Copyright Investing.com
Wall Street sees an invincible monopoly. The customers see an exit strategy. Imagine owning the only gas station on a highway where every car was specially built to run only on your premium fuel. For two decades, that’s been NVIDIA’s (NASDAQ: ) business model in artificial intelligence. Their chips run on a proprietary system that costs millions to escape once you’re locked in. But in the data centers of Amazon (NASDAQ: ), Google (NASDAQ: ), Microsoft (NASDAQ: ), and Meta (NASDAQ: ), something is changing: they’re building their own gas stations. And redesigning their cars to run on regular unleaded. The market still values NVIDIA at $5 trillion because it assumes those golden handcuffs are unbreakable. But the handcuff keys are being mass-produced. NVIDIA Year-to-Date Chart (2025): Shares are trading near all-time highs above $200 The Two-Customer Problem That Could Sink A Trillion-Dollar Company NVIDIA revealed something remarkable in its latest SEC filing: just two unnamed customers account for 39% of the company’s revenue. Customer A (likely Microsoft) accounts for 23% of everything NVIDIA sells. Customer B (probably Amazon) accounts for 16%. To put that in perspective: if McDonald’s got 39% of its revenue from two people, you’d call that a crisis, not a customer base. The twist: those mega-customers are buying NVIDIA’s chips while simultaneously designing their own chips to replace them. Your two biggest customers are building competing products. The math is simple and severe: If Microsoft (Customer A) and Amazon (Customer B) shift even half their AI work to custom chips over the next three years, NVIDIA loses more than $10 billion in annual revenue. Not from a new competitor entering the market. From customers walking out the door. The Empire Built on Software Is Becoming Software-Optional For years, the defense of NVIDIA’s dominance went like this: "Sure, Amazon can build a cheaper chip, but good luck replicating two decades of software that every AI developer already knows." NVIDIA’s CUDA software platform was supposed to be an unbreakable lock-in. Learning it required years of specialized programming expertise. Switching away meant rewriting millions of lines of code. But OpenAI (ironically, one of NVIDIA’s biggest success stories) built a tool called Triton that lets any developer write high-performance code that runs on both NVIDIA chips and their competitors’ chips. They built a universal translator. It’s like realizing your iPhone apps can suddenly run on Android with no modifications. The ecosystem advantage just evaporated. When Your Pricing Power Becomes Your Biggest Vulnerability NVIDIA’s gross margins hit 78% when chips were scarce and customers were desperate. Server configurations that cost $350,000 in 2023 were suddenly $600,000 in 2025, a 71% price increase in two years. This is a monopoly extracting maximum rent before the competition arrives. In April 2025, HSBC downgraded NVIDIA with a blunt assessment: the company is losing its ability to raise prices. When they tried to charge more for the next generation of chips, customers simply said no. When your biggest customers can build comparable alternatives at half the cost, your premium pricing reflects the fact that they haven’t finished building their alternatives yet. Those alternatives are now 70% complete. The Market Is Splitting: NVIDIA Owns the Shrinking Half The AI chip market is bifurcating into two completely different economies: Building AI models (where NVIDIA dominates): Creating GPT-5 or the next breakthrough model requires cutting-edge horsepower and massive computing clusters. This is NVIDIA’s fortress. Blackwell chips maintain a technical lead measured in years. Running AI models (where custom chips are winning): Once you’ve built ChatGPT, you need to run it billions of times per day for users around the world. This requires efficiency, low power consumption, and cost optimization rather than bleeding-edge performance. Custom chips are purpose-built for efficiency. And the "running AI" business is 10x bigger than the "building AI" business. As the AI market matures from "build everything" to "run everything efficiently," the revenue mix shifts toward the segment where NVIDIA’s performance advantage matters least and custom silicon’s cost advantage matters most. By 2028, industry projections suggest nearly half of all AI computing will run on custom chips built by the cloud providers themselves. Where NVIDIA Actually Wins (And Where It Doesn’t) NVIDIA is transforming from "the AI infrastructure monopoly" to "a leading semiconductor company with strong but normalizing profit margins." The company has opportunities in software licensing, robotics, edge computing, and international markets where geopolitical concerns require third-party suppliers. But profit margins in these segments will be 40-50%, not the 70%+ NVIDIA enjoys today on data center chips. What NVIDIA won’t maintain: 70%+ profit margins on data center chips while its four largest customers are actively manufacturing competing products. The Risk Wall Street Won’t Price In If the biggest customers are building alternatives, and the software lock-in is breaking, and pricing power is gone, and customer concentration is at crisis levels... why is NVIDIA still trading like an invincible monopoly? Because the market always lags in reality. Consensus still prices NVIDIA as if its dominance is permanent. But the mechanics of margin compression are already in motion. They just won’t show up in quarterly earnings until it’s too late for investors to exit gracefully. NVIDIA Valuation Scenarios: The Math Behind the Risk When you model out three scenarios through 2028, the math exposes the asymmetry. These scenarios isolate three key drivers: earnings growth, margin compression, and multiple re-rating. Bull Case ($220 target, +57% upside): NVIDIA’s software and robotics pivot succeeds. Custom chips from hyperscalers remain limited to internal workloads. The company maintains 60%+ gross margins in data centers while expanding into new markets. Earnings compound at 25-30% annually. Multiple stays elevated at 40-45x forward earnings. This scenario requires custom chip adoption to stall and software licensing to become a material revenue stream. Base Case ($100 target, -29% downside): Gradual total addressable market erosion as hyperscalers shift 30-40% of workloads to custom chips by 2028. Gross margins compress by 200-300 basis points annually. Earnings growth moderates to 15-20% as revenue mix shifts toward lower-margin segments. Multiple re-rates to 25-30x earnings as investors recognize NVIDIA as a cyclical chip company rather than a permanent monopoly. This is the scenario HSBC hinted at with its April 2025 downgrade. Bear Case ($60 target, -57% downside): Rapid hyperscaler substitution accelerates. Microsoft, Amazon, Google, and Meta collectively shift 50%+ of AI workloads to custom chips within three years. Gross margins compress to 55-60% as pricing power erodes and NVIDIA competes on cost rather than performance. Earnings growth slows to single digits. Multiple compresses to 15-18x as the market prices in sustained margin pressure and cyclical semiconductor dynamics. The bull case requires several low-probability outcomes to align simultaneously. The base and bear cases simply require current trends to continue. That asymmetry is the risk Wall Street refuses to price in. The Bottom Line: The Margin Supercycle Is Ending NVIDIA will remain dominant and profitable. Dominance and profitability are different from being invincible and irreplaceable. The margin compression is already happening: Pricing power is gone (confirmed by bank downgrades in April 2025) Software lock-in is optional (new tools already in production) Customer concentration is existential (39% revenue from two customers) Custom chips are scaling (projected 45% market share by 2028) Tech history has a pattern: When your largest customers become your competitors, and when software becomes hardware-agnostic, margins compress. It happened to Intel with the rise of ARM processors. It happened to Cisco when cloud providers built their own networking equipment. It’s happening to NVIDIA right now. When analyst models shift from "NVIDIA is irreplaceable" to "NVIDIA is a cyclical chip company with maturing growth," the stock’s valuation multiple will compress from 50x earnings to 25-30x. If growth simultaneously slows from 40% to 15%, the base and bear case scenarios play out. The HSBC downgrade in April 2025 was a signal that the market is beginning to recognize what the data already shows: the margin supercycle is ending.
