Top Tweets for #cloudedjudgement

In early 2023, I first introduced the 'AI Equation' to measure ROI. It's been gaining traction in investment circles. Reflecting on 2023, the thesis didn't unfold as expected. Here's my candid review and analysis: Does the thesis still stand? The 'AI Equation' Explained: 1⃣ GPU to Cloud Revenue: Thesis: $NVDA's data center revenue forecasted at $46B in 2023 and $80B in 2024 should translate to $20-40B in cloud AI revenue (14 months payback, see math below). Reality Check: The top 3 CSPs reported only <$3B in AI revenue in FY23. FY24 consensus is $230B, a $40B increase (+20%) from FY23. The AI rev is not reflected in the forecast. 2⃣ GPU to End Application: Thesis: Assuming a 20% ratio, an $80B GPU investment should yield $400B in 'AI rev'. Reality Check: Actual 'AI rev' is <$3B today (public co. & startups), indicating a nascent stage in AI's financial impact. Satya Nadella once humorously remarked, "If we trust the GPU forecasts, US GDP should 3x". Clearly, he's doing similar calculations! The question remains: Why hasn't cloud's AI revenue reached expected levels yet? 1⃣Time lag & scaling challenges for H100 clusters: It's challenging: ramping up GPU clusters has proven to be a bigger engineering challenge. $MSFT and $ORCL talked about the significant hurdles, impacting AI rev timelines. It’s time-consuming: Most were confused by H100 delivery vs. GA (General Availability) vs. 'live at scale'. CSPs today are still at 'GA with limited scale'. 2⃣ AI revenue definition, A100 vs. H100: A100 is a small portion: The rational question is "If H100 is slow to ramp, why isn't AI rev from A100 reflected in the numbers either?". Bc A100 represents only $5-10B of $NVDA's est. $46B data center revenue for FY23. This equates to <$2B in AI revenue for CSPs. A100 isn't incremental rev: Available since 2020, A100 doesn't contribute much to FY23 YoY growth. This might explain why $AMZN's IR only disclosed "AI rev $ is in line with $MSFT's", without committing to a YoY growth rate. 💡So where does it leave us on the "AI equation"? The thesis is still intact. There should be a lot more H100 live at scale in 1H24. The expected $30B in additional AI rev could double CSPs' growth to ~35%, significantly higher than the 20% consensus. This is a trillion-dollar question in the age of AI. I'm open to all the discussions and insights on the topic.