our 2024-W46 call. We grade verified on the public Anthropic MCP directory crossing the threshold and on at least two of the major LLM vendors shipping client-side support.

7. Voice agents become the default support channel for at least three top-twenty GCC banks. Production deployment, not pilot. We grade verified on three named institutions with publicly cited agent deployments. Partial if two of three are deployed and the third is in pilot.

8. DIFC's AI license framework crosses fifty licensed entities. The framework was nascent through 2024. We expect it to catch the sovereign-AI flywheel in 2025. We grade verified on the public DIFC register crossing fifty entities, with at least ten of those being foreign-incorporated subsidiaries.

9. The default code model inside Cursor flips to Anthropic Claude as the published default setting, and stays there through year-end. This is the continuation of our 2024-W39 call. We grade verified on the published default and the absence of a sustained reversal.

10. Apple Intelligence underperforms its Q1 product launch by a visible margin against the September 2024 keynote claims. A delayed Siri Pro release, a feature retrenchment, or a quiet pricing revision. We grade verified if any one of these three is publicly confirmed by Apple before mid-year.

Methodology notes

We assign each call a confidence band when we publish. We do not adjust the band after the fact. Our grade is binary at the end of the year. Verified, partial, or wrong. We publish a graded scorecard each quarter and a final scorecard at year-end.

We also commit to the inverse: if a forecast lands more than ten percent off our distribution targets in either direction, we will publish a methodology revision. A 90% verified rate is not credible and we will say so. A 50% verified rate means we are guessing and we will say so.

What this means for the Gulf

Three structural reads frame 2025 in this part of the world.

Sovereign capital becomes the price-setting AI dollar. PIF, ADQ, Mubadala, ADIA, and QIA collectively price more frontier AI rounds in 2025 than any single US institutional buyer. Founders who can land a credible Gulf anchor in Q1 will out-raise their San Francisco peers on both terms and valuation.

Arabic capability becomes a commercial moat. Through 2024 the Arabic-AI conversation was a research story. In 2025 the production work begins, and the GCC institutions that have already invested in sovereign training infrastructure will own customer relationships the global labs cannot touch from a New York or London base.

Regulation becomes a feature, not a friction. The EU AI Act and US sectoral approaches are inflexible and slow to adapt. DIFC, ADGM, and Bahrain's framework are explicitly designed for iterative revision. Operators who pick GCC jurisdiction in 2025 will ship products in eighteen months that European competitors will be explaining to regulators in eighteen months.

We will grade each of these ten calls in 2025-W25 mid-year audit and again at year-end. That scorecard is the load-bearing piece of our 2025 program. The point of public predictions is that they are gradable in public, and we intend to be graded.