"I bet the company, but I didn't think I would lose." That was Morris Chang on the $10 billion he committed in 2013 to build 20nm capacity on the promise that one customer—Apple—would fill the fabs. He was right. Apple's A8 shipped in 2014, and TSMC never looked back. Apple's annual spend at the foundry grew from $2 billion that year to $24 billion in 2025, peaking at 25% of all TSMC revenue. For more than a decade, Apple's checkbook funded the yield-learning curve on every leading node since 20nm, and in exchange it got something money can't usually buy: the right to set the terms.
Look at the wafer math now and you can see those terms being rewritten in real time. According to a Chinese-language report, a 5nm-class wafer is approaching roughly 60,000 yuan, with TSMC said to be raising prices across 7nm, 5nm, 3nm, and 2nm amid AI demand—though the details could not be independently verified. UBS reportedly expects another wafer price hike as early as 2027 and lifted its target 13% before earnings; Morgan Stanley followed with a 12% bump. This is not how a foundry behaves toward its founding anchor tenant. A supplier protecting its biggest, most loyal customer holds price to keep the relationship sticky. A supplier raising price across the board, into a backlog it can't clear, no longer needs to court anyone.
Last week this column traced TSMC's margin story and the cost of its offshore fabs—Arizona, Japan, the capex ramp that compresses free cash flow even as it builds the future. This week the story moves down to the customer side of the ledger, because the same scarcity that fattens TSMC's margins is quietly reordering who controls the leading edge. And the clearest signal isn't in the price hikes. It's in Apple's sudden scramble.
The same scarcity that fattens TSMC's margins is quietly reordering who controls the leading edge. And the clearest signal isn't in the price hikes. It's in Apple's sudden scramble.
Apple is reportedly compressing the 2nm node to a two-generation, roughly two-year lifecycle and chasing 1.4nm for flagship SoCs as early as 2028. The framing in the trade press is that Apple is being "forced" off 2nm by AI. That's half right. What actually changed is documented in SemiAnalysis's Apple-TSMC model: Apple's share of N2 drops to 48%—the first time in a decade it is not the dominant launch customer on a new node. Sit with that. Since 20nm, Apple's node-launch share ran consistently above 50%, in some cases near 100%. It funded the ramp, it ate the early yield pain, and it got first call on the output. At N2, for the first time, someone else is in the front seat.
That someone is Nvidia. TSMC's high-performance-computing revenue grew from 36% in 2020 to 58% in 2025—61% of revenue in Q1 2026—while smartphone revenue shrank from 46% to 29%. The model projects Nvidia will consume more N3 wafers than Apple by the fourth quarter of 2027. The platform shift has already happened. For over a decade, Apple was the only company on earth capable of pre-funding bleeding-edge capacity at scale. Nvidia's AI cash changed that. Now two customers can fund TSMC's roadmap, and the moment a monopoly buyer becomes a duopoly of buyers, the seller's leverage stops being theoretical.
Here's the subtle part, and it's where the panic narrative oversells itself. Apple's N2 share didn't collapse because Apple lost a fight. TSMC's A16 was architected for HPC—backside power delivery, gate-all-around, thermal envelopes tuned for accelerators, not phones. Smartphones will largely skip A16. Apple's share falls because the node wasn't built for it. That's not displacement; it's an admission. The leading edge is no longer designed around Apple's product cadence. Apple's mobile roadmap is riding shotgun on a vehicle Nvidia is now steering. When the anchor tenant has to architect its launch timing around whether the next node was built for someone else's workload, the anchor tenant is no longer the anchor.
Which brings us to the tell: the Intel gambit. Washington floated an Apple-Intel chip deal last week, and the logic is seductive—Intel rebuilding foundry credibility, Apple hunting for capacity TSMC can't spare. Tim Cook said in April that supply constraints had held back iPhone sales. But read what the analysts actually say about timing. Malcolm Penn of Future Horizons calls it a "shotgun wedding" and puts the absolute best case at two to three years before first chips flow: two years to design an SoC of this complexity, four months of production cycle time to volume. Apple would likely want Intel's 14A, which TECHnalysis pegs at 2028 or 2029. Daniel Newman of Futurum thinks volume Apple silicon at Intel is unlikely before late 2027 or early 2028, starting with non-critical parts in a MacBook Air or iPad Pro. And Paul Meeks's line lands hardest: investors are "pricing in perfect execution by Intel, which is a company that hasn't delivered for about 20 years."
You do not negotiate a shotgun wedding to a foundry that's two-plus years from volume and two decades removed from delivery if your incumbent has spare capacity and a reason to discount you. You do it when your incumbent has neither. The Intel flirtation is not Apple gaining optionality. It's the behavior of a buyer that has lost pricing power and is shopping the market to remind its supplier that alternatives exist—even when those alternatives are, in Penn's words, "a huge leap of faith and commercial and financial risk." Samsung is on roughly the same clock: its 1.4nm slipped from 2027 to 2029 while it stabilized 2nm and landed Tesla. The credible second sources are all years out. Apple knows this. Floating the deal is the leverage play of someone with little leverage left.
Does 1.4nm fix it? The same SemiAnalysis model says Apple reclaims 67% node share on A14—its highest since N3—because TSMC is designing A14 for both mobile and HPC from the start, with a separate Super Power Rail variant for the accelerators. So Apple gets its front seat back. But look at what had to be true for that to happen: TSMC had to choose to build the node for both customers. Apple's recovery on A14 is granted by TSMC's roadmap design, not extracted by Apple's checkbook. That's the whole inversion in one line. Reclaiming 67% on A14 doesn't rebalance power. It resets the same dependency one node down, on terms TSMC sets.
This is the structural truth the memory side of the industry already learned: scarcity hands the supplier durable pricing power without anyone needing to collude. It isn't a conspiracy at TSMC; it's the arithmetic of being the only firm with 70% share, priority access to EUV, and a customer list where two giants now bid for the same wafers. The leading edge has become a seller's market even for the buyer who built it. Apple's $300 billion in supplier capex and 8,000 engineers don't change the seat assignment when the constraint is fab output and the price keeps climbing into demand that won't quit.
Chang bet the company and didn't lose. A decade later, the company he bet no longer needs to bet on Apple. The scramble to 1.4nm, the Intel shotgun wedding, the compressed 2nm lifecycle—none of it is a customer expanding its options. It's a tenant discovering the lease renews on the landlord's terms now.