AMD: The Selloff Tested the Thesis. The Thesis Held.

Five days ago I said hold AMD at $521.95 — the Anthropic deal was real but the valuation was stretched, and I wanted to see concrete capacity commitments before moving to accumulate. The market then ran a live stress test: a sector-wide AI rout knocked AMD down 18% to $429.56 by Wednesday, and on Thursday AMD announced a second infrastructure deal — up to 2.5 gigawatts with Core Scientific — and the stock bounced 13% in a single session. The selloff was macro. The bounce was thesis. That distinction matters.

The Call

HOLD — upgrading to ACCUMULATE on any pullback below $460.

The July 26 call was HOLD at $521.95 with two conditions to upgrade: either a pullback toward $450 where the multiple would price in more execution risk, or concrete evidence that the "up to 2 GW" Anthropic ceiling was converting into firm capacity. Both have now partially triggered. The stock traded as low as $424.03 on Wednesday — through my $450 add level — and the Core Scientific deal provides the infrastructure evidence. The Thursday bounce to $485.39 took back some of the opportunity, but the setup is better today than it was at $521.

Why Now

Three things changed since Sunday's HOLD call:

1. The sector selloff created the entry I was waiting for — briefly. AMD fell 17.7% from Friday's $521.95 close to Wednesday's $429.56 close, a three-day cascade driven by a global semiconductor rout: SK Hynix dropped 20%, South Korea's KOSPI fell 12%, and Leopold Aschenbrenner's hedge fund was reportedly forced to unwind all public stock positions after steep AI losses. None of this was AMD-specific. This was a liquidity event, not a thesis event.

2. AMD announced the Core Scientific deal on July 28 — the missing piece. AMD secured more than 500 megawatts of U.S. data center capacity beginning in 2027, expandable to 2.5 GW, through a partnership with Core Scientific (NASDAQ: CORZ). The deal includes collaboration on physical infrastructure design and deployment of Instinct GPUs, EPYC CPUs, and ROCm software. AMD also receives market-priced warrants to purchase CORZ common stock, subject to commercial conditions. This is the second major infrastructure commitment in eight days — Anthropic on July 22, Core Scientific on July 28 — and it directly addresses the execution risk I flagged: AMD now has a path to physical capacity for its AI deployments, not just a chip design and a promise.

3. Q2 earnings are Tuesday, August 4. Four trading days from today. This is the next hard catalyst and the first chance to hear management quantify the Anthropic and Core Scientific commitments on the record.

The Evidence

The Core Scientific deal is structurally different from Anthropic. The Anthropic partnership was about a marquee customer committing to AMD silicon. The Core Scientific deal is about AMD securing the physical infrastructure — power, land, buildings — to actually deliver. In AI, the bottleneck is increasingly not chip design but the ability to bring compute online. AMD is now attacking both sides of that problem.

The warrants signal AMD's conviction. AMD didn't just sign a capacity agreement — it took warrants in CORZ. That's a chip company putting financial exposure behind its infrastructure partner. It aligns incentives and suggests AMD views Core Scientific's buildout capability as strategically important, not just commercially convenient.

The selloff was indiscriminate. AMD fell 18% in three days alongside the entire AI complex. NVDA, the sector's anchor, declined as well. When the selling is macro and the buying is company-specific — as Thursday's 13% AMD bounce on the Core Scientific news demonstrated — the signal is that the market differentiates between thesis risk and liquidity risk. AMD's thesis risk actually decreased this week.

Key figures (as of July 30 close):

Sources: SEC EDGAR 10-Q (period ended 2026-03-28), AMD IR, Core Scientific IR. Price data pulled 2026-07-31.

Levels & Triggers

Action levels: - $460 or below: ACCUMULATE. At that level the trailing P/E drops to ~130x and the market is pricing in meaningful execution risk on both deals. The evidence says that risk is declining, not rising. - $520+ (pre-earnings): TRIM back toward HOLD. If the stock recovers to where it was before the selloff without earnings confirmation, the easy money from this week's dislocation is gone. - $380: Thesis-break level. A decline to this level would imply either a deal falling apart or a fundamental deterioration in AI capex demand. Would revisit the entire position.

Catalysts to confirm: - Q2 earnings (Aug 4): First look at Q2 revenue, and more importantly, management's framing of the Anthropic and Core Scientific commitments. Any quantification of expected 2027 revenue from these deals would be a strong positive. - MI450 ramp details: Any disclosure on TSMC capacity allocation or yield timelines for MI455X. - ROCm/Claude optimization milestones: The software gap with CUDA remains the key competitive risk. Evidence of Claude workloads running on ROCm at scale would be material.

Kill condition: If Q2 earnings reveal that either the Anthropic or Core Scientific deal is materially smaller than the "up to" headline implies — or if AMD guides Q3 data center revenue below Q1's implied run rate — the thesis breaks. The deals are the thesis right now; if they shrink, so does the case.

Horizon

12-18 months. The first Anthropic GW deploys in H1 2027. Core Scientific's 500+ MW starts coming online in 2027. The revenue from both deals is a 2027 story, but the stock will price it in advance — or punish the absence of progress — well before then. The Aug 4 earnings call is the next checkpoint.

The Bear Case

The strongest argument against AMD right now: the "up to" problem hasn't gone away. Both the Anthropic deal (2 GW, $5B) and the Core Scientific deal (2.5 GW) carry ceiling language. The firm floor is much lower: Anthropic's first GW in H1 2027, Core Scientific's 500+ MW starting 2027. If the actual deployments land at the floor rather than the ceiling, the revenue impact is a fraction of what the headlines imply. Meanwhile, AMD is trading at ~137x trailing earnings — a multiple that demands the ceiling, not the floor. The $5B equity investment in Anthropic is a private, illiquid bet with AMD's entire cash position ($5.6B as of Q1) as the backing. If AI capex spending slows before these deals convert to revenue, AMD is long commitments and short flexibility. The Thursday bounce was encouraging, but one good day doesn't de-risk a 137x multiple.

What I'm Watching

Tuesday, August 4: AMD Q2 FY2026 earnings. This is now the single most important data point for the thesis. I'm listening for three things: (1) Q2 data center revenue vs. Q1's implied run rate, (2) any quantification of Anthropic or Core Scientific revenue expectations, and (3) management's language on MI450 ramp timing. If the print is strong and the guidance is confident, the ACCUMULATE trigger at $460 may not survive the week.

Axelrod Research is independent equity analysis, for information only. This is not investment advice. Positions may exist in securities discussed. Past performance is not indicative of future results.

Sources

• AMD IR — AMD and Anthropic Strategic Partnership — Jul 22, 2026
• Core Scientific IR — AMD Infrastructure Partnership — Jul 28, 2026
• SEC EDGAR — AMD 10-Q (Q1 FY26) — Filed May 6, 2026
• CNBC — Aschenbrenner Hedge Fund Unwinds AI Positions — Jul 30, 2026
• Quartz — AMD Secures 2.5 GW AI Data Center Capacity — Jul 28, 2026
• Investing.com — AMD Sector Selloff Analysis — Jul 28, 2026