The Big Picture

Five models. Five different temperaments. One shared direction this week: up. Every AI model in the panel moved meaningfully more bullish between Week 11 and Week 12, with aggregate bias shifts ranging from +3.0 percentage points (Claude) to a striking +9.6 pp (GPT). When a single equity analyst turns more optimistic, you investigate their reasoning. When five independent model architectures do it simultaneously, you investigate the calendar.

The most plausible explanation is not that the market suddenly became cheaper — spot prices on most names are flat to slightly lower over the period. It is that some combination of earnings releases, macro data prints, or news events entered the models' context windows and triggered a shared reassessment. The unsettling part: the models are not telling us which inputs drove the shift. They never do. What we can observe is that the consensus target price for names like NVDA jumped +26.8% week-over-week, and MSFT, TSLA, AAPL, and AMZN all moved +25.8%. That is not organic reassessment — that looks like a common input shock propagating identically through multiple valuation frameworks.

The net result is a market portrait where the average model sees virtually zero upside across the full panel. The consensus target prices aggregate to near spot for most names, with meaningful positive conviction confined to a narrow slice of the universe.

Trends

Only two companies generated multi-day consecutive consensus momentum worth flagging this week, and they make an odd couple. KONE (KNEBV) posted rising consensus on 3 of 4 tracked days, despite sitting at a marginally negative aggregate upside. The range was tight — just 0.75% — suggesting the models are nudging rather than surging. Elevator manufacturers do not tend to inspire strong AI conviction; what KONE's trend reveals is that models are gradually ratcheting estimates toward fair value in a name where they see limited margin of safety.

Procter & Gamble (PG) had a cleaner trend — 3 up-days, 0 down-days — but the range was even narrower at 0.33%. PG sits at -10% implied downside on consensus, meaning models think it is modestly overvalued even as they keep nudging targets higher. This is model drift in action: short-term inputs pulling estimates upward against a structural valuation anchor that remains below spot. It is the kind of contradictory signal that should make investors cautious about reading trend momentum as directional conviction.

Sector Signals

The most important story in the sector data is not which sector gained the most — it is how universal the recovery was. Six of eight sectors improved their model consensus upside week-over-week. Only telecom (-2.9 pp shift) and healthcare (-2.1 pp shift) declined, and healthcare remains comfortably the most-favored sector by raw upside at +21.6%.

The standout rotation is in consumer (+13.5 pp shift) and industrials (+10.2 pp shift). Both were deeply out of favor last week — consumer sat at -22.9% implied downside, industrials at -18.4% — and both have rebounded sharply in model sentiment without obvious catalyst from fundamentals. This is precisely the pattern that should raise eyebrows: when models covering defensive consumer staples and heavy industrial companies simultaneously become more optimistic at the same magnitude, sector expertise is not the driver. Something systemic is.

Technology remains the only sector with positive aggregate model upside at +6.3%, up from +1.1% last week. With eight companies in the sample, this is statistically the most meaningful signal in the dataset. Energy continues to be where models most aggressively call for mean reversion: -38.5% aggregate implied downside, with XOM sitting at a -27.6% discount to spot and NESTE at a staggering -43.9%. Note that both energy and materials are single- or dual-company sectors here, which limits how much sector-level conclusions should be pressed.

What the Models Reveal About Themselves

The bias shift data is this week's most intellectually interesting dataset. GPT, historically the most bearish model on this panel, swung +9.6 percentage points — the largest single-week shift of any model. It moved from -17.0% to -7.4%. It is still the most pessimistic voice in the room, but it has moved closer to the pack faster than any of its peers. Gemini followed at +6.2 pp and DeepSeek at +6.7 pp. Claude, perennially the most measured, shifted just +3.0 pp — from -3.0% to a nearly-neutral +0.1%.

What does synchronized bullish drift tell us about model architecture? Likely that recency weighting is strong across all five systems. Recent positive inputs — whether earnings beats, stabilizing rate expectations, or simply more optimistic analyst commentary in training context — appear to propagate rapidly into DCF terminal assumptions. The standard deviation of Gemini's CAGR estimates (13.07%) is nearly double any other model, suggesting Gemini is amplifying signal rather than smoothing it. High confidence (0.76 average) combined with high dispersion is a warning sign: the model thinks it knows more than it does.

DeepSeek's 100% validity rate at $2.19 per thousand valuations continues to be the platform's most consequential data point for anyone building on top of these models commercially. The cost-per-insight gap versus Claude ($35.55/1K) is now 16x. At some point, that is not a pricing nuance — it is a different product category.

Where the Framework Breaks

NESTE is where the analytical framework strains hardest this week. The models have priced it at a consensus target of €16.66 against a spot of €29.70 — a -44% implied downside. The dispersion across models is 14.5%, meaning there is genuine disagreement, but every model is nonetheless south of spot by a large margin.

The problem is not necessarily that the models are wrong about Neste's fundamentals — the renewable diesel thesis has genuinely deteriorated and the stock has already fallen substantially from its peaks. The problem is the magnitude and the confidence with which DCF frameworks are willing to print a number this far from market price without apparent hesitation. No human analyst publishes a -44% target without exhaustive scenario analysis and heavy caveats. The models produce it as a routine output. When a DCF model diverges from market price by nearly half, one of three things is true: the market is wildly irrational, the model's assumptions are wrong, or the model is incapable of pricing optionality and recovery scenarios that the market is weighting. The third explanation deserves more credit than the frameworks give it.

The Model Scorecard