AMD's $11.2B Quarter Is the Easy Print — Helios, the Anthropic 2GW Deal, and a 64x Forward P/E Decide Whether This Rally Has a Second Half

Advanced Micro Devices's $11.2B midpoint guide is essentially consensus — published estimates cluster at $11.28B (Tickeron), $11.34B (Investopedia) and $11.3B (LSEG). The bar for a beat is roughly $11.45–11.50B, and a miss on revenue is hard to construct given Q1 already printed $10.25B (+38% YoY) with Data Center alone at $5.8B. The interesting question is mix: management guided non-GAAP gross margin to ~56% in Q2, up from 55% in Q1, and the call needs to show Data Center GM holding north of mid-50s as NVIDIA's Blackwell GB300 cycle bottomed at 72.4% GAAP in NVDA's Q2 FY26 (Aug 27, 2025).

On July 22, 2026, Advanced Micro Devices and Anthropic announced up to 2GW of MI450/MI455X GPUs in Helios rack-scale systems, with the first gigawatt shipping H1 2027. AMD will also invest up to $5B in Anthropic equity as part of the deal — and Anthropic's Series H (May 2026) priced at a $965B post-money valuation on a $47B run-rate revenue (Anthropic IR). AMD is buying into a customer that just raised at ~20x revenue, with a hardware commitment the WSJ describes as 'tens of billions of dollars.'

Each Helios rack is a 72-MI455X system connecting 432GB HBM4 per GPU (31TB total), tied together by UALoE fabric running over Broadcom's 102.4 Tbps Tomahawk 6 switches (Broadcom shipped TH6 in June 2025). That makes Broadcom the connective tissue of both AMD and Nvidia rack-scale strategies — Nvidia's NVL72 depends on the same switch silicon for scale-out. The AMD-specific layer is TSMC's CoWoS packaging: AMD reserved 61,000–75,000 CoWoS wafers for MI400/MI450 in 2026 against an industry total of ~120,000–140,000 wafers/month per industry tracking.

Industry estimates peg Helios at ~$5M per rack, ~$1M above a comparable Nvidia Vera Rubin NVL72 system. AMD's pitch is that 50% more HBM4 capacity and 15% more FP4 compute per rack deliver better tokens-per-dollar for frontier inference — but at the corporate level, AMD's TTM gross margin is 50.3% vs NVDA's 74.1%. Even if MI450 ASPs land in the $35,000–$40,000/GPU range (analyst estimates), the ~24-point gross margin gap means every dollar of MI450 revenue converts to a fraction of the gross profit NVDA prints on Blackwell/Rubin.

The bear case is that Blackwell's first-year gross margin trough (NVDA dipped to 72.4% GAAP / 72.7% non-GAAP at Blackwell ramp) replicates at AMD as MI450 hits volume — but AMD has less cushion because it starts from a 50.3% TTM blended GM, and the MI455X 12-stack HBM4 cost is the single biggest variable. The bull case is that the 56% Q2 guide → 58%+ by Q4 holds, and FY27 GM expansion unlocks the 2x EV/sales multiple AMD currently commands.

NVDA's Q1 FY27 (Apr 26, 2026) printed $81.6B revenue (+85% YoY) with Data Center at $75.2B (+92%), and Q2 FY27 is guided to $91B. AMD's full-year FY26 consensus is $47–50B vs NVDA's run-rate of ~$320B+. The numeric gap is 6–7x — AMD's bull case is not 'catch NVDA,' it's 'multiply revenue 3x by FY28 from a smaller base.' That's the framing management has been telegraphing since the OpenAI 6GW deal in October 2025 and the Meta Platforms 6GW in February.