Altera’s Agilex 2X puts programmable chips to work on AI and robotics

Altera has announced Agilex 2X, a new family of programmable chips aimed at AI infrastructure, robotics and other systems that need to process data quickly. The notable split: software support is available now, but the first Agilex 27 engineering samples are not expected until the second half of 2027.

Watch Desk analysis

What happened

SiliconANGLE reports that the portfolio includes four families: Agilex 27 for intensive data processing and radio-frequency workloads; Agilex 25 for uses including drones and low-latency sensors; Agilex 23 for robotics, industrial automation and automotive systems; and Agilex 22 for functions such as motor control and bridging interfaces.

The first family, Agilex 27, is designed for high throughput and bandwidth. Altera claims a 1.8-times improvement in performance per watt over Agilex 7, and says its HyperFlow on-chip network is designed to move multiple terabits of data per second. Those are company claims, not customer results. SiliconANGLE’s report also says support is available in Quartus Prime 26.3. QuartusIQ, Altera’s AI development companion, is available through Quartus Prime Pro Edition 26.3.

Why it matters

AI hardware is not only a question of how much raw computing a system can deliver. Robots, drones and industrial equipment also have to gather and move sensor data, often within tight limits on power, size and response time. Programmable chips can be adapted to handle those changing jobs, rather than locking every function into a fixed design.

For developers, the other bet is that AI-assisted engineering could make programmable hardware less daunting to use. Altera reports faster design cycles with QuartusIQ in internal benchmarks, but engineers still have to validate designs. The commercial question is whether the software helps more teams reach working hardware, not merely whether it can make a promising demo look brisk.

Our read

This is a substantive roadmap for computing that connects AI and robotics to real hardware, not just a fresh chip name on a slide. But much of the product family remains a promise for now: Quartus support is here, while Agilex 27 samples are expected in late 2027. The performance figures will matter most when customers can test complete systems under real workloads.

What to watch

• Whether Agilex 27 engineering samples arrive in the expected second half of 2027.
• How the performance-per-watt claim holds up in customer systems.
• Whether QuartusIQ helps engineers build and validate designs faster in practice.

Discussion spark: For AI and robotics systems, is adaptable programmable hardware worth the extra design work, or should developers favour fixed-function chips wherever possible?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.