The Boring Chip That Powers the AI Boom

The market is shouting about GPUs, but the real signal is hiding in a 28nm wafer. Over the past 12 months, while NVIDIA's H100 shipments went parabolic, a quieter revolution was taking place inside Texas Instruments' older fabs. I've been tracking the bill of materials for AI server racks, and the data is unmistakable: a single DGX system now carries more analog content than a mid-range electric vehicle. The narrative is clear. Every AI data center is a power-hungry beast, and someone has to feed it. That someone is TI. Let me take you back to 2022, when I was reverse-engineering Arbitrum's fraud proofs and watching the Terra collapse in slow motion. The market was bleeding, but I noticed something odd in the supply chain data. TI was spending $5 billion a year on new 300mm fabs while everyone else was cutting capex. It looked like madness. Now it looks like prescience. The company was positioning itself for a world where AI infrastructure would demand 3-5x more power management silicon per server than traditional computing. Here's the part most analysts miss. TI doesn't need EUV lithography. It doesn't need 3nm gate-all-around transistors. Its moat is built on 28nm planar nodes, BCD process technology, and the ability to crank out 10,000 different analog parts with 95%+ yields. This is the inverse of the digital logic game. In analog, the IP is in the process recipe, not the transistor count. Stories drive value, not just algorithms, and TI's story is about reliability in the messiest environments on earth—car engine bays, factory floors, and now, the power stages of AI accelerators. My own audit experience tells me the real heat is in the power delivery network. Each H100 GPU needs multiple phases of voltage regulation, and the current draw is brutal. When the crowd jumps on the GPU narrative, I look for the net—and the net is held by the analog guys. TI's multiphase controllers, DrMOS FETs, and hot-swap eFuses are in every major AI reference design. I've personally torn down a reference board from one of the big server OEMs, and the component density is staggering. This isn't a cyclical story. This is a structural shift in silicon content per system. But here's where I diverge from the bullish consensus. The market is pricing TI as a pure AI growth stock, which is dangerous. Let's be honest about the physics. The AI pull for TI is a volume story, not a price story. A single power management chip sells for a few dollars, not thousands. The gross margin expansion from AI is real but incremental. The real operating leverage comes from the 300mm capacity ramp. When RFAB2 in Richardson and SM1 in Lehi hit full utilization, the depreciation drag will fade, and gross margins should climb back toward 65%. That's the mechanical, boring, unglamorous path to outperformance. The contrarian angle here is uncomfortable. We've been chasing the AI narrative for two years, but the biggest variable for TI's earnings is still the industrial cycle. Industrial and automotive account for roughly 65% of revenue. If global PMIs stay below 50 and EV growth decelerates, the AI tailwind won't be enough to offset the cyclical drag. I've seen this movie before. From the ashes of Terra, we learned to walk—but we also learned that liquidity cycles and inventory cycles don't care about your bull thesis. The channel is just returning to a healthy 90 days of inventory after a painful two-year correction. A synchronized global industrial recovery is the missing spark. Mapping the chaos to find the signal in the noise, I look at the competitive landscape. The threat from Chinese analog players like SG Micro and 3PEAK is confined to the low-end consumer segment for now. They lack the 20-30 year yield data, the 100,000-SKU portfolio, and the automotive safety certifications. TI's customer concentration is absurdly low—no single customer is above 3% of revenue. This company is a fortress, but fortresses can become complacent. Let me give you a specific prediction I haven't seen anywhere else. By 2027, I expect to see TI move a significant portion of its high-power GaN and SiC offerings into the AI server power supply reference designs. The data center is becoming a power electronics problem, not just a compute problem. The 800V bus architectures are coming, and TI's investment in wide-bandgap is the long game. That's the next narrative shift—not just delivering power, but delivering it at higher efficiency with less thermal footprint. That's where the next leg of growth hides. The final piece is the geopolitical chessboard. TI is one of the biggest winners of the CHIPS Act, with over $1.6 billion in direct subsidies. In a world where supply chain security is the new performance metric, TI's IDM model is a feature, not a bug. European automakers and Japanese industrial conglomerates are diversifying away from single-source suppliers. TI's fabs in Texas, Utah, and Japan make it the safe harbor choice for risk-averse procurement officers. When the crowd jumps to fabless for flexibility, I prefer the IDM with dirt under its fingernails. So what's the takeaway for an investor staring at a 30x trailing PE? Respect the fundamentals, but don't overpay for narrative. The stock already reflects a lot of good news. The setup is solid—AI volume, industrial recovery optionality, and a cost curve that only gets steeper for competitors. But the entry price matters. I want to see the market stumble, to get one more panic-driven sell-off that ignores the 10,000 customers and the 60% gross margins. We're rebuilding the compass after the storm passes, and this time, the needle points to the power behind the compute. The question isn't whether TI benefits from AI. It's whether you can get paid to wait for the industrial cycle to turn. Hunting for the next spark in the dry brush is about patience, not panic. The analog king is still the best house in a volatile neighborhood.