Raymond James just upgraded AMD to Strong Buy. The rationale reads clean: AMD is eating Intel's server lunch, the EPYC line is compounding, and the AI tailwind is real. But the market is buying a story about AMD's execution. What it's actually buying is a bet on TSMC's capacity allocation, Intel's manufacturing delays, and the quiet assumption that Intel's 18A node will arrive late — or not at all.
Let's be blunt: AMD's "technical edge" is not its own. It's rented from TSMC. AMD designs, but TSMC prints. This is a fabless advantage that has a hidden tax — one that appears when you look at the global liquidity map. AI demand is squeezing the same 3nm and 5nm lines that AMD's entire server roadmap depends on. NVIDIA and Apple get priority. AMD gets whatever's left. The market is pricing AMD like it owns the factory. It doesn't.
Intel, on the other hand, is being written off as a structural loser. That's the consensus, and consensus is a lagging indicator. The real story is that Intel is the US's only shot at a credible advanced-node factory. Washington knows it, the Department of Defense knows it, and the CHIPS Act is effectively a state-backed subsidy to keep Intel's bleeding alive until the strategy turns. The market sees a value trap; I see an option on geopolitical chaos.
The Process Gap Is a Mirage
Let's get the numbers on the table. AMD is on 5nm and 4nm for the Zen 4/Zen 4c generation, and Zen 5 is moving to 3nm. Intel is stuck on Intel 7 for its server chips — that's roughly an enhanced 10nm. Intel 4 is running in client Meteor Lake. Its 18A node, which it claims will match TSMC's 2nm, is targeted for 2025 production. On paper, AMD leads by one to two process nodes in the server market. In practice, this lead is TSMC's yield curve, not AMD's.
I've audited enough contracts to know the difference between a technical roadmap and a shipping roadmap. TSMC's 3nm yields have climbed above 80% in 2024, and that maturity de-risks AMD's entire lineup. Intel's 18A is reportedly sitting at 60-70% yield in early stages. That's not a functional process. That's a lab experiment. The gap means Intel's foundry unit — Intel Foundry — cannot be profitable at scale until those yields hit 80%+ and stay there. That's a multi-year process. Every quarter that 18A slips, AMD's process advantage extends.
The packaging story matters here, too. AMD's Chiplet architecture is a structural weapon. It's not just about die size or cost — it's about flexibility and yield. AMD can mix and match CCDs and IODs across product tiers without re-spinning the entire design. Intel's EMIB and Foveros are technically impressive, but they're built for monolithic, high-cost designs. In a world where AI chips are sucking up CoWoS packaging capacity, AMD's Chiplet approach is a way to route around the bottleneck. The market doesn't price this design freedom. It should.
But here's the uncomfortable part: AMD's Chiplet advantage is locked to TSMC's output. The more AMD grows, the more it competes for the same limited advanced-node capacity as NVIDIA and Apple. In 2024, TSMC's 3nm and 5nm lines are running above 90% utilization. Every MI300 shipped is a trade-off against an H100. AMD's is not in the driver's seat. It's in the passenger seat, with a good map.
The Liquidity Illusion in AI Demand
AI server demand is the engine that's driving AMD's narrative. And the numbers are seductive: server CPU revenue could grow from $50 billion in 2024 to $70 billion by 2028. AI servers need 2-3x more CPUs per unit than traditional boxes. AMD's EPYC has been winning in this segment because of its performance-per-watt and price advantage over Intel's Xeon. This is real. I've seen the procurement sheets from cloud providers. They're not buying Intel's chips out of loyalty.
But let's apply the forensic lens. The AI investment boom is a liquidity event masquerading as a technology revolution. Hyperscaler capex is flooding into data centers, and a portion of that is speculative, driven by the fear of missing the next platform shift. When the AI commercialization curve doesn't match the spending curve — and it won't, because the hype cycle always overshoots — the capex spigot will tighten. AMD's data center segment is 40-45% of its revenue. That's a concentrated bet on a single narrative. When the correction comes, AMD's stock will trade like a leveraged AI play, not a diversified semiconductor company.
Intel's in a different position. It has a broad customer base across PCs and legacy enterprise. That's not a growth engine, but it's a ballast. The AI impact on Intel's CPU business is more diluted, but that's also a risk: the AI boom is lifting AMD's boat more than Intel's. If the AI bubble pops, AMD's falls harder. The asymmetry is not in AMD's favor.
The Geopolitical Shell Game
Regulation and geopolitics are the hidden variables that will decide this race. AMD's dependence on TSMC is a single point of failure, and it's concentrated in Taiwan. In a conflict scenario, AMD has no backup foundry. Intel, however, has factories in Arizona, Ohio, New Mexico, and Oregon. The US government has a strategic interest in keeping Intel alive — not because Intel is the best, but because it's the only American advanced process. The CHIPS Act is essentially a subsidy to Intel, but it's a forced hedge for the US.
The real asymmetric risk is the export control regime. Intel makes 25-30% of its revenue from China, while AMD gets 15-20%. The US is tightening export controls on AI chips, and AMD's MI300 is restricted. Intel's Gaudi 3 is also restricted. But Intel's broader PC and server CPU line is not fully cut off from China. AMD's exposure is more concentrated in data center, which is exactly the segment that's being restricted. So, the trade war hits AMD harder relative to its revenue base.
Meanwhile, China's domestic substitution is not a joke. With the Big Fund III injection of $344 billion, domestic x86-compatible (Hygon) and ARM-based (Phytium, Kunpeng) chips are gaining ground. Intel is the bigger target — it has more mature market share to lose. But AMD's share in China is a smaller pie, and the marginal loss matters less. The real chess game is that Intel's China revenue is a political hostage. AMD is a better geopolitical position — which is why the market rewards it.
The Contrarian Angle: The Intel 18A Timeline Is the Key Variable
Now, the part most analysts miss. The Raymond James upgrade is implicitly a bet that Intel's 18A doesn't scale in 2025. If 18A yields and ships on schedule — with the performance that Intel claims — the entire AMD thesis gets shaken. The process gap narrows, and Intel's new products (Granite Rapids successor) become a real threat. The market has written off Intel's execution, but Intel has a history of pulling rabbits out of hats in the last minute. The risk is not zero.
But here's the more subtle point: AMD's process advantage is a race against the TSMC clock, not just Intel. TSMC's 2nm GAA process is scheduled for 2025-2026, and AMD's Zen 6 will ride it. If TSMC's 2nm yields are better than Intel's 18A — which is likely — AMD stays ahead. The real question is whether TSMC's capacity allocation will favor AMD or NVIDIA. The market doesn't price this properly.
The Final Word: Who Actually Has the Leverage?
Let's break down the balance of power. AMD is a higher-margin, high-ROIC business — 50%+ gross margins, and ROIC above WACC. It's creating value. Intel is destroying value — ROIC below WACC, negative free cash flow, and a giant capex burden. In the short term, the market is right: AMD is the better business.
But the market is also projecting the current momentum linearly. It's ignoring the structural fragility of AMD's supply chain, the AI spending peak, and the geopolitical risk. And it's pricing Intel's as a zero — but Intel is not a zero. It's a politically protected, national-security-backed option. If Intel's 18A ships on time, AMD's edge is gone in two years. If it doesn't, AMD's a good story for a few more years — but the ARM threat is coming from the outside.
In a way, this is the same old cycle: hype is just liquidity with a distorted memory. The market sees AMD's share gains and forgets that AMD's process is a rented one. The market sees Intel's losses and forgets the government is its biggest shareholder.
The Takeaway
Buying AMD is a bet on TSMC's capacity and a bet against Intel's execution. Both of those could be correct — for the next 12-18 months. But this is a crowded trade. The smart money should be watching Intel's 18A yield curve, not AMD's next earnings. And in a macro environment where AI spending is the only visible engine, distraction is the tax we pay for novelty.
Volatility is the price of entry. And the next correction will be triggered by a failed Intel yield check — or an AI capex miss — not by an AMD product failure. AMD is a better business. But it's not a moat. It's a window. And windows close.
The next cycle will not belong to the best product, but to the most resilient supply chain. AMD doesn't have one. Intel is building one. That's the contrarian play that no one's pricing yet.
It's not about who wins the tech. It's about who survives the geopolitics. And that's a game where the factory's worth more than the code.
Hype is just liquidity with a distorted memory. But the factory, in the end, is the only thing that's real.