“Technology itself does not decide what the future looks like. People do.” – Lisa Su – CEO of AMD
Technological change never arrives as a purely technical event. A chip, model, platform or network creates possibilities, but its consequences depend on who controls access, which incentives shape deployment, what institutions permit and how workers, customers and citizens adapt. That distinction matters most when computing becomes general-purpose. Artificial intelligence can alter research, manufacturing, healthcare, education and public administration, yet none of those outcomes follows automatically from higher performance. Choices about design, ownership, distribution and accountability determine whether capability becomes broad productivity or concentrated power.
Lisa Su’s position carries particular weight because her career sits at the intersection of engineering and industrial strategy. As chair and chief executive officer of AMD, she has led the company’s transformation into a high-performance and adaptive computing business, with a growing emphasis on AI infrastructure.7 Her public arguments place hardware within a much larger system: processors, memory, networking, software, data centres, developers and customers must work together before an impressive laboratory result can become a dependable service. AMD’s recent strategy therefore presents AI not merely as a model competition, but as an ecosystem and supply-chain contest.
The practical implication is that human agency operates at several levels. Engineers decide which problems to optimise and which constraints to accept. Executives decide where capital is allocated and whether platforms remain open or become tightly controlled. Governments decide how export controls, competition rules, safety standards and public procurement shape markets. Organisations decide whether automated outputs support professional judgement or quietly replace it. Individuals then encounter the consequences through employment, prices, services and access to information. The future is consequently produced by a chain of decisions rather than by an abstract force called technology.
From invention to infrastructure
AI makes this chain unusually visible because advances in software depend on physical resources. Training and running models require accelerators, processors, memory, networking, electricity, cooling and specialised engineering. Su told the US Senate Commerce Committee in 2025 that leadership in AI requires leadership in the infrastructure that powers it, while identifying open ecosystems, domestic semiconductor capacity, talent and a balance between security and adoption as strategic priorities.4 The argument shifts attention away from theatrical demonstrations and towards the less visible decisions that determine who can build and use advanced systems.
AMD’s own product narrative reflects that shift. The company describes an end-to-end portfolio spanning data-centre CPUs and GPUs, PC processors and embedded or adaptive systems, alongside open software and collaboration with partners.1,4 This approach is strategically rational in a market where performance is not the only bottleneck. Customers also need systems that can be procured, integrated, programmed, secured and operated at acceptable cost. An accelerator that cannot be supported by software, networking and skilled people may have impressive specifications but limited economic value.
That does not mean technical capability is secondary. It sets the boundaries of what people can choose. More efficient computing can reduce the cost of inference and enable applications that were previously impractical. Better systems can expand scientific simulation, improve industrial control and assist workers with complex information. Yet increased capability also enlarges the range of possible misuse, from surveillance and fraud to automated discrimination and insecure decision-making. Human choice is therefore not a claim that technology is weak; it is a claim that power must be governed.
The governance problem
Responsible deployment requires more than asking whether a model is accurate in a benchmark. The relevant question is whether the complete socio-technical system performs acceptably in its real setting, including its data, users, incentives, interfaces and escalation procedures. The National Institute of Standards and Technology’s AI Risk Management Framework organises this work around four functions: govern, map, measure and manage.6,14 Those functions imply continuing responsibility before deployment, during operation and after an incident, rather than a single compliance check at launch.
NIST also distinguishes different human roles in AI systems, ranging from fully manual arrangements to systems that act autonomously, defer to experts or provide an additional opinion.3 This distinction is important because the phrase human oversight can conceal very different realities. A person who can technically override a system may lack the time, information or authority to do so. Conversely, requiring approval for every low-risk action can create fatigue and encourage automatic rubber-stamping. Effective agency depends on clearly assigned responsibility, meaningful intervention powers, monitoring and training suited to the consequences of error.
There is a further objection to optimistic accounts of choice: many decisions are made under commercial pressure. A hospital may adopt an imperfect tool because budgets are constrained. A small business may accept a platform’s terms because alternatives are unavailable. A worker may rely on automated scheduling because refusing it risks lost hours. In such conditions, consent is not always equivalent to control. Competition policy, labour protections, public standards and transparent procurement can widen the practical choices available to people, making agency an institutional property rather than merely an individual attitude.
Why the debate matters for AMD
For AMD, the principle has a direct strategic consequence. If AI adoption spreads across industries, the winning position will depend not only on supplying fast silicon but on enabling a diverse community of developers and organisations to build useful applications. Su has argued for open ecosystems because interoperability can reduce barriers to entry, encourage competition and broaden innovation.4 The case is commercially self-interested as well as civic: a wider developer base can increase demand for a platform, while customers are less likely to commit when they fear dependence on a single vendor.
Open ecosystems, however, are not automatically open in practice. Access may still be limited by pricing, scarce capacity, proprietary interfaces, software maturity, energy requirements or shortages of technical staff. A nominally interoperable system can remain difficult to migrate if applications depend on vendor-specific optimisations. The meaningful test is whether customers can switch, audit, adapt and operate systems without prohibitive cost. That is where technical architecture meets market structure, and where claims about empowerment must be tested against evidence.
The same tension appears in predictions about scale. AMD and other industry participants anticipate rapidly rising demand for compute as AI moves from experimentation into enterprise workflows and more autonomous software agents.10,13 Such growth could produce significant productivity gains, but forecasts are not outcomes. Productivity depends on redesigning work, distributing gains, protecting quality and ensuring that people possess the skills to use new tools. If investment flows mainly into infrastructure while training, safety and access lag behind, the resulting future may be technologically advanced but socially narrow.
The most durable reading is therefore neither technological determinism nor human exceptionalism. Machines can reshape incentives, capabilities and institutions so profoundly that old forms of decision-making become impossible or uneconomic. People still decide which systems to build, who receives their benefits, whose risks are tolerated and what remedies exist when systems fail. That responsibility extends from semiconductor design and supply-chain policy to the final workplace interface. Treating those choices as unavoidable would itself be a choice, and usually one that favours whoever already controls the infrastructure.
References
1. AMD and its Partners Share Their Vision for ‘AI Everywhere, for … – 2026-01-06 – https://newsroom.amd.com/news/amd-and-its-partners-share-their-vision-for-ai-ev/
2. Artificial Intelligence Risk Management Framework (AI RMF 1.0) – https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf
3. App. C: AI Risk Management and Human-AI Interaction – AIRC – 2026-06-18 – https://airc.nist.gov/airmf-resources/airmf/appendices/app-c-ai-risk-management-and-human-ai-interaction/
4. Preserving U.S. Leadership in the Race for AI – AMD – 2025-05-08 – https://www.amd.com/en/blogs/2025/preserving-us-leadership-in-the-race-for-ai.html
5. AMD CEO Lisa Su Declares ‘AI Is for Everyone’ in CES 2026 With Guests from OpenAI, Luma AI, Liquid AI, World Labs and More – 2026-01-05 – https://www.techtimes.com/articles/313772/20260105/amd-ceo-lisa-su-declares-ai-everyone-ces-2026-guests-openai-luma-ai-liquid-ai-world-labs.htm
6. AI RMF – AIRC – 2026-06-18 – https://airc.nist.gov/airmf-resources/airmf/
7. Dr. Lisa Su – AMD Chair and Chief Executive Officer – 2026-06-26 – https://www.amd.com/en/corporate/leadership/lisa-su.html
8. AMD in the AI Era: CEO Lisa Su on Product Innovation … – https://www.sixfivemedia.com/content/amd-in-the-ai-era-ceo-lisa-su-on-product-innovation-leadership-and-the-future
9. AI Risk Management Framework – Palo Alto Networks – https://www.paloaltonetworks.com/cyberpedia/ai-risk-management-framework
10. AMD Lisa Su CES Las Vegas 2026 – 2026-01-06 – https://www.rev.com/transcripts/amd-at-ces-2026
11. Lisa Su on AMD’s Strategy for Growth and the Future of AI – https://time.com/7026241/lisa-su-amd-ceo-interview/
12. [PDF] Artificial Intelligence Risk Management Framework: Generative … – https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
13. AMD CEO On ‘Next Phase Of AI’ Investments In EPYC, Ryzen … – 2026-01-22 – https://www.crn.com/news/components-peripherals/2026/amd-ceo-on-next-phase-of-ai-investments-in-epyc-ryzen-gpus-and-partners-in-2026
14. Executive Summary – AIRC – NIST AI Resource Center – 2026-06-18 – https://airc.nist.gov/airmf-resources/airmf/0-ai-rmf-1-0/
15. Framing the Risk Management Framework: Actionable Instructions by NIST in The “Measure” Section of the AI RMF – 2023-04-13 – https://epic.org/framing-the-risk-management-framework-actionable-instructions-by-nist-in-the-measure-section-of-the-ai-rmf/
