$1.86. That was AMD's stock price in mid-2015. What happened between those two data points is one of the most dramatic turnarounds in technology history — and it wasn't luck. She bet everything on a single CPU architecture called Zen, outsourced manufacturing to TSMC, and told Wall Street to be patient. AMD doesn't make chips. It designs them — obsessively, expensively, brilliantly — and then hands the blueprints to TSMC in Taiwan, which does the actual manufacturing on the most advanced production lines on Earth. It's also why AMD's fate is partially in someone else's hands, but we'll get to that. The money comes from four places, and the mix has shifted dramatically in just three years. This is the crown jewel now. Pensando data processing units handle networking offload. Three years ago, this segment was half its current size. Semi-custom APUs power every PlayStation 5 and Xbox Series console sold worldwide. The console contracts provide predictable multi-year revenue but carry thinner margins than enterprise products. This is the Xilinx inheritance — FPGAs, Versal adaptive SoCs, Alveo accelerators. These go into telecom base stations, fighter jet avionics, automotive ADAS systems, medical imaging equipment, and industrial automation. The margins are excellent. The downside is cyclicality: telecom spending collapsed in 2023-2024, dragging this segment down before it recovers. The unusual aspect of AMD's economics is the margin trajectory. Gross margins have climbed toward 52-54% as the revenue mix tilts from low-margin console chips toward high-value data center products. The FY2025 results benefited from an AI infrastructure spending boom. Whether that spending level is sustainable is a question AMD can't answer alone. It does not manufacture any of them. The capital that doesn't go into factories goes into design engineering. It's Amazon. Amazon is doing something different. Every chip Amazon designs internally is a chip it doesn't buy from AMD. And Amazon is AMD's single largest customer category. Meta designs custom inference silicon. AMD can't sue them into buying EPYC. It can't lock them in with proprietary software the way NVIDIA does with CUDA. Now, Intel. The oldest rivalry in semiconductors — 55 years of it. Intel still ships more total server CPUs than AMD in absolute volume. It still has deeper enterprise relationships built over decades. EPYC went from near-zero server share in 2017 to an estimated 30-35% of x86 server shipments by 2025. If they do, AMD's share gains plateau. If they don't, AMD pushes toward 40-45% and the x86 server market effectively becomes a duopoly where AMD is the premium choice. My judgment: Intel recovers partially but not fully. AMD keeps gaining, just more slowly. Then there's NVIDIA in AI accelerators. AMD's pitch here is honest but limited: "You need a second supplier, and we're the only credible one." That's not a claim of superiority. It's a claim of necessity. NVIDIA's hardware is better today. NVIDIA's software network is vastly deeper. AMD exists in AI because the market structure demands an alternative, not because AMD has earned dominance through technical superiority. Where AMD wins decisively: platform breadth. That matters for customers managing complex infrastructure who want fewer supplier relationships. The fabless model shapes the financial profile in fundamental ways. Every major AI framework was improved for CUDA first. Every university teaches CUDA. Every enterprise AI team has pipelines built on CUDA libraries. AMD cannot manufacture a single advanced chip without TSMC. Not one. The CoWoS advanced packaging bottleneck in 2023-2024 already demonstrated this — AMD couldn't get enough AI accelerators built fast enough because packaging capacity was constrained. The third issue is regulatory. China represents enormous AI chip demand, and AMD is legally prohibited from serving much of it. That's a permanent addressable-market reduction that no amount of product innovation can fix. Intel can't do GPUs or FPGAs at AMD's level. NVIDIA can't do CPUs. Qualcomm can't do servers. Xilinx couldn't do any of it without AMD's distribution and platform integration. But breadth alone isn't a defense. That's not a marketing trick. Then there's the TSMC relationship. Every dollar of R&D goes into design, architecture, and software rather than keeping a factory running. Intel bears that factory burden. AMD doesn't. AMD now has this validation at every major cloud provider. Nobody currently has all six. The dominant wager is AI infrastructure. The AI play has three layers. AMD's accelerators compete on memory capacity and capacity — the MI300X offers 192GB of HBM3, which matters for large language models that need to fit in GPU memory. Second, software: ROCm needs to reach the point where enterprises can deploy AMD hardware without rewriting their CUDA-based pipelines. The supporting bets are simpler. EPYC keeps gaining server CPU share — AMD went from near-zero in 2017 to an estimated mid-30s percentage of x86 server shipments. Ryzen AI targets the emerging AI PC category where on-device inference creates upgrade demand. The Xilinx portfolio serves long-cycle embedded markets that provide margin stability when consumer segments get choppy. That's the metric that tells you whether the AI bet is working or whether AMD remains primarily a CPU success story with AI aspirations. The CPU side is nearly settled. The irony is, None of that is uncertain enough to lose sleep over. That's the irony Lisa Su has to solve. Santa Clara, 1969. The founding thesis was simple: the semiconductor industry needed a second-source supplier for Intel's chips, and someone technically capable should provide it. For its first two decades, AMD operated largely in Intel's shadow, manufacturing compatible versions of x86 processors under licensing agreements that gave Intel legal cover for market dominance claims while giving AMD revenue. The ATI Technologies acquisition in 2006 brought graphics processing capabilities that would prove essential two decades later when GPUs became the computational substrate for machine learning. At the time, it looked like an expensive bet on gaming. In retrospect, it positioned AMD to compete in AI compute before AI compute was a market category. AMD sold its Austin campus. It laid off thousands of engineers. What remained was a pure design firm with a single viable architectural bet — Zen — that Lisa Su and her engineering team had to execute flawlessly. If AMD's software stack crosses that line — call it the point where a Fortune 500 AI team can deploy Instinct accelerators without hiring dedicated porting engineers — then data center GPU revenue doubles by 2028 and AMD becomes a $50-60 billion revenue company. EPYC owns 30-35% of x86 server shipments and Intel would need three consecutive flawless generations to reverse that — something Intel hasn't managed since Haswell. This is two very different businesses wearing the same label. When those companies increase capital spending, AMD's numbers look spectacular. The company designs CPUs, GPUs, and adaptive computing products for data centers, personal computers, gaming consoles, and embedded systems. The company that should worry Lisa Su most isn't NVIDIA. But Intel has been executing poorly since roughly 2015, and AMD exploited every stumble. The question is whether Intel's new leadership can ship competitive products on a modern process node. That's a viable position — it generates billions in revenue — but it's fragile in a way that the CPU business isn't. No other company ships x86 CPUs, discrete GPUs, AI accelerators, FPGAs, and data processing units from a single vendor. The competitive position is the strongest it's been since the Athlon 64 era. Let me be direct about what keeps AMD's leadership up at night: CUDA. The embedded business recovers as telecom spending normalizes. The near-death years of 2012 through 2016 forced choices that determined the modern company. It spun off its manufacturing operations as GlobalFoundries.