How AMD's 11k engrs in India shape its next chips
AMD has more than 11,000 engineers in India, working across every business of the chipmaker. Their brief reaches into computing’s toughest problems: making processors faster while squeezing them into less space and holding down electricity consumption. India is also home to one of AMD’s five CPU design centres worldwide; the other four are in the US.
The latest results of this engineering work are headed into products used all over the globe. Local teams have delivered CPU designs and chiplets (the smaller pieces assembled into a processor) for advanced manufacturing processes. These will go into forthcoming EPYC server and PC products scheduled for late 2026 and early 2027. Earlier this year, they also delivered chiplets for Instinct, AMD’s chips for demanding AI calculations.
EPYC processors power servers, the computers behind cloud services and business applications, including offerings from AWS and Microsoft Azure. Ryzen is AMD’s processor family for personal computers, appearing in laptops from Dell, HP, Lenovo, and others. The engineering choices made here therefore matter both to someone working on a laptop and to companies running large data centres.
Across those markets, the challenge is to make each new generation do more within tight physical and electrical limits. “With every generation, we challenge ourselves to bring down power consumption, make the area more compact, and still keep the performance high enough for applications,” says Jaya Jagadish, country head and senior vice president of engineering at AMD India.
That balancing act is becoming harder. For years, improvements in chip manufacturing helped make designs smaller and more efficient. Those gains are becoming harder to extract, Jagadish says, putting more responsibility on engineers to rethink how a processor is put together.
One answer is to divide it into chiplets. The parts doing the most demanding calculations can use the newest manufacturing technology, while components handling connections and moving data can use older, cheaper processes. This lets designers spend on advanced manufacturing where it delivers the greatest benefit.
AMD’s team in India works on bringing these differently manufactured pieces together in a single package. “We then connect these chiplets and make them work like a single chip,” Jagadish says. Making that combination function requires close collaboration with colleagues in the US.
The same search for efficiency extends inside the CPU itself. Customers want different things from its cores: some prioritise speed, others need a compact design or cannot afford its power consumption. Engineers must decide which features a particular application needs and where they can save space and energy.
One area handled by the Indian team is what AMD calls a “lean core”. Jagadish says engineers take a main core and reduce its area by 50%, significantly cutting power consumption, then work on features to improve performance. The target is enough performance for the intended task within a smaller, more economical design. Some customers also seek specific security features.
Turning such decisions into a working product takes several kinds of engineering. While US teams define the broad architecture, colleagues here assess whether those ideas can be realised, develop the detailed design and prepare it for manufacture. On Ryzen AI and server chips, they lead substantial parts of the work that brings the processor’s different functions together.
That includes arranging the circuitry, checking that the design behaves as intended and building in features that allow manufactured chips to be tested for defects.
Such responsibility took years to build. AMD’s Indian operation began more than two decades ago. Jagadish recalls starting the CPU group with about 25 people, initially checking designs and handling some implementation work. Finding experienced processor engineers was difficult, so AMD began developing the skills itself. Over time, the group added disciplines and took responsibility for complete core development.
“We started hiring talented engineers with a strong aptitude for learning and training them ourselves,” she says. “This was not something that had been passed down to us; it was something we were creating.”
Now AI is changing how that workforce operates, but Jagadish expects the engineering problems to become more demanding as the tools improve. The ambition is to tackle problems that cannot be solved today. That makes the accumulated knowledge behind those chip designs especially valuable.
“Nobody can replace the expertise you develop in core engineering,” she says. “Only when you have that foundation can AI be highly productive for you.”
The latest results of this engineering work are headed into products used all over the globe. Local teams have delivered CPU designs and chiplets (the smaller pieces assembled into a processor) for advanced manufacturing processes. These will go into forthcoming EPYC server and PC products scheduled for late 2026 and early 2027. Earlier this year, they also delivered chiplets for Instinct, AMD’s chips for demanding AI calculations.
EPYC processors power servers, the computers behind cloud services and business applications, including offerings from AWS and Microsoft Azure. Ryzen is AMD’s processor family for personal computers, appearing in laptops from Dell, HP, Lenovo, and others. The engineering choices made here therefore matter both to someone working on a laptop and to companies running large data centres.
Across those markets, the challenge is to make each new generation do more within tight physical and electrical limits. “With every generation, we challenge ourselves to bring down power consumption, make the area more compact, and still keep the performance high enough for applications,” says Jaya Jagadish, country head and senior vice president of engineering at AMD India.
That balancing act is becoming harder. For years, improvements in chip manufacturing helped make designs smaller and more efficient. Those gains are becoming harder to extract, Jagadish says, putting more responsibility on engineers to rethink how a processor is put together.
One answer is to divide it into chiplets. The parts doing the most demanding calculations can use the newest manufacturing technology, while components handling connections and moving data can use older, cheaper processes. This lets designers spend on advanced manufacturing where it delivers the greatest benefit.
AMD’s team in India works on bringing these differently manufactured pieces together in a single package. “We then connect these chiplets and make them work like a single chip,” Jagadish says. Making that combination function requires close collaboration with colleagues in the US.
The same search for efficiency extends inside the CPU itself. Customers want different things from its cores: some prioritise speed, others need a compact design or cannot afford its power consumption. Engineers must decide which features a particular application needs and where they can save space and energy.
One area handled by the Indian team is what AMD calls a “lean core”. Jagadish says engineers take a main core and reduce its area by 50%, significantly cutting power consumption, then work on features to improve performance. The target is enough performance for the intended task within a smaller, more economical design. Some customers also seek specific security features.
Turning such decisions into a working product takes several kinds of engineering. While US teams define the broad architecture, colleagues here assess whether those ideas can be realised, develop the detailed design and prepare it for manufacture. On Ryzen AI and server chips, they lead substantial parts of the work that brings the processor’s different functions together.
That includes arranging the circuitry, checking that the design behaves as intended and building in features that allow manufactured chips to be tested for defects.
Such responsibility took years to build. AMD’s Indian operation began more than two decades ago. Jagadish recalls starting the CPU group with about 25 people, initially checking designs and handling some implementation work. Finding experienced processor engineers was difficult, so AMD began developing the skills itself. Over time, the group added disciplines and took responsibility for complete core development.
“We started hiring talented engineers with a strong aptitude for learning and training them ourselves,” she says. “This was not something that had been passed down to us; it was something we were creating.”
Now AI is changing how that workforce operates, but Jagadish expects the engineering problems to become more demanding as the tools improve. The ambition is to tackle problems that cannot be solved today. That makes the accumulated knowledge behind those chip designs especially valuable.
“Nobody can replace the expertise you develop in core engineering,” she says. “Only when you have that foundation can AI be highly productive for you.”
Next Story