ENYO™SEMICONDUCTOR ← All insights
Insights · Own the Customer, Own the Code · 2010–2019 · 12 Nov 2026 · 2 min read

Part 4: The Customer Builds Its Own Chips

In this decade the largest chip buyers decided to design their own chips. Amazon, Google, Microsoft and Apple all built in-house silicon teams, and the fastest way to build such a team was to buy a startup.

Nov 12, 2026

Part 4 · 2010–2019 · The Customer Builds Its Own Chips

In this decade the largest chip buyers decided to design their own chips. Amazon, Google, Microsoft and Apple all built in-house silicon teams, and the fastest way to build such a team was to buy a startup.

Thanks for reading! Subscribe for free to receive new posts and support my work.

Amazon acquired Annapurna Labs in 2015. Intel bought Nervana and Movidius in 2016, and Habana Labs in 2019 for about $2 billion. The pattern repeated across the industry, as startups that solved a specific problem for a large customer became that customer’s internal chip team.

Two forces drove this shift, and we see both of them at work today. The first was physics. In the product data we reviewed, the transistor count of the largest chips doubled roughly every two years in the 2000s, and that slowed to nearly three years in the 2010s. General-purpose chips stopped improving fast enough on their own, and specialised chips for artificial intelligence, networking and storage filled the gap.

The second force was demand. Cloud companies and phone makers knew exactly what their workloads needed, and because they owned the customer and the data, they could define the chip. A startup that matched their need had a clear buyer from the first day.

Many of these teams came from Israel, including Annapurna, Habana and Anobit. Israel’s chip startups did not own factories. They had experienced engineers, close links to US buyers and a strong habit of starting companies, and they built a global position through design alone. That is a model we should study closely.

Bangladesh’s telecom sector offers a parallel that is close to home. Operators do not buy chips directly, but they buy network equipment, home gateways and smart meters in large volumes, and they set the specifications. Whoever owns the demand can shape the chip, as the cloud companies did.

For Bangladesh the lesson is that we should start from a real buyer. A design team that solves a defined problem for an operator, a utility or an equipment maker has a path to revenue. A design team without a buyer has only a technology.

Thanks for reading! Subscribe for free to receive new posts and support my work.

Get new Enyo analysis by email. Subscribe on Substack