Nvidia PE ratio hits 7-year low as markets sell off

Nvidia shares are trading at about 19.6 times expected earnings over the next 12 months, the chipmaker’s lowest forward price-to-earnings (PE) multiple since early 2019, as global markets sold off on fears of war in the Middle East and higher inflation.

The stock is down nearly 20% from its record close in October, fell 2.2% on March 27, and is on track to drop about 10% in Q1.

Investors have also questioned whether heavy AI infrastructure spending by customers such as Microsoft, Alphabet, and Amazon will take longer than expected to translate into higher profits.

The decline has cut more than US$800 billion from Nvidia’s market value to about US$4 trillion, and proprietary trader Dennis Dick at Triple D Trading said fast-moving AI shifts raise the risk that today’s hardware leaders could be disrupted.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Custom silicon is putting pressure on Nvidia

A shift is underway as Nvidia customers build in-house or custom chips, often called custom silicon, meaning chips designed for a company’s own needs rather than bought off-the-shelf.
Alphabet trains its most advanced model, Gemini 3, entirely on internally designed Tensor Processing Units (TPUs), which are AI chips Google built to run its machine-learning workloads 1.
Building this hardware in-house, meaning a company makes key components itself rather than buying them from suppliers, gives Alphabet AI compute at an estimated 20% of the cost of buying high-end Nvidia GPUs 1.
Anthropic plans to use up to 1 million Google TPUs, while Meta is reportedly in talks for a multi-billion dollar deal to use them in its data centers 1.
Broadcom is speeding the trend by co-designing custom chips for Google and others, and forecasts put it at a 60% share of the AI server compute application specific integrated chip (AISC) market, a segment for custom AI chips built for servers, by the end of 2027 2.

AI competition now hinges on power use and full-stack control

More than one hardware route now exists, with Google’s Gemini on custom TPUs competing against Microsoft’s partnership with OpenAI on its Azure cloud 1.
Winning now comes from tuning the full stack of hardware, software, and models rather than relying on the strongest chip alone.
The biggest constraint is moving from chip supply to basics such as electricity availability and the time needed to build new data centers 3.
Hyperscalers, the largest cloud providers such as Amazon, Microsoft, and Alphabet, are set to spend about $700 billion on infrastructure this year, yet AI demand may still outrun supply through 2027 because of physical limits 3.