The Big Picture

Five models, 23 companies, five trading days. The aggregate signal from 575 individual valuations this week is unmistakably cautious. Three of the five models are net bearish — average upside below zero — and the two that remain technically bullish, GPT at +1.1% and Gemini at +0.9%, are barely above the waterline. The ensemble consensus across the full universe sits close to flat, with the median company trading within a few percentage points of its AI-derived fair value.

What makes this interesting is not the level but the direction of travel. Every model except Gemini grew more pessimistic week-on-week. That synchronized drift toward caution is not noise — it reflects something the models are collectively absorbing from the same earnings releases, macro data, and market prices. When five architecturally distinct systems all lean the same way, the signal deserves respect even if the magnitude is modest.

The cap rate story is equally telling. DeepSeek and Grok both cap 47% of their terminal values — nearly half their outputs hit the ceiling before they can express their full bearishness. Claude, by contrast, caps only 35.7%. This means Claude's -0.4% average upside is a relatively unfiltered view, while DeepSeek's -5.7% is the visible portion of a potentially more severe underlying pessimism. The cap is a compression artifact, and ignoring it leads to systematic misreading of model conviction.

Trends

Three names show multi-day directional momentum in model consensus this week. Neste Oyj is the standout: four consecutive rising days, zero down days, a +15.3% target price revision over the week, and yet the stock still sits -18% below consensus. The models are aggressively upgrading their view of Neste even as the stock remains deeply underwater on their own estimates. That combination — rising conviction plus persistent undervaluation — is exactly the pattern worth watching.

Johnson & Johnson shows a quieter version of the same dynamic: three up days, one down, a modest 1.83% range. The models are incrementally warming to JNJ without drama. Metso Oyj rounds out the trio with three clean up days and the tightest range of the three at 1.02% — a slow, steady rehabilitation of a name that remains below consensus target.

None of these trends are explosive. But in a week where the dominant direction was pessimism, finding three names where model conviction is consistently rising is a meaningful counterpoint.

Sector Signals

The sector rotation picture this week is a study in divergence. Industrials improved the most, gaining +3.6 percentage points of model favor to reach -4.7% average upside — still negative, but meaningfully less so than last week's -8.3%. Energy also recovered ground, shifting +2.6pp to -20.9%. Both moves suggest the models are partially unwinding prior pessimism on cyclicals, though both sectors remain deeply discounted in absolute terms.

The deterioration is concentrated in Materials (-4.7pp shift, now at -20.7% upside) and Technology (-2.5pp, falling from +4.7% to +2.2%). The technology slide is the more consequential signal: this is the largest sector by company count (8 names) and the only one that has been consistently net positive. Watching it erode toward neutral is the week's most important sector story.

Healthcare remains the models' favorite sector at +20.5% average upside, though even here conviction slipped -2.0pp. Note that single-company sectors — Materials is just UPM, Telecom is just Elisa — should be read as individual stock calls, not genuine sector signals.

What the Models Reveal About Themselves

The model evolution data this week is a masterclass in behavioral divergence. GPT's bias collapsed from +4.5% to +1.1% — a -3.4pp swing that is the largest single-week shift in the cohort. GPT was the most bullish model last week; it is now the joint-most bullish alongside Gemini, but barely. Something in this week's inputs — likely earnings data from the large US tech names — caused GPT to substantially revise its optimism downward.

Gemini did the opposite. It moved from -1.4% to +0.9%, a +2.3pp improvement, making it the only model to turn net bullish this week. Gemini also carries the highest average confidence score at 0.77 and the widest CAGR standard deviation at 11.29% — it is simultaneously the most self-assured and the most dispersed in its forecasts. High confidence with high dispersion is a combination that should give investors pause. It suggests Gemini is not hedging; it is making bold, differentiated calls and expressing certainty about them.

Claude's shift from +0.2% to -0.4% is numerically small but directionally significant — it crossed the zero line into net bearish territory. Claude also generates the most tokens by a wide margin (933,151 versus DeepSeek's 716,718) and costs 19x more per thousand valuations. The extra verbosity does not appear to translate into meaningfully different conclusions at the aggregate level, though it may matter at the individual stock level.

Where the Framework Breaks

NVIDIA is the cleanest illustration of where DCF-based AI valuation hits its limits. The consensus target price is $203.22 against a spot of $211.50 — a -3.9% implied downside — with a dispersion score of exactly 0.000. Every model agrees, to the decimal, on the same answer. That unanimity is not a sign of analytical clarity; it is a sign that all five models are anchoring to the same set of consensus inputs and applying near-identical discount rate assumptions. When dispersion is zero, you are not getting five independent opinions. You are getting one opinion, repeated five times.

The same phenomenon appears in TSLA (dispersion 0.000) and XOM (0.000). These are not the simplest businesses to value — they are among the most contested. Zero dispersion on Tesla is, frankly, an anomaly that should trigger skepticism about the independence of model outputs on high-profile, heavily-covered names.

The Model Scorecard