Friday, August 28, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

AMD has lost a significant compute allocation to Nvidia after SpaceX selected Nvidia accelerators over AMD alternatives for its next-generation AI infrastructure.

This reallocation demonstrates accelerating hyperscaler and aerospace preference for Nvidia silicon, tightening the competitive moat for AMD in the high-performance training and inference markets.
Trade pressSlicast · August 22, 2026 · US · Source: Google News
importance 88

Elon Musk’s decision to build SpaceX’s (NASDAQ:SPCX) AI infrastructure around NVIDIA’s (NASDAQ:NVDA) Vera Rubin platform, reportedly securing 20% of NVIDIA’s next-generation capacity, is the kind of development that gives AMD investors pause. It denies Advanced Micro Devices (NASDAQ:AMD) a marquee customer just as CEO Lisa Su is working to convince Wall Street that its Instinct roadmap can compete with Jensen Huang’s offerings.

The market reacted swiftly. Shares closed at $469.46 on Thursday, marking a 13.77% decline over the past month despite a strong quarterly earnings report.

The key question is whether SpaceX represents a structural weakness or merely a single lost deal in a market AMD projects will approach $2 trillion by 2030. Evidence points toward the latter interpretation.

Musk’s primary rationale centers on supply allocation. SpaceX requires guaranteed access to a platform controlled by NVIDIA, which ultimately decides how capacity is distributed. This grants NVIDIA leverage that AMD cannot yet replicate: the ability to turn hardware scarcity into exclusivity for the world’s largest AI buyers.

Furthermore, SpaceX aligns with a specific customer profile. As a single-tenant buyer philosophically aligned with Musk’s xAI, its Grok training workloads benefit from running on a unified hardware stack.

While losing SpaceX is notable, interpreting it as proof that AMD is faltering ignores a critical reality: AMD has actually won this cycle. The company reported second-quarter revenue of $11.54 billion, a 50.1% year-over-year increase, with data center revenue surging 107% to $6.72 billion. That segment now generates $2.1 billion in operating income, a significant swing from a loss a year ago tied to China export charges.

The customer pipeline supporting AMD’s growth trajectory is robust. The company has secured deployment commitments from OpenAI, Meta (NASDAQ:META), and Anthropic, whom Su described as “strategic anchor customers.” Additionally, AMD has a two-gigawatt agreement with Anthropic for its MI450 processors within the Helios architecture, alongside large-scale deployments on Microsoft (NASDAQ:MSFT) Azure.

Speaking with analysts, Su stated, “We enter the second half with strong momentum as EPYC demand accelerates, Instinct deployments scale and Helios begins to ramp.”

NVIDIA retains preferential access to hyperscale capital and continues to secure clients like SpaceX through allocation control rather than pure benchmark performance. That advantage is real, and AMD’s current forward price-to-earnings ratio of 65 assumes it will narrow. Nevertheless, AMD’s growth challenge remains firmly intact. The data center segment is projected to more than double year over year by 2027, and management has raised its long-term targets rather than lowering them.

Market sentiment reflects cautious neutrality. Reddit sentiment scored a neutral 48 this week, with the most upvoted thread titled “Looking for different perspectives, sell or reduce my position in AMD.” Given the recent stock pullback, that caution is understandable.

SpaceX is a loss worth acknowledging, but it represents a single client within a broader market where AMD has already locked in multi-gigawatt commitments from the firms actively developing frontier AI models. The AI infrastructure race is being contested, which is fundamentally different from one company falling apart.

Contact [email protected] for any questions or corrections.

Read the original
AMD has lost a significant compute allocation… · Slicast