AI Signals — Week 46, Nov 09–13, 2026
- Every single model turned more bearish this week — a rare unanimous drift that suggests the models are collectively repricing something structural, not reacting to noise.
- TSLA sits at -58.5% implied downside and AAPL at -42.1%, yet both carry zero dispersion — meaning the models agree completely on their bearishness, which is arguably more alarming than disagreement.
- TIETO commands a +102% consensus upside at zero dispersion too, creating a mirror-image puzzle: unanimous conviction on both the most loved and most hated names.
- GPT is now the most bearish model at -5.5% average upside with the tightest CAGR range of any model, suggesting it has converged on a cautious, low-variance worldview.
- Financials lost 4.4 percentage points of model favor in a single week — the sharpest sector deterioration in this dataset — while materials quietly recovered the most.
The Big Picture
Five models. Five bias shifts. All negative. That is the headline this week, and it deserves to be read slowly.
When a single model drifts bearish, you look for an idiosyncratic explanation — a prompt sensitivity, a data artifact, a provider-side quirk. When all five move in the same direction simultaneously, the explanation has to be systemic. The models are not talking to each other. They share no memory across sessions. Yet claude dropped 1.2pp, deepseek dropped 1.6pp, grok dropped 1.2pp, gemini dropped 0.8pp, and gpt dropped 0.7pp. The only common input is the underlying company data — earnings revisions, price moves, macro signals embedded in the financials. Whatever is in that data, all five models read it the same way this week.
The aggregate picture is one of a market that AI valuation models now view with meaningful skepticism. Of the 24 companies in the universe, the consensus target price implies downside for a majority. The average upside across models ranges from -5.5% (gpt) to +6.6% (gemini) — a spread that itself tells a story about how differently these systems are calibrated, even when they agree on direction.
Sector Signals
The most consequential move this week was in financials, where the model-implied upside fell from -6.8% to -11.2%, a shift of -4.4pp. With four companies in the sector — including JPM at -11.7% implied downside — this is not a single-stock distortion. The models are collectively reassessing what banks and insurers are worth at current prices. Note that SAMPO sits at -42.5% implied downside with a target price revision of -5.6% this week; that is a model genuinely moving its estimate, not just a price effect.
Materials was the week's quiet winner, recovering +5.6pp — though with only one company (UPM, which also received a +5.2% target price upgrade), this is a single-stock signal dressed up as a sector trend. Treat it accordingly.
Technology continued to gain favor, now at +10.7% average implied upside versus +8.0% last week. The sector carries eight companies and genuine breadth, making this the most credible positive signal in the data. GOOGL at +47.6% and MSFT receiving a +4.5% target price revision are doing real work here. The contrast with NVDA at -20.1% and AAPL at -42.1% within the same sector label is a reminder that "technology" is not a monolith — it is a container for wildly divergent model views.
Consumer deteriorated further to -23.3%, with PG at -17.9% and JNJ classified alongside it. The models see consumer-facing defensives as expensive, full stop.
What the Models Reveal About Themselves
The most structurally interesting data point this week is gpt's CAGR standard deviation of just 2.77 — less than half of gemini's 6.3 and well below claude's 5.22. A tight CAGR distribution means gpt is applying similar growth assumptions across very different businesses. That is not necessarily wrong, but it is a form of intellectual compression: the model is not differentiating aggressively between a mature telco and a hypergrowth platform. Combined with its 21% cap rate — the highest of any model — gpt is systematically discounting terminal values more heavily than its peers. The result is a model that is structurally bearish not because it is pessimistic about near-term earnings, but because it trusts the future less.
Gemini, by contrast, has the widest CAGR dispersion (6.3) and the highest average upside (+6.6%). It is doing more differentiation, rewarding growth stories more generously. Its 13.2% cap rate is the second-lowest. Gemini is the optimist in the room, and it is optimistic in a considered way — through higher growth assumptions, not lower discount rates.
Claude remains the most expensive model at $50.73 per 1,000 valuations — roughly 17x the cost of gpt and 7x gemini. Its latency of 39.9 seconds per valuation is an order of magnitude slower than gpt's 3.6 seconds. Whether that cost buys meaningfully better signal is the central question this platform exists to answer.
Where the Framework Breaks
ORNBV (Orion Oyj B) is the week's most instructive anomaly. Its consensus target price equals its spot price exactly — €103.10 — producing 0.0% implied upside and a null dispersion value. Then it receives the largest target price revision in the dataset at -24.6%. These two facts cannot both be true in a stable system. What almost certainly happened: the models revised sharply downward this week, but the consensus target — being an average across the week's sessions — landed precisely at spot by arithmetic coincidence, masking the directional violence underneath. The -24.6% revision is the real signal. The 0.0% upside is a statistical artifact. When a framework produces a number that looks like equilibrium but is actually the average of chaos, it is worth pausing before trusting the output.
TSLA and BRK-B both show 0.0 dispersion — but for opposite reasons. Tesla at -58.5% with zero dispersion means every model, every day, agreed it is dramatically overvalued. Berkshire at -8.6% with zero dispersion suggests similar unanimity on modest overvaluation. Zero dispersion is not always a sign of confidence; sometimes it is a sign that the models have all anchored to the same prior and stopped exploring.
The Model Scorecard
| Model | Avg Upside | Bias Shift | Cap Rate | Valid % | Cost/1K |
|---|---|---|---|---|---|
| claude | +4.9% | -1.2pp | 12.7% | 98.3% | $50.73 |
| deepseek | -1.1% | -1.6pp | 16.0% | 99.2% | $2.99 |
| gemini | +6.6% | -0.8pp | 13.2% | 95.0% | $25.98 |
| gpt | -5.5% | -0.7pp | 21.0% | 99.2% | $7.16 |
| grok | -1.9% | -1.2pp | 16.8% | 99.2% | $8.49 |