AI Signals — Week 42, Oct 12–16, 2026

The Big Picture

Five models. Five bearish shifts. Not one exception.

That kind of unanimity is worth pausing on. In any given week, you'd expect at least one model to drift bullish while others cool — idiosyncratic noise in the prompting, different sensitivity to the same earnings data, the usual. Instead, Week 42 delivered a clean sweep: claude slipped from +1.7% to +0.9%, gemini from +3.7% to +2.0%, deepseek from -1.2% to -2.0%, gpt from -8.1% to -10.0%, and grok from -5.2% to -6.6%. The ensemble's center of gravity has moved, and it moved in one direction.

The aggregate consensus across all 24 companies sits at a wide range: from +103% implied upside on TIETO to -56% on TSLA. That spread is not a sign of model confusion — it reflects genuine valuation polarization. The models collectively see a market where a handful of Finnish mid-caps look deeply cheap and several high-profile US mega-caps look dangerously expensive. Whether that view is correct is a question for the market. What it tells us about the models is that they are not anchoring to price momentum; they are anchoring to DCF fundamentals, and right now those fundamentals are telling a very different story than the tape.

Trends

The trends array is empty this week. No company sustained a multi-day consecutive move in consensus direction across the five-day window. That absence is itself informative: the models are not building conviction on individual names. Revisions are happening — see NOKIA's +16.5% target-price jump and ORNBV's -10.5% cut — but they are not cascading into sustained directional momentum. The models are adjusting and then going quiet, rather than doubling down.

Sector Signals

Healthcare's -9.3 percentage point shift is the week's most dramatic move, and it deserves scrutiny. With only two companies in the sector bucket, a single name's repricing can swing the aggregate hard — so this is not a broad sector call so much as a concentrated revision. Still, the direction matters: the models moved from a moderately bullish +19.1% implied upside to a barely-positive +9.8% in one week. Something in the underlying cash flow or discount rate assumptions changed materially for at least one of those names.

Materials — a single-company sector, which means the data point is UPM alone — deteriorated further to -22.3% from -16.8%. UPM also appears in the tp_changes list with a -4.5% pure estimate revision, confirming this is a genuine model downgrade, not a market-price artifact.

Telecom is the lone bright spot, nudging up +3.1 points to +31.9% implied upside. Again, single-company (the data maps to ELISA in the FI telecom slot), so read it as a stock-specific signal rather than a sector thesis. Industrials, financials, energy, consumer, and technology all drifted slightly negative — a broad, shallow deterioration rather than any sector-specific shock.

The technology sector's move from -0.3% to -1.9% is modest in absolute terms but covers eight companies, making it the most statistically meaningful shift of the week. The models are, on balance, incrementally more skeptical of technology valuations than they were seven days ago.

What the Models Reveal About Themselves

The inter-model spread is the story this week. GPT at -10.0% and Gemini at +2.0% are looking at the same 24 companies and arriving at conclusions separated by 12 percentage points. That gap is not explained by different cap rates alone — GPT's cap rate of 18.3% is the highest in the panel, meaning it is hitting the sanity ceiling more often and having those estimates pulled down. But cap rates alone don't explain a 12-point divergence in average upside.

The deeper signal is in the CAGR assumptions. GPT projects an average earnings CAGR of 4.9% with a tight standard deviation of 3.0 — it is a cautious, low-dispersion model that clusters its growth assumptions conservatively. Gemini, by contrast, runs at 7.5% CAGR with a 6.3 standard deviation, meaning it is willing to assign very high growth to some names and very low to others. Claude matches Gemini's 7.5% CAGR mean but with tighter dispersion (5.41), landing it in the middle of the pack on upside.

Deepseek warrants a structural note. Since the provider-side switch to reasoning mode on 2026-07-31, deepseek's median gap has migrated from roughly -8% toward ~0%. This week's -2.0% reading is consistent with that post-switch range — it is no longer the reflexively bearish outlier it once was, though it remains below the bullish models. Pre- and post-July 31 deepseek comparisons should be treated as spanning a series break; the model you are reading today is behaviorally different from the one in the historical archive.

All five models show zero capped estimates this week, confirming the v8 sanity guard is functioning as designed — wide enough to let genuine dispersion through, tight enough to catch absurdities.

Where the Framework Breaks

TSLA and BRK-B both show dispersion of exactly 0.0 — perfect model agreement. For Berkshire, that makes intuitive sense: a conglomerate with transparent book value and a long earnings history is the kind of name where DCF models converge. But Tesla at 0.0 dispersion with a -56.1% consensus downside is a different animal entirely. Tesla is a high-uncertainty, high-narrative stock where reasonable analysts disagree violently. The fact that five structurally different AI models have converged to identical bearish conclusions suggests they may be sharing a common blind spot — perhaps all trained on the same corpus of skeptical analyst notes, perhaps all applying similar terminal-growth penalties to capital-intensive EV manufacturers. When models agree perfectly on a controversial name, that is not confidence. That is a correlated error waiting to happen.

TIETO presents the mirror-image problem: +103% implied upside with 0.0 dispersion. The models unanimously see it as worth roughly double its current price of €18.64, yet the market disagrees by a factor of two. Either the models are missing something the market knows — liquidity, structural decline, a rights issue — or the market is missing something the models see. Zero dispersion at extreme upside is a flag, not a conviction signal.

The Model Scorecard