AI Signals — Week 11, Mar 09–13, 2026
- Every model thinks the market is overvalued — average upside across all five models is negative, ranging from GPT's brutal -17% verdict to Claude's relatively sanguine -3%, a 14-percentage-point gap that tells you more about model personality than market reality.
- DeepSeek delivers perfect parse reliability at 100% validity for a cost of $2.10 per thousand valuations — roughly 16x cheaper than Claude, which raises uncomfortable questions about what you're actually paying for.
- Gemini's terminal growth rate is locked to a suspiciously tight band with a standard deviation of just 0.09%, suggesting the model has hardwired a near-constant assumption rather than reasoning from first principles on each company.
- GPT is the only model to peg terminal growth at exactly 2.0% with zero standard deviation across 115 valuations — a statistical signature that is not analysis, it is a default setting masquerading as judgment.
The Big Picture
When five independent AI models are asked to value 23 companies across a full trading week and every single one of them arrives at a negative average upside, that is not noise. That is a signal — though perhaps not the one you expect.
The consensus target price sits below spot for the majority of names. AAPL trades at $255.76 against a model consensus of $177.26. TSLA at $395.01 versus $252.97. XOM at $153.53 with models pointing to $91.60. The models are not being mildly cautious; they are, in aggregate, calling a significant chunk of the US equity market expensive by traditional DCF standards.
The interesting question is whether this reflects genuine fundamental overvaluation or a structural limitation of DCF frameworks applied to companies where intangible compounding — brand moats, platform network effects, AI optionality — simply does not fit neatly into a terminal growth rate and a WACC. Probably both. But the consistency across models is striking enough to warrant attention, not dismissal.
Among the brighter spots: GOOGL and META each carry positive upside readings and saw target price upgrades of +11.1% and +12.4% respectively week-on-week. The models, it seems, are warming to the idea that AI infrastructure owners have pricing power the broader market hasn't fully priced in. NOKIA, by contrast, was savaged — its consensus target fell -25.1%, the sharpest single-name revision of the week.
What the Models Reveal About Themselves
The most revealing number this week is not an upside figure. It is GPT's terminal growth standard deviation: 0.0%. Exactly 2.0% for every single one of 110 valid valuations. This is not a model reasoning about long-run nominal GDP growth for each company — it is a template with a hard-coded field. A Finnish telecom operator and a US megacap cloud platform, treated identically on the most sensitive lever in a DCF. That lever, incidentally, explains roughly 40-60% of intrinsic value in most models.
Gemini is not entirely innocent here either. Its terminal growth standard deviation of 0.09% is so tight it borders on the same sin. The difference is that Gemini at least wobbles slightly, and it compensates with the highest confidence score in the cohort at 0.76 — though one has to ask whether confidence built on near-identical assumptions is confidence at all.
DeepSeek is the week's operational standout. Perfect validity, the second-lowest latency at 24.7 seconds, and a total cost of $0.24 for the entire week's run. At $2.10 per thousand valuations, it undercuts Claude by a factor of 16. Claude, meanwhile, produces the most generous market view at -3.0% average upside and the widest CAGR dispersion — which reads as genuine per-company reasoning rather than formula application. You may be paying for that nuance, or you may be paying for verbosity. Distinguishing between the two is the hard problem.
Grok posts the highest cap rate at 48.7% — nearly half of all valuations hitting the ceiling constraint — which suggests its raw DCF outputs are systematically aggressive and the framework is doing significant work to rein them in.
Where the Framework Breaks
The NVDA case is quietly devastating. The model consensus target price is $159.81 against a spot of $183.14 — a -13% implied downside — and dispersion across models is 0.0, meaning every model converges on roughly the same bearish number. Perfect agreement on a company whose fundamental story is, arguably, the hardest in the market to value right now.
This is where DCF frameworks structurally fail. NVIDIA's valuation is not a function of its current free cash flow discounted at a 9-something WACC. It is a function of whether AI compute demand remains insatiable for the next seven years, whether custom silicon from hyperscalers erodes its moat, and whether its software ecosystem creates the kind of lock-in that makes the gross margin durable. None of that fits in a terminal growth rate. The models' unanimous bearishness on NVDA is less a valuation call and more a confession that they are using the wrong tool for the job.
The Model Scorecard
| Model | Avg Upside | Cap Rate | Valid % | Confidence | Cost/1K |
|---|---|---|---|---|---|
| claude | -3.0% | 43.4% | 92.2% | 0.62 | $34.38 |
| deepseek | -11.1% | 40.0% | 100.0% | 0.68 | $2.10 |
| gemini | -5.6% | 37.4% | 100.0% | 0.76 | $10.17 |
| gpt | -17.0% | 40.9% | 95.7% | 0.66 | $15.53 |
| grok | -7.2% | 48.7% | 98.3% | 0.71 | $14.47 |