AI Infrastructure Strategy: Chloe Ma of Epic Semi
Chloe Jian Ma — Chief Business Officer, Epic Semi
Chloe Ma is Chief Business Officer at Epic Semi, a leader whose career spans Cisco, Juniper Networks, Mellanox, Intel, SiFive and Arm. Her central message is that AI is moving from training models to executing them, so businesses now need complete, deployment-ready AI infrastructure that is efficient, open, and under their own control, rather than a pile of individual chips.
Chloe describes her career as following the evolution of infrastructure itself. She began at the peak of the dot-com bubble at Cisco, the networking giant of the internet era, and then moved through Juniper. She then saw infrastructure leadership shift to Amazon Web Services, Google Cloud, and Microsoft, which increasingly worked directly with semiconductor companies.
That shift took her to Mellanox, a chip and module company where she could engage hyperscalers directly. A desire to understand different workloads, from big data analytics and databases to storage and emerging machine learning, then led her to compute companies, starting with Intel. She later joined Silicon Valley startup SiFive, where she met the founders of Epic Semi, and spent about four years at Arm building Edge AI solutions for chip companies. She joined Epic Semi about a year before the interview.
What drew her in was the chance to combine everything she had built into a full-stack, end-to-end AI infrastructure offering, and a mission she says is set by the CEO's vision: helping industries and nations create and deploy technology under their own control.
According to Chloe, building a proof of concept is now easy. She mentions hiring a college-freshman intern over the summer to build AI agents for marketing and sales efficiency. The harder question is keeping such agents running reliably and integrated with existing work.
She points out that not everything is token generation by a large language model. Agentic tasks still need to query traditional databases and open familiar files such as PowerPoint and Word, and GPUs cannot do this alone. Software integration with existing workflows, plus security and authentication, is a big gap between a flashy demo and production deployment.
On the hardware side, she says GPUs and other accelerators are not easy to deploy, manage, and integrate with existing IT infrastructure. Companies that want data sovereignty and on-premises deployment often lack IT teams sophisticated enough to extract performance from a heterogeneous architecture.
Chloe names three expectations. First, businesses rarely have time to stitch together many small components, so they demand a complete, deployment-ready system for running applications and agents and experimenting with new ideas.
Second, the measurement is changing. In her view, token throughput per GPU is not the point; what matters is the number of completed tasks per dollar of total cost of ownership (TCO), judged against an end-to-end business problem such as increasing European sales.
Third, businesses want more choice and control. She calls Nvidia a great company, but says relying on a single vendor worries many businesses, who want options that prevent lock-in.
Chloe says she had spent little time in the Middle East and GCC region before joining Epic Semi, but was fascinated by what she observed. She notes that Silicon Valley AI companies made their first big revenues there, and that the US took in Saudi Arabia and the UAE, accompanied by prominent CEOs and major deals.
She identifies several regional advantages: A national ambition to become a significant player in AI infrastructure. Abundant oil reserves that can be turned into electricity, giving low power costs compared with California, Europe, or Japan, which matters as AI raises power-supply concerns. Capital increasingly directed at AI data centers rather than only US companies and real estate. A desire to preserve language and culture, reflected in training their own Arabic language models.
Chloe also says that security concerns, heightened by the war, strengthen the push for sovereignty, even though the conflict caused some disruption. She describes three levels of sovereignty: data and model sovereignty, infrastructure sovereignty, and operational sovereignty, meaning the ability to run purchased equipment without complete reliance on a third party.
Chloe frames Contrail AIX as Epic Semi's answer to the shift from the AI training era to the AI execution era. She says that Nvidia CEO Jensen Huang validated the vision: the agentic era needs CPUs, other inference accelerators, and GPUs.
The Contrail AIX server CPU combines traditional CPU processing and AI acceleration on one chip with shared memory and storage, so an agentic workload can move easily between tasks like opening a PowerPoint or querying a database and LLM reasoning. Public product information from the company describes 32 RISC-V CPU cores, 16 AI acceleration cores, and up to 150 TOPS (INT8) in a unified architecture.
The platform is built on RISC-V, an open architecture. Chloe contrasts it with the proprietary Nvidia GPU, x86, and Arm architectures, and notes that it is governed by a nonprofit in Switzerland. RISC-V was founded in 2015 as the RISC-V Foundation and is incorporated today as the RISC-V International Association in Switzerland. She says Epic Semi happens to be the first with a real server-grade CPU-plus-AI compute platform on it. Her point is that any country or company with talent, investment, and desire can build chips on the architecture, for example, Brazil, Malaysia, or Germany, without changing the software on top.
The discussion closes on the idea that the real test of AI is whether businesses can use it for meaningful work at the right cost and with the right level of control. As the host put it, AI is moving rapidly from experimentation toward execution, and the infrastructure underneath is becoming as important as the AI itself. Listeners can connect with Chloe on LinkedIn or visit Epic Semi at epicsemi.com.
Comments
No comments yet — be the first to share your thoughts.