#亚马逊巨额融资与AI投资

JUST IN: Amazon beats Q2 earnings expectations ▸ EPS: $5.75 vs $1.82 expected ▸ Revenue: $200.6B vs $196.5B expected $AMZN +7.9% after hours https://t.co/bEhzHouehv

Texas and the Midwest are about to capture more than half of all of the new US AI infrastructure development: Amazon, Google, Meta, and Microsoft are expected to spend nearly $700B combined in 2026 on AI infrastructure, up more than 60% from 2025. Where this money is going exactly is to AI data center real estate that houses the specific infrastructure needed to train, deploy, and deliver AI apps and services. This requires 3 things: - A dedicated grid connection of at least 100 megawatts (training requires more power). - A land parcel of 100 acres or more. - A cool climate that reduces the cost of cooling (chips heat up often). Established markets like Northern Virginia and Silicon Valley have run out of available power, and that is pushing new construction inland into places like Texas. The thing that’s going to determine how suitable a place is for AI infrastructure is going to be based on these 6 things: - Power infrastructure (30%) - Land availability (20%) - Network connectivity (15%) - Labor market (15%) - Business environment (10%) - Climate and resources (10%) And based on these markers, here’s what the top markets look like in the US on a score range of 2.9-4.2:

Amazon runs hundreds of millions of forecasts every day to move packages. Uber makes pricing decisions every few minutes across millions of rides. Banks score billions of card transactions every day for fraud. All of this runs on structured data, not language. That's the step change we set out to bring with Synthefy. Today, we're releasing Synthefy Nori, our first tabular foundation model. For over a decade, structured data prediction meant the same grind: train an XGBoost or LightGBM model, tune it, ship it, watch it drift, retrain. Synthefy Nori ends it. Pass your labeled rows as context, get predictions back in seconds. No training. No tuning. Synthefy Nori Highlights: > #1 on aggregate across 96 datasets, ahead of TabPFN-3 > Beats tuned XGBoost & LightGBM on ~80% benchmark datasets, with zero training or tuning > 6M params, ~1/10 the size of its peers > Fully open source See next tweet for white paper, Github, Hugging Face

I gave my @NousResearch Hermes agent $100,000 last night and this morning it went long Amazon stock and calls AMZN +7% after hours It’s executing against my long running AI thesis with a ton of context, connections and data to support it https://t.co/ADTirjjCDU