The AI Chip Three Kingdoms — Nvidia · AMD · Intel Deep Dive
Who survives and who takes the throne in H2 2026
① The star of 2026's AI chip rally isn't Nvidia. Intel +195% and AMD +140% led the market while Nvidia managed just +8.8%.
② In the first week of July, Samsung's earnings triggered a reckoning — the chip sector plunged 6.8% in a day as 'AI bubble' warnings turned real. This is a valuation re-rating phase.
Follow-up: the Nvidia-lagging pattern held for a while — four straight quarters of beating and falling — until the August 26, 2026 Q2 print broke it on FY2028 guidance. See Inside Nvidia's Q2.
③ Each company faces a different decisive battle — Nvidia must defend against custom silicon, AMD must ramp its MI accelerators, and Intel's fate hangs on 18A yields.
① The rally changed its face
If 2024–2025 was Nvidia's solo run, 2026 looks completely different. Year-to-date: Micron +224%, Intel +195%, ARM +171%, AMD +140% — and Nvidia at +8.8%. The company that should be the biggest AI infrastructure beneficiary has become the sector's black sheep.
This is not simple rotation. It's a structural signal that the AI chip market's center of gravity is dispersing from 'GPU monopoly' toward memory, foundry, and custom silicon. And in July's first week, warning lights flashed across the entire rally: Samsung beat consensus by 'only' 6%, global chips sold off in unison, and Bank of America formally flagged bubble risk. This report dissects the real strength of the big three from inside that storm.
② Nvidia — the paradox of the throne
First, to be clear: Nvidia's business has not deteriorated. Data center revenue hit $75.25B last quarter, up 92% YoY. The problem is the stock had already priced in more than that.
- The custom silicon invasion — On July 8, Apple announced a $30B multiyear chip agreement with Broadcom. After Google (TPU) and Amazon (Trainium), now Apple: Big Tech's migration from Nvidia GPUs to in-house designs is accelerating, with Broadcom the biggest beneficiary.
- DeepSeek-driven demand uncertainty — Reports say China's DeepSeek is developing its own AI chip. As efficient models spread, the era of 'just buy more GPUs' may be ending.
- The curse of the base effect — At a world-leading market cap, even 92% growth becomes 'expected growth.' The room for surprise is narrow.
The CUDA software moat remains intact, and 92% data center growth is overwhelming in absolute terms. +8.8% YTD is the number of a resting stock, not a broken one. But a 2024-style explosive re-rating is hard to expect from here.
③ AMD — from challenger to Wall Street's top pick
AMD is having a career year. In late June, Cantor Fitzgerald raised its target from $500 to $700 and named AMD its top pick in compute — ranked above both Nvidia and Broadcom, implying 35%+ upside from the current $517.
The core thesis is the MI-series AI accelerator ramp. Hyperscalers fatigued by Nvidia's supply constraints and pricing wanted a second source, and AMD took the seat. Training still belongs to Nvidia, but in inference AMD is winning real share on price-performance.
The risks are equally clear. In the July 7 selloff AMD fell 6.5%, worse than the sector average — a +140% stock carries that much valuation sensitivity. And the custom silicon invasion isn't just Nvidia's problem: if Big Tech goes in-house, AMD's 'second supplier' logic weakens too.
④ Intel — the last gate of the comeback story: yields
The biggest reversal story of 2026 is Intel, up roughly 270% in the first half alone. Foundry separation hopes, 18A process milestones, and government support stacked up.
But cracks appeared in July. Intel fell -9.7% on July 7 alone — nominally part of the Samsung-triggered selloff, but with two Intel-specific negatives layered on top.
- 18A-P yield problem — The critical next-gen node has entered risk production, but current yields fall short of commercial profitability. Profitable yields may slip to late 2026 or 2027.
- The price-hike paradox — On July 6, Intel announced selective price increases on consumer and Xeon server processors. Meant to offset supply-chain costs, the market read it as a confession of margin pressure.
The +270% rally pre-priced a 'foundry renaissance' future. For that future to arrive, Intel must pass through the 18A yield gate. Every delay can shake the stock as violently as this week did. This is the single most dangerous type of stock to chase at highs.
⑤ Head-to-head — investment points at a glance
| Metric | Nvidia (NVDA) | AMD | Intel (INTC) |
|---|---|---|---|
| Price (7/8) | $204.12 | $517.41 | $110.24 |
| YTD return | +8.8% | +140.3% | +195.5% |
| Key weapon | CUDA ecosystem · training GPU dominance | MI accelerators · inference value | 18A node · US-based foundry |
| Biggest risk | Custom silicon · expectations burden | Rich valuation · weakening #2 logic | 18A yield delays · margin squeeze |
| Street view | Still dominant in absolute scale | Cantor top pick (PT $700) | Some see 60% upside (HSBC) |
| Character | Maturing into stable growth | High-beta momentum | Turnaround bet |
⑥ The 'AI bubble' debate — what makes this correction different
July 7's selloff was a textbook valuation re-rating. Samsung's results weren't bad — a 6% beat. The problem: a stock up 150% this year wasn't satisfied by that. Micron fell as much as 13% intraday, erasing roughly $138B in market value, and reports of SK Hynix moderating HBM expansion stoked doubts about whether AI demand has peaked.
Bank of America's warning, summarized: "AI infrastructure valuations have stretched well beyond historical norms; if capex-to-revenue conversion disappoints even modestly, bubble risk becomes real."
The decisive difference from the 2000 dot-com bubble is that the earnings are real. Nvidia's +92% data center growth and Micron's +345% revenue growth exist on income statements. Today's debate isn't 'is the demand fake' but 'how long can this growth rate last' — which also means corrections like this week's will repeat whenever deceleration signals appear.
⑦ A practical playbook for Korean retail investors
Stability-first — Skip single names; use semiconductor ETFs like SOXX/SMH to spread across all three plus equipment makers, cushioning single-stock -10% shocks like this week's.
Growth-oriented — Dollar-cost into AMD and Nvidia. With the July correction still unfolding, split entries into 3–4 tranches rather than one buy.
Aggressive — An Intel turnaround bet requires tracking 18A yield news religiously. SOXL (3× leverage) suffers maximal volatility decay in regimes like this — short-term trades only.
Universal checklist: ① Keep semiconductors under 30% of your portfolio ② Avoid new entries in the volatility windows around earnings ③ US chips move in sync with SK Hynix and Samsung — manage your total semiconductor exposure across US and Korean holdings combined.
The lesson of 2026's chip market: the champion doesn't always win. While Nvidia rested, Intel and AMD nearly tripled — and this week put that entire rally on trial. What's needed now isn't prediction but allocation: rather than betting on one company's victory, mixing the 'all three survive' scenario (ETFs) with individual-winner bets (staged entries) is the realistic optimum for retail investors.
Why training demand and inference demand call for different hardware is explained from the ground up in How LLMs Work. For the vocabulary that shows up on earnings calls, see the infrastructure cluster in 50 AI Terms. For how this three-way fight extended beyond GPUs into the developer-platform layer, see Inside Nvidia's Hugging Face Acquisition.
※ This report is for informational purposes as of July 9, 2026 and is not investment advice. All figures verified against Yahoo Finance (yfinance) and public reporting. Investment decisions and responsibility rest with the investor.
※ This report is provided for informational and educational purposes only and does not constitute a recommendation to buy or sell any security.
New research, when it lands
Subscribe and the next deep dive comes to you, along with the daily market brief — free, unsubscribe anytime.