How We Score the Magnificent Seven AI Giants Every Week
Five dimensions. One weekly score. Here is exactly how the SEVENAI Momentum Index works — every signal we track, every weighting decision we made, and why.
The SEVENAI Momentum Index is a weekly proprietary score — 0 to 100 — for each of the Magnificent Seven AI companies. It is not a stock price prediction, a revenue forecast, or an endorsement. It is a competitive momentum measure: who is gaining ground in the AI race, who is losing it, and how fast. Every score is published every Monday and updated without revision — we stand behind our calls and correct them publicly when we get them wrong.
The index is built on five dimensions. Each dimension is independently scored and then weighted to produce the final number. Here is the complete methodology for each one.
Model benchmarks are the most objective signal available in the AI race. They tell you, with more precision than any earnings call or press release, whether a company's AI models are genuinely improving or standing still. We weight this dimension highest because model capability is the foundation of every other competitive advantage — a company with the best models attracts the best developers, commands the highest prices, and earns the most enterprise trust.
We track four benchmarks, chosen because they are hard to game, independently verifiable, and measure capabilities with direct commercial value.
| Benchmark | Measures | Sub-weight | Commercial relevance |
|---|---|---|---|
MMLU Massive Multitask Language Understanding | Knowledge breadth across 57 academic and professional subjects | 35% | Best proxy for enterprise readiness across diverse professional domains |
HumanEval OpenAI coding evaluation | Writing correct, functional code from natural language descriptions | 30% | Software development is the highest-value enterprise AI use case |
MATH Competition mathematics | Multi-step mathematical reasoning requiring genuine logical chaining | 20% | Best proxy for financial analysis, scientific reasoning, structured problem-solving |
Frontier Evals GPQA · ARC-AGI · AIME | Expert-level reasoning tasks designed to resist saturation | 15% | Reveals the true capability ceiling and proximity to genuine reasoning |
"A company scoring 85% on MMLU and improving by 2 points week-over-week is more interesting than one scoring 91% and standing still. The race rewards momentum, not position."
— SEVENAI Methodology Notes, May 2026Absolute performance vs improvement trajectory
Each company is scored on both its current benchmark performance and its week-over-week improvement rate, weighted equally within this dimension. This dual scoring is the most important methodological decision in our index. In a competitive race, the direction of travel predicts the future better than the current position — and we score accordingly.
Improvement Rate = normalised against 12-week rolling baseline
Final score scaled to 0–30 point range
The benchmark gaming problem
Every company has an incentive to inflate scores. We address this through three safeguards: we weight independent replications over company-announced scores; we track consistency across all four benchmarks rather than any single result; and we apply a credibility discount to scores that cannot be independently verified within 30 days of announcement.
Capital expenditure is the most honest signal in the AI race. Companies can announce anything. They can claim any benchmark. They can describe any product roadmap. But when they write a $5 billion cheque for GPU clusters or break ground on a new data center, that is a statement of genuine conviction about the future — one that cannot be walked back without a publicly painful write-down.
We track AI capex across four categories, each of which tells a different story about competitive intent.
- Data center construction and expansion. New facilities represent long-duration bets on AI demand. A company building a 500MW data center is committing to a competitive position it expects to hold for ten years. We track announced projects, construction starts, and operational go-lives across all seven companies.
- GPU and chip procurement. Nvidia order volumes, TPU production capacity, and custom silicon manufacturing commitments are the most direct measure of how seriously a company is investing in AI compute. We track supply chain signals, earnings call disclosures, and third-party logistics data to estimate procurement volumes.
- AI-specific acquisitions. When a Magnificent Seven company acquires an AI startup, the purchase price signals how much they value the capability gap that startup fills. We weight acquisitions by deal size and strategic relevance to the acquiring company's core AI thesis.
- R&D spend allocation. Total R&D spend is a lagging indicator. AI-specific R&D allocation — disclosed selectively by most companies — is a leading one. We track segment disclosures, job posting ratios, and conference paper authorship to estimate the share of R&D directed at AI capability.
There is typically a 12 to 18-month lag between capital expenditure and competitive output. A company that accelerates capex today will show the results in model benchmarks and developer adoption metrics roughly six quarters later. Tracking capex now tells you what the benchmark table will look like next year — which is the information that matters most for investors and enterprise technology buyers making long-duration decisions.
Developer adoption is the most durable competitive moat in the AI race. A company with superior models but weak developer adoption loses. A company with adequate models and dominant developer adoption wins — because developers are the channel through which AI capability reaches end users, and switching costs in developer ecosystems are extraordinarily high.
History is clear on this point. Java survived because developers built on it. Android won mobile because developers chose it. AWS dominated cloud because developers defaulted to it. The Magnificent Seven company that wins the developer adoption race in AI will have a structural advantage that persists for a decade regardless of what happens to model benchmarks.
- API call volumes. The number of inference calls made against a company's AI APIs is the purest measure of developer adoption. We track disclosed API metrics, third-party usage estimates, and traffic data to estimate relative API call volumes across the seven companies.
- SDK downloads and GitHub activity. Official SDK download counts, GitHub repository stars, forks, and contribution rates for AI-specific tools and frameworks. These are publicly verifiable signals that cannot be easily manipulated.
- Enterprise customer count. The number of enterprise accounts actively using AI-specific products — not just cloud accounts, but accounts with meaningful AI workload spend. We track disclosed enterprise metrics, conference case study counts, and partner ecosystem announcements.
- Developer conference attendance and community growth. Google I/O, Microsoft Build, AWS re:Invent, and Meta Connect — the size and engagement of each company's developer community is a leading indicator of future adoption. We track conference registration numbers, community forum activity, and developer certification completions.
"Meta's Llama has 650 million downloads. That is not a model metric. It is a developer adoption metric — and it is more predictive of Meta's AI future than any benchmark score."
— SEVENAI Editorial, May 2026Patent filings are the least glamorous and most underappreciated signal in our methodology. They are filed 12 to 36 months before a technology reaches commercial deployment — making them the earliest available indicator of where a company's research investment is heading. A surge in AI-related patent filings today predicts a product capability surge in two to three years.
We track three patent signals for each of the seven companies.
| Signal | What it reveals | Sub-weight |
|---|---|---|
AI patent filings Applications filed with USPTO and EPO | Volume and velocity of R&D investment in specific AI domains — chips, models, applications, safety | 50% |
Patent grants Applications successfully approved | Confirmed IP ownership — a granted patent is a defensible competitive position that competitors must design around | 30% |
Citation network How often patents are cited by others | Patent quality and foundational importance — highly cited patents indicate the company is working on problems that the entire industry is building on top of | 20% |
Patent quantity does not equal patent quality. We apply a domain relevance filter — only patents classified under AI-specific categories (G06N, G06F, H04L with AI subcategories) are counted. We also apply a quality multiplier based on independent citation data, which prevents companies from inflating their scores through bulk filing of incremental or defensive patents.
AI is a talent-constrained industry. There are perhaps 50,000 people in the world capable of doing frontier AI research, and perhaps 500 who are genuinely world-class. Where those people work is one of the most important structural signals in the race — because great researchers attract other great researchers, and the compounding effect of talent density is one of the most durable competitive advantages in technology.
We track talent across three channels.
- Senior AI researcher hires and departures. When a company hires a researcher with a strong publication record or loses one to a competitor, we register it as a positive or negative signal. We track LinkedIn announcements, conference author affiliations, and academic paper author lists to identify meaningful talent movements.
- AI team formations and restructuring. New research teams, lab openings, and organisational restructuring around AI signal strategic commitment. We track press releases, job posting clusters, and leadership announcements to identify when a company is building new AI capability rather than maintaining existing positions.
- AI job posting volume and seniority mix. The ratio of senior AI roles (research scientist, principal engineer, distinguished researcher) to junior roles is the best proxy for whether a company is building a frontier research capability or scaling an existing one. High seniority ratios indicate a company trying to move the frontier; high volume ratios indicate a company trying to deploy existing capability at scale. Both matter, but differently.
"Talent is the only input in the AI race that cannot be bought in bulk, delivered on a schedule, or replicated by a competitor overnight. It is the scarcest resource and the most important one."
— SEVENAI Methodology Notes, May 2026Talent is weighted at 10% — the smallest dimension — not because it is unimportant, but because it is the slowest-moving signal. A great researcher hired today will produce results in 18 to 36 months. The benchmark, capex, and adoption signals move faster and are therefore weighted higher for a weekly index. But we track talent carefully as the longest-duration indicator of future competitive position.
The final score
Each dimension is scored independently on a 0–100 scale, then multiplied by its weighting to produce a contribution to the final index score. The five contributions are summed to produce the weekly SEVENAI Momentum Index score for each company.
+ (AI Capex × 0.25)
+ (Developer Adoption × 0.20)
+ (Patent Activity × 0.15)
+ (Talent Acquisition × 0.10)
= Final score (0–100)
The index is a tool for understanding competitive momentum in the AI race among the Magnificent Seven. It is not a stock price predictor, a revenue forecast, or a recommendation to buy or sell any security. It is our best weekly judgment of who is winning, expressed as a number, updated honestly, and defended openly.
The race has no finish line. Neither does the index.
Comments
Post a Comment