# MIT study finds AI can already replace 11.7% of U.S. workforce

---

MIT: Most AI useless. Also MIT: AI can already replace large part of workforce.

---

Listen up, regards. Ex-Googler here. Did my EE degree at MIT before spending years in Mountain View, so I’m going to try and explain this slowly so even you lot can understand why this TPU FOMO is absolute garbage. I am still bullish on GOOG in the long run and have quite a bit of my net worth tied up in the stock. HOWEVER: You guys are drooling over "TPU" like you actually know what an ASIC is. A Tensor Processing Unit is a scalpel; an Nvidia GPU is a Swiss Army Knife with a chainsaw attached. Here is the engineering reality: TPUs are specialized ASICs. They are decent at matrix math for specific internal workloads, but individually? They are slow as sh*t compared to a Hopper or Blackwell GPU. To match Nvidia’s raw throughput, Google has to daisy-chain thousands of these things together. It is not "efficient", but represents a massive hardware tax and latency headache that doesn't show up on the spec sheet. THE MOST IMPORTANT PART WHICH YOU FORGET is the Software Stack. Nvidia doesn't just sell chips but CUDA. The entire planet's AI infrastructure, from Robotics to AVs, is built on Nvidia’s stack. You don't just "switch" to TPUs. Porting a massive, production-level AV stack to run purely on Google’s custom silicon is an engineering nightmare. Nvidia has a moat wider than your wife’s boyfriend’s ego because of this software lock-in. Google buying Nvidia chips while making TPUs isn't a sign Nvidia is dying. It is a hedge. It is called circular supply chain management. No massive hyperscaler wants a single point of failure. Even if Nvidia loses 10% or 20% market share to these internal chips, the Total Addressable Market for compute is expanding so fast it doesn't matter. Nvidia loses a slice but the pie is getting 10x bigger. Stop FOMOing into GOOGL thinking they just killed Jensen. They didn't. They'll both do well but NVIDIA is still king. End of rant

---

Listen up, regards. Ex-Googler here. Did my EE degree at MIT before spending years in Mountain View, so I’m going to try and explain this slowly so even you lot can understand why this TPU FOMO is absolute garbage. I am still bullish on GOOG in the long run and have quite a bit of my net worth tied up in the stock. HOWEVER: You guys are drooling over "TPU" like you actually know what an ASIC is. A Tensor Processing Unit is a scalpel; an Nvidia GPU is a Swiss Army Knife with a chainsaw attached. Here is the engineering reality: TPUs are specialized ASICs. They are decent at matrix math for specific internal workloads, but individually? They are slow as sh*t compared to a Hopper or Blackwell GPU. To match Nvidia’s raw throughput, Google has to daisy-chain thousands of these things together. It is not "efficient", but represents a massive hardware tax and latency headache that doesn't show up on the spec sheet. THE MOST IMPORTANT PART WHICH YOU FORGET is the Software Stack. Nvidia doesn't just sell chips but CUDA. The entire planet's AI infrastructure, from Robotics to AVs, is built on Nvidia’s stack. You don't just "switch" to TPUs. Porting a massive, production-level AV stack to run purely on Google’s custom silicon is an engineering nightmare. Nvidia has a moat wider than your wife’s boyfriend’s ego because of this software lock-in. Google buying Nvidia chips while making TPUs isn't a sign Nvidia is dying. It is a hedge. It is called circular supply chain management. No massive hyperscaler wants a single point of failure. Even if Nvidia loses 10% or 20% market share to these internal chips, the Total Addressable Market for compute is expanding so fast it doesn't matter. Nvidia loses a slice but the pie is getting 10x bigger. Stop FOMOing into GOOGL thinking they just killed Jensen. They didn't. They'll both do well but NVIDIA is still king. End of rant

---