Raymond James just slapped a Strong Buy rating on AMD with a headline that reads like a coronation: a clear path to challenging Intel's CPU dominance. The stock got its pop. The narrative got its wings. But I've spent enough time on both sides of the trade to know a borrowed advantage when I see one.
AMD doesn't own a fab. It owns a relationship with TSMC. And that relationship is the entire thesis โ the bull case, the bear case, and every trade in between. We didn't see the TSMC capacity squeeze coming until the AI gold rush made it impossible to ignore. When NVIDIA and Apple are fighting for the same 3nm slots, AMD's so-called "process advantage" becomes a queue position. That's not analysis. That's a coin flip.
Here's what the upgrade report gets right, and what it conveniently leaves out. I've stress-tested enough protocols in my time โ including a three-week audit of an AMM that nearly blew up on a reentrancy bug โ to know that the real risk is never the thing everyone's staring at. It's the dependency hiding underneath.
The Data Stack
Let's lay the actual numbers on the table. AMD's server CPU share sits at roughly 25%. Five years ago it was 5%. The flywheel is real: EPYC genuinely wins on power-per-watt, and the hyperscalers โ Microsoft, Google, Amazon, Meta โ keep rotating more of their fleet to AMD. The margin structure confirms it. AMD's gross margins run 52-55%; Intel is down to 40-42% and bleeding further. Intel's foundry business is a cash furnace. Capital expenditures eat 30-35% of revenue โ $250 billion across Ohio, Arizona, New Mexico, Oregon. AMD spends 5-8% of revenue on capex. That's the structural gap between owning factories and renting someone else's.
But here's where the Strong Buy thesis gets fragile.
The Single Point of Failure
AMD is fabless. Its entire process edge is TSMC's manufacturing maturity. That means AMD's competitive moat is TSMC's capacity allocation โ a decision AMD doesn't control. Right now, TSMC's 3nm and 5nm lines are running at over 90% utilization. Who's first in line? NVIDIA, which owns the AI training market with 80% share. Apple, the most loyal customer in the business. AMD? Third place, hoping for leftovers.
This is the same trap I watched in the 2020 DeFi summer. Projects flashed APYs that made eyes water, and I saw the TVL climb โ the liquidity was subsidized. The moment the incentive programs stopped, the users evaporated like a rug pull in slow motion. AMD's process advantage is a similar subsidy. It's borrowed from TSMC's execution. The market prices the share gains, but it doesn't price the queue position. If TSMC has to choose between feeding NVIDIA's AI appetite and AMD's EPYC ramp, the choice is obvious.
I ran a stress test on a protocol once โ the bonding curve looked airtight until I simulated a flash loan attack on the withdrawal function. We patched it before mainnet, saving $15 million in TVL. The lesson stuck: trustless code requires rigorous testing, not faith. The same applies to a fabless chip company. The trust is in TSMC's capacity allocation, and that's not a contract โ it's a favor.
The Intel Wildcard
Here's the counter-intuitive part the report barely touches. Intel's 18A โ the 1.8nm-class node โ is slated for production in the second half of 2025. Intel is the first customer for ASML's High-NA EUV machines. If 18A yields climb above 80%, the process gap between Intel and TSMC closes to zero. AMD's window of process superiority shrinks from a comfortable two-node lead to a one-cycle advantage. The market has priced Intel as a legacy laggard. The upgrade thesis quietly assumes Intel 18A will stumble โ but the delivery record on Intel's new roadmap is a coin flip.
Early yield reports put Intel 18A at 60-70% in late 2024. That's not great. But Intel has a government-sponsored balance sheet that AMD doesn't. The CHIPS Act is pumping $85 billion in direct subsidies and another $11 billion in loans into Intel's foundry expansion. The depreciation drag will hit Intel's margins by two to four points in the 2026-2028 window โ that's real โ but the US government has a strategic interest in a sovereign process node. Taiwan's geopolitical risk is AMD's risk. The Intel foundry is a hedge AMD can't buy.
And that's the piece the upgrade thesis ignores. AMD's supply chain is a single source: TSMC, Taiwan. If the strait gets tense, AMD has no backup fab. Intel has its own fabs, its own capacity, its own depreciation headaches โ but also its own survival plan. The market's not pricing that asymmetry. It's pricing the earnings momentum.
The ARM Elephant
The bigger, unspoken threat isn't Intel. It's ARM. AMD and Intel are fighting the last war for x86 dominance, but the castle is burning. Amazon's Graviton, NVIDIA's Grace, Microsoft's Cobalt โ ARM-based CPUs are chewing through cloud-native workloads with better performance per watt. The hyperscalers aren't just buying chips anymore; they're designing their own silicon. That's a buyer-power shift that erodes the entire x86 duopoly.
I saw this pattern in the 2021 NFT boom โ everyone was fighting over the minting platforms, but the real value was the provenance layer underneath. Same story here. The real value isn't in AMD versus Intel. It's in the foundry layer, the architecture layer, and the geopolitical positioning that nobody's pricing.
The Valuation Trap
AMD's valuation is telling: 40x trailing earnings, 8x sales. The market has already priced AMD as the AI winner. Intel sits at 1.5x price-to-book โ a classic value trap. Low multiple, structural losses, a foundry business that burns cash like a DeFi project burning treasury. The market's not wrong to favor AMD on fundamentals. But the valuation leaves no room for the three variables that could break the thesis: AI demand softening, Intel 18A delivering, or ARM accelerating its encroachment.
Based on my experience working with institutional partners on ETF custody solutions โ the tension between compliance and decentralization โ the smart money is already positioned. They're not betting on AMD's CPU architecture. They're betting on TSMC's yield curve. They're betting on the manufacturing layer, not the design layer. The Strong Buy upgrade is just the institutional rubber stamp on a narrative that's already been price in.
The Takeaway
The AMD bull case is real for the next four to five quarters. The product momentum is real. The margin story is real. But the structural risk is a dependency disguised as a strength. Watch the 18A yield numbers the way you'd watch a protocol's TVL after the incentives get pulled. If Intel's yields hold above 80%, the AMD premium shrinks. If TSMC's capacity gets squeezed by NVIDIA and Apple, AMD's growth story hits a wall. The market prices the narrative. It doesn't price the queue position. That's the edge.
We didn't see the single-source fragility until it was already inside the trade. The code doesn't lie โ but the dependency does. Trust no one, verify everything, and know who actually owns the chips.