AI Signals — Week 40, Sep 28–Oct 02, 2026

The Big Picture

The week of September 28 presents a peculiar tableau: every sector in the coverage universe moved toward higher model-implied upside simultaneously. That kind of unanimity is not a signal of broad analytical insight — it is more likely a signal of shared inputs, shared macro assumptions, or a gentle drift in how the models are pricing risk. When everything gets a little better at once, nothing has actually been learned.

The aggregate picture is one of a deeply divided panel. Gemini sits at +4.5% average upside, Claude at +2.4%, while GPT anchors the bearish end at -7.2% — a spread of nearly 12 percentage points between the most and least optimistic models running identical data through the same valuation engine. That gap is not noise. It reflects genuinely different assumptions about growth, discount rates, and terminal value — and it should make any investor pause before treating "AI consensus" as a monolithic view.

The company-level data sharpens the picture further. At one extreme, TIETO trades at €18.36 against a model consensus target of €38.00 — an implied upside of 107% with zero dispersion. At the other, TSLA at $354.11 faces a consensus target of $158.23, implying -55.3% downside, also with zero dispersion. Two unanimous verdicts, at opposite poles of the valuation spectrum. The models are not uncertain here — they are collectively, confidently extreme.

Trends

No multi-day consecutive momentum trends were recorded this week. The absence of trending signals is itself informative: the models are not building conviction in any single direction on individual names. Estimates are moving, but not persistently. This is a week of recalibration, not momentum.

Sector Signals

The sector rotation data tells a story of uniform, modest improvement — and that uniformity is the story. Every one of the eight tracked sectors posted a positive shift in model-implied upside versus the prior week. The magnitudes are small: consumer led with +3.3pp, followed by materials at +2.0pp, with healthcare bringing up the rear at +1.0pp.

But interpret these shifts carefully. The sector_rotation figures blend market price movements with estimate revisions — if prices fell modestly across the board, upside gaps would mechanically widen even with no change in model targets. A week where markets drifted lower would produce exactly this pattern. The signal is real but ambiguous.

What is less ambiguous is the absolute level of conviction by sector. Healthcare sits at +32.0% implied upside across just two companies (ORNBV and JNJ — though JNJ is itself -11.5% underwater, suggesting the sector average is being pulled hard by Orion's extraordinary +83.9% gap). Energy remains the most structurally bearish sector at -26.6%, driven by XOM at -30.3% and FUM1V at -48.3%. The models have a consistent, cross-model view that energy assets are priced for a world the DCF framework does not recognize.

Note that materials (UPM alone), telecom (ELISA alone), and several other single-company sectors should be read as individual stock views, not genuine sector signals. A sector of one is a stock call wearing a sector label.

What the Models Reveal About Themselves

The most interesting behavioral story this week is Grok's sudden lurch toward optimism. Its bias shifted +1.8pp — the largest single-week move in the panel — taking it from -3.5% to -1.7%. Gemini moved +1.7pp, from +2.8% to +4.5%. These are not trivial adjustments. Together they suggest that two of the five models responded to something in the week's inputs — macro data, earnings revisions, or prompt-level context — in a way that GPT and Claude largely did not.

GPT barely moved (+0.3pp), remaining the most persistently bearish model at -7.2% average upside. Its confidence score of 0.59 is also the lowest in the panel — GPT is not just bearish, it is uncertain about its bearishness. That combination is analytically interesting: a model that sees downside but hedges its own conviction.

Claude remains the most expensive model by a significant margin at $45.59 per 1,000 valuations — roughly 18x the cost of DeepSeek at $2.56. Its average latency of 33.8 seconds per valuation is also an outlier; every other model completes in under 13 seconds. Whether that additional compute translates into superior insight is a question this data cannot fully answer, but Claude's +2.4% upside sits comfortably in the middle of the pack — not obviously better calibrated than Gemini's +4.5% at less than a quarter of the cost.

DeepSeek deserves a standing note. Following the provider-side switch to a reasoning-mode variant on July 31, its median gap moved from roughly -8% toward near-zero. This week's -0.5% is consistent with that post-switch behavior. Comparisons of DeepSeek's current stance to pre-July 31 readings should be treated as spanning a series break — the model name is the same, but the underlying inference behavior is not.

Where the Framework Breaks

The zero-dispersion cases are where the framework reveals its limits most starkly. TIETO at +107% upside with dispersion of 0.0, and TSLA at -55.3% with dispersion of 0.0, and BRK-B at +3.8% with dispersion of 0.0 — these are not cases where the models have converged through independent analysis. They are cases where the models have converged to the same answer for the same structural reason, likely because the inputs and valuation mechanics leave little room for differentiation.

For TSLA, the verdict is particularly striking. A -55.3% implied downside with zero disagreement means every model, every day this week, produced a target price clustered tightly around $158. At a spot price of $354, that is not a valuation gap — it is a chasm. The framework is essentially saying: at current prices, no defensible DCF produces a buy case. That may be correct. But zero dispersion on a name this complex and this narratively contested should trigger skepticism about whether the models are genuinely analyzing Tesla or simply anchoring to the same set of consensus inputs and arriving at the same mechanical output.

The Model Scorecard