Earnings & Pattern recognition

I'll take a stab here - it feels like you're relying on the 1st degree absolute results and hoping for a rule of thumb rather than interpreting why those results are important. 

Beating EPS consensus estimates by 5% in itself is sort of meaningless. For MMHFs primarily, my interpretation has been: deconstruct the biz into its main drivers (what drives the stock price), understand what the current stock price today says about those key value drivers, understand where those KPIs are likely to head in future results = is there going to be a result that will reinforce/disprove the key arguments one way or another (shift the market's perception on this name, either incrementally or significantly). Then do that for your entire coverage, force rank, hedge out low convictions, and exploit high convictions and/or skewed risk/rewards. 

EX: lets say I have a fresh spin off, and the business typically grows topline by 3-4%/yr and has EBIT margins of 15%. Mgmt. makes the argument that independently they can grow closer to 5%-6%/yr and EBIT margins should be closer to 18%-20%. Based on comps / DCF we know that companies like the former typically trade in the mid-teens P/E while the latter are high teens low 20s P/E (prob even more if we see a consistent LDD EPS growth trajectory over the next few years). Right now the stock price is embedding the former's metrics. Over time, you believe that the company can execute according to the plan and turn into the latter. 

This company can beat EPS by 5% but it was all driven by a one-time recurring contract, and margins were actually flat, which maybe suggests there is not much margin leverage here =/= not a great result. Or they can show minimal topline growth in the quarter with decent margin expansion still, and a solid guide, and even though the results weren't incredible, the market is ready to warm up to this story now with a little more evidence, and we get a meaningful re-rate.

Another example that comes to mind is ALGN right now - this company is trying to re-capture the growth metrics of its earlier years, but the competitiveness of the end market continues to evolve a bit, and there is more uncertainty re: TAM + durability of growth + at what level from here on out. As such, it is likely to swing off of a couple key metrics in earnings because it will have a meaningful impact on the market's expectations of what does topline growth look like for the next few years. 

So your question about just looking at consensus numbers is one step removed from interpreting what the results say about the businesses valuation longer term, and what is embedded in the stock price today. If anything, you can reverse engineer "consensus" by just looking at the stock price, the company's past track record ("normalized earnings" comes to mind), and peers. So as the other poster said, maybe some AI stocks came somewhat in-line with consensus or buyside expectations, but each result reinforced the trajectory for now (market is an extrapolation machine). My understanding is that its not always about a big hair variant perception to EPS estimates, and can even be just a marginal 2nd derivative improvement on consensus longs. 
 

Feel like I was redundant here but hopefully this helps.