AMD: The Anthropic Deal, a $5B Stake, and the First Real Crack in Nvidia's Monopoly

Single-stock deep dive · Published July 22, 2026 (~12:05 pm ET) · @dailyanalysts

The one-line thesis: AMD just signed its fifth blue-chip AI compute customer (Anthropic, up to 2GW + a $5B equity stake) on the same morning the market was busy convincing itself AI hardware demand is about to collapse. That contradiction is the opportunity. The Kimi-K3 "cheaper-inference-kills-chips" panic is, on a relative basis, an AMD tailwind — and the 2027 order book is now large enough to de-risk the very ROI fear crushing the group. My call: accumulate AMD on weakness $495–525, HIGH CONVICTION, 12-month target $640 (stretch $700), invalidation on a weekly close below $460. Do not chase $556 into the August 4 print.

1. What happened today (the catalyst)

At 9:00 am ET, on the opening morning of its "Advancing AI 2026" conference in San Francisco, AMD and Anthropic announced a strategic partnership (AMD IR press release). The terms, straight from the primary source:

Up to 2 gigawatts of AMD Instinct MI450 Series (MI455X) GPUs deployed in AMD Helios rack-scale systems, with the first gigawatt beginning in the first half of 2027. Built on EPYC "Venice" CPUs, Pensando networking, ROCm software.
AMD commits a strategic equity investment of up to $5 billion in Anthropic — AMD's first-ever investment in an AI lab. Per the WSJ/TradingKey reporting, the capital tranches unlock as deployment milestones are hit.
A multi-year engineering collaboration: Anthropic will use Claude to accelerate AMD's ROCm software development (a direct attack on Nvidia's CUDA moat), and AMD adopts Claude across its engineering teams.
Builds on Anthropic's existing use of AMD MI355X GPUs — this is an expansion, not a cold start.

This landed 48 hours after a second deal: Microsoft agreed to deploy AMD Helios racks in Azure (7/20 release), joining Meta, OpenAI and Oracle.

The tape's verdict: AMD opened down ~3% (a classic sell-the-news reaction — the Anthropic deal had been telegraphed by Jefferies for weeks), then reversed violently to trade $556.08, +2.14% on the day (intraday range $526.60–$558.42, a bullish outside-reversal candle). Nvidia rose +2.87% to $213.24 alongside it — the market is reading this as the pie growing, not a zero-sum knife fight.

2. Company overview & business model

AMD is now a full-stack AI infrastructure vendor, not just a CPU house. Four pillars: Data Center (EPYC server CPUs + Instinct GPUs — now the growth engine and, as of Q4 2025, generating more GPU revenue than CPU revenue in a quarter), Client (Ryzen PC/AI-PC), Gaming (semi-custom consoles + Radeon), and Embedded (Xilinx FPGA). The bull case rests almost entirely on Data Center Instinct GPUs displacing a sliver of Nvidia's ~88–92% share.

Q1 FY2026 actuals (last reported)

Metric	Value	Note
Total revenue	$10.25B	+37.9% YoY
Non-GAAP EPS	$1.37	Beat $1.29 consensus
Data Center revenue	$5.78B	+57% YoY
Q2 guide	~$11.2B	~46% growth

Next print: Q2 FY2026 earnings August 4 (after close), consensus EPS $1.63, revenue $11.39B. That is the near-term binary — not this deal.

3. The order book most people are underweighting

Here is the point the "AI is a bubble" crowd keeps skipping: AMD has now assembled a customer roster that, two years ago, was 100% Nvidia territory. Cumulative disclosed Instinct commitments:

Customer	Commitment	Structure / timing
OpenAI	6 GW	Warrant for 160M AMD shares (~10%) at $0.01, vesting on milestones; first 1GW MI450 in H2 2026
Meta	up to 6 GW	~$100B framework (Feb 2026), custom MI450 silicon
Oracle	50,000 MI450 GPUs	Helios supercluster
Microsoft	Azure Helios racks	Announced 7/20/2026
Anthropic	up to 2 GW	+ $5B AMD equity stake; first 1GW H1 2027

The Street consensus (and AMD's own commentary) points to roughly ~5 GW of AMD compute deploying in 2027 across these names. Lisa Su has publicly guided to >35% Data Center revenue CAGR and a long-term EPS target above $20 (Financial Analyst Day, Nov 2025), with "leading customer forecasts exceeding our initial expectations." The consequence: AMD 2027 revenue is becoming a book-of-business problem, not a demand-guessing problem. That visibility is exactly what the Kimi-K3 selloff says the sector lacks.

4. The contrarian insight: Kimi K3 helps AMD (relatively)

On July 16–17, China's Moonshot released Kimi K3, a 2.8-trillion-parameter open-weight model rivaling OpenAI/Anthropic frontier models while reportedly using fewer advanced chips. JPMorgan called it "DeepSeek 2.0." The SOX fell ~4% that Friday and is down ~19% from its June 22 peak — flirting with a technical bear market. The fear: cheaper models → less capex → less chip demand.

Here is what consensus is missing. Kimi K3 is fundamentally an inference-economics story, not a training story. And two second-order effects cut in AMD's favor:

Jevons paradox. Cheaper inference doesn't shrink total inference demand — it explodes it. More tokens served = more accelerators deployed. Anthropic's Tom Brown said it plainly today: "map the right workloads to the right hardware." That is a multi-vendor, inference-weighted future — precisely AMD's opening.
AMD's architecture is memory-heavy, which is what inference wants. The Helios rack (72× MI455X) carries ~31TB of HBM4 and claims ~1.5x the memory capacity of Nvidia's competing Vera Rubin NVL72 rack. Memory bandwidth and capacity are the binding constraint for large-context inference. If the workload mix tilts from training toward inference — which cheaper Chinese models accelerate — AMD's relative competitiveness improves.

So the very headline gutting the group is, on a relative basis, an argument for the #2 player, not against it. That is the trade almost nobody is framing correctly this week.

5. Competitive reality — don't oversell it

I follow the evidence to its honest end, and the honest end is: Nvidia is still winning, and AMD is still #2.

Dimension	AMD (Helios / MI455X)	Nvidia (Vera Rubin NVL72)
Data-center GPU share	~8–12%	~88–92%
Rack memory	~31TB HBM4 (~1.5x)	~20TB HBM4
Reported rack cost	~$5.0–5.5M	~$3.5–4.0M
Software moat	ROCm (catching up; now Claude-assisted)	CUDA (entrenched, ~15-yr lead)
Q1 DC growth	+57% (off ~$5.8B base)	+92% (off ~$82B base)

AMD's rack is more expensive per unit and its software ecosystem is years behind. The MI455X still has to hit mass production and pass Anthropic's performance validation before H1 2027 — that is real execution risk, not a formality. CUDA remains the reason a startup defaults to Nvidia. The Claude-for-ROCm collaboration is the single most strategically interesting piece of today's deal precisely because it attacks that moat, but it will take quarters to prove.

6. Valuation — expensive, but not indefensible

Metric	Value
Price (7/22, ~12pm ET)	$556.08
Market cap	~$883B
P/E (TTM)	~176x (EPS TTM $3.05)
Forward P/E (consensus)	~69x
Revenue growth (YoY)	~35%
Gross margin	50.3%
Beta	2.51 (high)
52-week range	$149.22 – $584.73 (high 6/30/26)
YTD / 1-yr	+131% / +209%

My read: 176x trailing is a headline scare number; the honest lens is forward. At ~69x forward and a credible path to >$20 EPS by 2027–2028, AMD trades at roughly ~25–27x its 2028 earnings power — not cheap, but reasonable for a >35% grower with a locked order book. The catch: a 69x forward multiple prices in flawless execution. Any guide wobble on August 4, any MI450 slip, or a China-export tightening (MI308 controls already cost ~$440M in FY2025 charges) resets the multiple fast. This is a high-beta (2.51) name in a hawkish tape.

Where the Street is

UBS (Arcuri): $700 (raised from $670, 7/15)
Goldman Sachs: $640
Stifel: $635 (+41% from $450, 7/11)
24/7 Wall St.: base $562.88, bull $629, bear $434
Citi: $460, Neutral (the valuation skeptic)

7. Risks (quantified)

It's a 2027 story. Zero of this deal hits the P&L before H1 2027. The stock already fell 3% on the actual news — announcement fatigue is real.
Circular financing. AMD invests $5B in Anthropic → Anthropic buys AMD chips. Same reflexive loop as the OpenAI 160M-share warrant. It flatters demand and it is exactly the structure bears point to when they call the AI complex a house of mirrors. Watch for this critique to gain traction if any lab's funding wobbles.
Macro. 10Y at 4.63% (near 2026 highs), Fed hike odds at 28% for July 29 (up from 11% a week ago) and 87% for at least one hike by year-end, WTI spiking to ~$86–94 on the Iran war / Houthi Red Sea threat. Rising rates compress high-multiple growth; a 2.51-beta name is the first to get hit.
Execution. MI455X must ship at volume and validate on Anthropic's workloads by H1 2027. ROCm is still behind CUDA.
Near-term swing factor: Alphabet and Tesla report tonight (after close). Alphabet's capex guide (already $190B for 2026, "significantly higher" 2027) is the read-through — raise it and chips rip; hold/cut it and AMD retests the low $520s.

8. Technical picture

AMD round-tripped from its $584.73 ATH (6/30) down ~11% into last week's Kimi-K3 washout, then bounced. Today's outside-reversal off $526.60 with a close attempt near $558 is constructive but sits below the ATH. Map:

Resistance: $584.73 (ATH). A clean break targets the $629–$640 analyst zone.
Support: $520–526 (today's low / gap), then ~$495–500 (recent shelf), then $460 (Citi target / structural).
Invalidation: a weekly close below $460 breaks the uptrend structure and would signal the order book is being repriced or macro has taken over.

9. Three-scenario framework (12 months)

Scenario	Prob.	Trigger	Price
Bull	30%	Alphabet/hyperscaler capex guides higher this week; Advancing AI Day 2 (7/23) delivers MI500 detail + another named customer; Aug 4 beat-and-raise. Inference-mix rotation favors AMD memory density.	$640–$700
Base	50%	Deal is "priced in"; AMD consolidates $520–585 into Aug 4; earnings beat but 2027 is the story; grinds higher on order-book visibility.	$600–$640
Bear	20%	Fed hikes 7/29, oil stays bid, 10Y >4.75%; high-beta AI compresses; Kimi-K3 capex-cut narrative reasserts; MI450 timing doubt. Weekly close <$460 pauses the thesis.	$430–$495

10. My opinion & recommendation

HIGH CONVICTION AMD — accumulate on weakness

Entry zone: $495–$525 (buy the dips; do NOT chase $556 into Aug 4)
Target: $640 (12 months), stretch $700
Invalidation: weekly close below $460
Timeframe: 6–12 months
Conviction basis (2+ signals): (1) Anthropic deal + Microsoft Azure Helios deal in 48h; (2) analyst target wall lifting to $635–$700; (3) the contrarian inference-mix tailwind from Kimi K3. That clears the HIGH CONVICTION bar.
Sizing: high-beta (2.51) — size at ~half a normal position given macro (Fed hike risk 7/29, oil). Add the second half only on a weekly close back above $585 (ATH breakout) or on a flush into $495–500.

My honest fair-value estimate is ~$620 today — roughly 12% above spot — justified by ~27x 2028 EPS on a de-risked order book, discounted for execution and circular-financing risk. I am bullish AMD but I refuse to buy the top tick of a name that just round-tripped 11% and reports in 13 days. Patience on entry is the edge here.

Pair trade / read-throughs: The bigger AMD's Helios deployment, the more HBM4 demand — reinforcing my open HIGH CONVICTION call on Micron (MU) (each Helios rack ~31TB HBM4, 1.5x Nvidia). AMD's $5B Anthropic stake is also a de-facto call option on Anthropic's rumored ~$965B IPO.

Summary — main thesis & actionable steps

The Anthropic deal isn't about 2026 revenue (it's zero). It's about 2027 visibility and cracking Nvidia's software moat via Claude-assisted ROCm.
The Kimi-K3 selloff logic helps the memory-heavy #2 player on a relative basis. That's the non-consensus point.
Buy $495–525, target $640/$700, stop on weekly close <$460. Don't chase $556.
Near-term swing = Alphabet capex guide tonight and AMD's own Q2 print Aug 4.
Own MU alongside as the HBM4 derivative.

Sources: AMD/Anthropic partnership press release (primary); AMD IR press-release archive (Microsoft Helios, Q1 results); Charles Schwab Market Update, 7/22/2026; 24/7 Wall St. price-target note; Business Insider on the Kimi-K3 chip selloff; AMD/OpenAI 6GW deal (warrant structure). Price/quote data via Finnhub as of ~16:00 UTC July 22, 2026. This is analysis and opinion, not personalized investment advice.