The Big Picture

The models turned meaningfully more optimistic this week — but the optimism is unevenly distributed and, in at least one case, suspiciously abrupt. Across all five models, average upside estimates range from a near-flat +0.4% (DeepSeek) to a comparatively ebullient +6.0% (GPT). The consensus target prices imply that most of the 23 tracked companies are either fairly valued or modestly overpriced at current levels. Of the full universe, only seven names carry positive upside consensus, and several of those — GOOGL at +0.5%, META at +0.5%, TIETO at +0.5% — are barely above the rounding threshold.

What this tells us is not that markets are efficient. It tells us that the models, collectively, are struggling to find compelling value in a universe that has already re-rated. The average cap rate — the share of valuations where the model's terminal value calculation hits a ceiling constraint — sits between 34% and 45% depending on the model. Nearly half of all valuations are being capped. That is a structural signal: these models are encountering companies whose implied growth trajectories strain the mathematical limits of DCF orthodoxy.

Trends

Nokia (NOKIA) is the sole entry in this week's trend table, and it earns its place with three rising consensus days out of four, within a 7.77% target-price range. For a stock that has spent years as a graveyard of analyst optimism, this is worth noting — not because the models have suddenly discovered a Nokia thesis, but because the consistency of upward revision across multiple sessions suggests something more than noise. The consensus target of €5.55 against a spot of €7.23 still implies -23% downside, so the trend is a softening of bearishness rather than a turn to conviction. But in model-behavior terms, three consecutive days of rising targets is a meaningful signal of recalibration.

Sector Signals

Technology leads all sectors in week-on-week improvement, with model upside shifting +8.3 points to reach +15.1% across eight companies. This is the largest single-week sectoral move in the current dataset. The caveat: technology also contains the widest dispersion of individual outcomes, and the sector's aggregate upside is heavily influenced by a handful of names with large target gaps. When GOOGL carries a consensus target of €433.77 against a spot of €280.92, the arithmetic flatters the sector average considerably.

Energy improved by +5.4 points but remains deeply negative at -33.1% implied upside — the models are not warming to energy so much as becoming slightly less cold. XOM at -29% and NESTE at -37% are structural drags. Healthcare is the only sector that deteriorated week-on-week, slipping -2.0 points to +20.0% — still the highest absolute upside reading of any sector, driven by ORNBV and JNJ (though JNJ's target price fell -12.4% this week, the sharpest single-name cut in the dataset).

Note that telecom contains only one company (ELISA) and materials only one (UPM). Single-company sectors should be read as individual stock signals, not genuine rotation.

What the Models Reveal About Themselves

The story of the week is GPT's +13.3-point bias shift — from -7.3% last week to +6.0% this week. No other model comes close: DeepSeek moved +4.6 points, Grok +5.1 points, Claude +3.1 points, Gemini a modest +1.3 points. GPT's swing is not a gentle drift; it is a discontinuity. When a model's average upside estimate moves by 13 percentage points in a single week without an obvious macro catalyst, the most likely explanation is sensitivity to input framing — small changes in how prompts are constructed, or how recent price data is presented, can produce outsized output swings in models that lack strong anchoring priors.

Gemini, by contrast, is the most stable this week. Its +1.3-point shift, combined with the highest average confidence score (0.73) and a CAGR standard deviation of 10.65% — the widest of any model — suggests a model that is confident in its directional calls but highly differentiated across names. High confidence plus high dispersion is an interesting combination: Gemini is not hedging toward the mean.

Claude's terminal growth rate stands out structurally. At 2.32% with a standard deviation of 0.46%, it is the only model that varies its terminal growth assumption across companies. Every other model locks in exactly 2.0% with zero deviation. Claude is doing something the others are not: adjusting long-run growth expectations to the specific company. Whether that produces better valuations is an empirical question, but it is unambiguously more sophisticated modeling behavior.

Where the Framework Breaks

NESTE is the most interesting failure case this week. Its consensus target of €17.98 against a spot of €28.36 implies -37% downside — the most bearish reading in the universe. Yet its target price *rose* +7.9% this week, the second-largest upward revision after META. Its dispersion score of 0.198 is the highest in the dataset by a wide margin, meaning the five models disagree more violently about Neste's fair value than about any other company tracked.

This is a framework stress test. Neste is a renewable fuels company caught between collapsing SAF margins, feedstock cost volatility, and a long-term energy transition narrative that may or may not materialize on any investable timeline. The models are not converging — they are diverging further. High dispersion plus a rising average target plus deeply negative consensus upside is a combination that defies clean interpretation. It suggests the models are not analyzing the same company; they are analyzing different versions of what Neste might become.

The Model Scorecard