Synthos Predictions · AI · Infrastructure layer

Is AI infrastructure priced in — or is there room to run?

Synthos Score: 83/100 — bullish, ▲ rising. The experts' case: power — not chips — is the binding constraint on the AI buildout, and demand still overwhelms global capacity to deliver compute. Distilled from 1,769 claims by 35 tracked voices, weighted by verified track record.

Synthos Research · synthosresearch.com · as of July 5, 2026 · updated weekly

⟳ This call last moved 12 days ago — see what changed 🔒

Synthos Score: 83
Feasibility: 84
Priced in: 35
Frontier Gap: +49
Disagreement: 15%

Synthos Score = precision-weighted expert conviction (bull-share of directional claims; each voice capped, feed-collapsed, recency-decayed). Feasibility = the science panel (bull-share of frontier-scientist claims). Priced-in v1 = trade-extension percentile of the power/datacenter basket (VRT, ETN, VST, CEG). The Frontier Gap = feasibility − priced-in — how much headroom the science gives the trade.

The dialectic — strongest case on each side

▲ The bull case

“Power generation, conversion, and cooling — not chips or model design — are the pacing constraints of AI; the grid takes years. Compute demand will never catch up to supply — smarter models perpetually need more compute.”

Jordi Visser · macro/AI strategist · Jan–Apr 2026 · tracked voice

▼ The bear case

“The ~$4T AI-IPO wave may mark a near-term peak in the capex trade — positioning is crowded and the trade likely needs 3–6 months of digestion.”

Jordi Visser (self-caution) + minority bear cohort · 15% of directional weight

Every claim traces to a dated source in the Synthos knowledge base. We show the disagreement — that’s the product, not a bug.

Prediction history — and when we changed our mind

Weekly recompute from launch; the line gets more valuable every month it exists.

Jul 05 2026 Prediction opened publicly at 83 (rising) — genesis entry.
Jun 2026 🔒 Conviction firmed on new hyperscaler capex + “time-to-compute” evidence.
Mar 2026 🔒 Bear-side weight rose briefly on IPO-digestion warnings, then faded.

What would change our mind

Capex guidance cuts from two or more hyperscalers in consecutive quarters — the demand signal breaking, not wobbling.
Idle compute appearing — GPU lease rates falling and utilization slipping (today the evidence says “no idle GPUs in the world”).
The power bottleneck resolving early — interconnect queues clearing / grid additions outpacing datacenter demand, removing the core scarcity thesis.

These falsifiers are logged with the call and graded in public. Honesty is the product.

Track this prediction.
One email the moment the Synthos Score moves — with the evidence that moved it. No noise, no spam, unsubscribe anytime.

So what — where this prediction points

A wide Gap on AI infrastructure favors the power & datacenter complex over the crowded chip trade — the picks-and-shovels behind the picks-and-shovels. Explore the evidence-backed surfaces:

★ Synthos One 🌳 Knowledge tree Deep dives (586) Macro regime