A conversation with Uma Mondal, Engineering SVP and AMI’s first Fellow, and Earle Philhower, Product Marketing Manager for AMI MegaRAC.
By 2030, the industry is on pace to spend nearly $7 trillion building out AI data center infrastructure, according to McKinsey. Every dollar of that build-out eventually lands on hardware that must boot, secure itself, and report its health before a single workload runs, and that job belongs to firmware. At AMI, a Lattice Company – which turned 40 last year – that job has quietly moved from background utility to load-bearing infrastructure.
To help give perspective, two people at very different points in their AMI careers – Uma Mondal, the company’s first Fellow and a 20-plus-year AMI veteran, and Earle Philhower, several months into the new role as Product Marketing Manager for AMI MegaRAC® Manageability Solutions – have shared their experiences working at AMI during this pivotal shift.
The View from a Newcomer
Before joining AMI in 2026, Earle Philhower spent his career in data storage technologies at companies including Intel, HGST and KIOXIA, where he was familiar with AMI’s work.
“AMI is kind of the gold standard for the boot and management of high-level storage systems. The storage systems we developed at my other companies were all controlled by servers and systems driven by AMI firmware. They had a great reputation in terms of stability and interoperability,” said Earle.
“I think the first time I used a server with AMI MegaRAC with an HTML5 KVM I nearly cried; it was such an improvement over some other proprietary ones,” he added.
What surprised him the most after joining AMI? Finding how long AMI had been working behind the scenes in the open source community. He also pushed back on the idea that open code invites easy competition. “The source code is not the entire product. The product for us is the code, the engineering support, and really the know-how and muscle memory of 40 years of doing this kind of stuff,” he said.
From One Optimized Box to a Rack Full of Everything
Earle’s sharpest read on why AMI’s 40+ year tenure matters now is architectural. “Earlier, your point of optimization was a single server with maybe a couple of CPU sockets, a couple of NICs, and maybe an accelerator,” he said. “Right now, we’re in the middle of a seismic shift in terms of how compute and networking is built and delivered. We’re seeing new CPU architectures in the data center like ARM and RISC-V, networking with 400Gbit+ Ethernet, NVLink, and even services built on top of it, like CXL.”
These technologies are enabling a new class of compute unit at rack-scale, where individual units outclass entire HPC clusters from not too long ago. Each new complex, heterogeneous system needs firmware for every component, with updates every six to 12 months as new silicon arrives and maintained with security patches for the rack’s five-to-seven-year lifetime.
AMI’s answer is a unified source-tree approach that covers the full firmware stack, from its AMI Aptio V UEFI BIOS to its AMI MegaRAC BMC firmware, so a fix or security update reaches every supported platform through one release path. Combined with partnerships with every major CPU and system designer, and early access to customer reference boards for new silicon, this lets AMI move a design from reference to customer platform in a significantly shorter time.
Forty Years of Institutional Memory
Uma Mondal’s career grew along with AMI. He was part of the team that designed the first MegaRAC hardware for Dell, working with a team of engineers that included AMI CEO Sanjoy Maity, many of whom now sit in AMI’s senior leadership. Speaking about that lineage, he said, “It says that AMI is fundamentally an engineering-driven company, one built by people who have lived through real failures, successes and silicon cycles.”
That experience and depth produced one of AMI’s more distinctive inventions, one which Uma was closely involved in. When Intel released Platform Firmware Resilience using an expensive FPGA, customers balked at the added cost. In response, Uma and his colleagues proposed implementing Root of Trust on the BMC’s unused secondary processor, “completely separate from the Main BMC OS,” using its existing crypto engine. Intel ultimately declined to put the required boot feature on its server roadmap, so the original product never shipped as planned. But the effort morphed into AMI’s current line of AMI Tektagon® security solutions, providing AMI with hardware-level Root of Trust knowledge that shapes its security architecture today.
Uma noted that particularly during the early years, much of AMI’s institutional knowledge was never written down. “So, the way we transfer it is through hands-on debugging, walking engineers through real issues and explaining the history behind decisions,” he said. AMI backs that up with its Engineering Information Portal (EIP), a searchable log of issues and resolutions running since roughly 2000, which the company has now paired with AI-based search in its AMI AMILiA™ solution, to more easily pass that deep experience and expertise to its customers.
Firmware’s New Job Description
Both agree that the role of firmware has fundamentally changed. In Uma’s words, firmware “is no longer passive; it is proactive.” It has become the root of trust for platform security, the control plane for power, thermals and system, and the integration layer that makes complex silicon and high-speed interfaces function as a cohesive system. Modern platforms combine PCIe Gen5 and Gen6, NVMe at scale, emerging standards like CXL, multi-host architectures, and increasingly sophisticated system management layers – all of which the BMC must orchestrate.
Earle sees the same shift from the market side. “From the outside, it looks like firmware’s done and there is nothing to innovate,” he said. But he doesn’t believe a single engineer at any server or GPU vendor thinks firmware is a solved issue. New standards keep arriving, rack generations keep turning over, and regulations like the EU’s Cyber Resilience Act now require years of ongoing security support regardless of whether the underlying technology changes.
A Place to Learn the Fundamentals
Put Earle and Uma’s accounts side by side and a single thread emerges: that firmware has moved from invisible plumbing to the control plane for AI factories, and AMI’s 40-year run and institutional memory have helped it earn its position as a leader in the firmware space. For new engineers weighing where to build a career, Uma advises that AMI is a great starting point to learn the fundamentals. “If you want to learn something about hardware or low-level software, this is the place,” he said. “With PCIe, CXL, GPUs and multi-host systems all converging in one BMC, this is where the real hard work is being done, and difficult engineering problems are being solved.”
As the industry continues to invest heavily in AI infrastructure over the next several years, AMI’s commitment to excellence is what keeps it booting. And with Lattice Semiconductor Corporation having successfully completed of its acquisition of AMI, AMI moves to a new phase in its trajectory to serve the critical convergence of hardware and firmware required by hyperscalers, OEMs, ODMs and neocloud providers to address the escalating demands of modern AI infrastructure.
