Two years ago, the AI conversation was mostly about chatbots. Today it is about power grids, chip supply, cloud contracts and who gets to run the world's AI workloads at the lowest cost. The people steering that shift are not a single club. They run very different companies, and they disagree on plenty. But five of them, Jensen Huang, Rene Haas, Andy Jassy, Satya Nadella and Dylan Field, each hold a piece of the puzzle. This is a plain-English look at what each one is doing in October 2026, and what it adds up to for the rest of us.
The AI Story Has Moved From Software to Infrastructure
For a while, AI looked like a software race. Whoever had the cleverest model won the headlines. That has quietly changed. Models still matter, but the bottleneck now sits underneath them: chips, memory, electricity and data center space. Industry estimates cited this year put big-cloud spending on AI infrastructure in the hundreds of billions of dollars for 2026 alone, far above 2025.
That is why this list of names makes sense. One builds the dominant AI hardware. One designs the processor architecture inside a huge share of the world's devices and servers. Two run cloud platforms that are racing to build their own silicon. And one leads the design platform where many of the resulting products get shaped. Read together, they show how the AI economy is actually being assembled.
Jensen Huang: Selling the “AI Factory”
Nvidia's founder and CEO has spent the past year repeating one idea in different ways: AI is not a feature, it is infrastructure. At CES 2026, he argued that companies are not printing new money to buy Nvidia systems. They are redirecting budgets that used to go to conventional servers into what he calls AI factories. It is a clever reframing, because it turns a scary capital bill into a replacement cycle.
At GTC in March, Nvidia introduced its Vera Rubin rack-scale platform, and Huang told the audience that revenue visibility for its current and upcoming AI systems could reach at least $1 trillion through 2027. On the February earnings call he said “compute equals revenues now,” his shorthand for why he thinks the spending is sustainable. Nvidia's market value was reported at around $4.3 trillion in early August.
Two other threads are worth watching. The first is “sovereign AI,” where governments fund their own national computing capacity. The second is physical AI, meaning robots and machines that act in the real world. Both widen the market beyond the usual cloud giants. The caution is that export rules on advanced chips and the push by customers to build their own silicon could trim Nvidia's lead over time. Huang's answer so far has been to keep selling complete systems rather than lone chips.
“ The AI race is no longer won by the best model alone. It is won by whoever can run it cheaply, at scale, every day. — The Empire Insight Editorial Board
Rene Haas: Arm Steps Out of the Phone and Into the Data Center
Arm built its reputation in smartphones, where its low-power designs are everywhere. Under CEO Rene Haas, the company is deliberately moving toward cloud and AI data centers. He has said as much, describing a shift away from being known mainly for phones. The logic is simple: when data centers run into power limits, efficiency becomes the thing everyone wants.
Haas's most interesting argument is about the humble CPU. As AI shifts from single prompts to agents that run continuously, someone has to manage memory, call tools, query databases and keep the whole system moving. That work mostly lands on CPUs, not GPUs. Arm says that makes processor cores more valuable, not less, and that the world will need several times more cores within the same power budget.
The numbers Arm shared this year back that story up. The company said data center royalty revenue more than doubled year on year in its latest quarter, that shipments of its Neoverse cores passed 1.5 billion, and that customer demand for its new Arm AGI CPU has gone past $2 billion across fiscal 2027 and 2028. That last product matters for another reason. It signals that Arm is not only licensing designs anymore but also offering finished silicon, a real change in its business model. Haas has said partners he checked with were supportive, though it is fair to note that some customers may still feel uneasy about a supplier moving closer to being a competitor.
Arm's own executives also point out that over half of the new cloud computing capacity added in the past twelve months ran on Arm-based processors. Amazon's Graviton and the in-house chips at Google and Microsoft all build on the architecture. In other words, Haas does not need to beat the cloud giants. He just needs to be inside what they build.
Andy Jassy: Amazon's $200 Billion Bet on Its Own Chips
Of the five, Andy Jassy has made the boldest public wager in dollars. Amazon said it plans roughly $200 billion in capital spending in 2026, mostly aimed at AI and infrastructure. In his April shareholder letter he explained the thinking: AWS's AI business had reached a $15 billion annual run rate, and he said a large share of the planned spending already had customer commitments behind it.
The heart of the plan is custom silicon. Amazon's Graviton CPUs and Trainium AI chips are being positioned as a cheaper path than relying only on Nvidia. Jassy said Trainium should save Amazon tens of billions of dollars in capital spending each year and give it a margin edge on inference. He also said the chips business was running at more than $20 billion a year and growing at triple-digit rates.
Demand shows up in the deals. In April, Amazon agreed to invest up to $25 billion more in Anthropic, which in turn committed to spend over $100 billion on AWS over ten years and secured access to as much as 5 gigawatts of capacity on Amazon's custom chips. Reports also describe a Trainium commitment from OpenAI beginning in 2027. Amazon has also deepened its relationship with Nvidia, so this is not a simple either-or story.
The risk is obvious. A $200 billion year only pays off if customers keep showing up and if the chips deliver the promised savings. Jassy's reply is that the cloud industry has been through a similar build-then-fill cycle before, and he likes how it turned out.
Satya Nadella: Microsoft Wants More Choice, Not Fewer Suppliers
Microsoft's approach under Satya Nadella is best described as “build your own, but keep buying.” The company launched its Maia 200 inference chip in January and, on October 12, unveiled Maia 300, its third-generation design. Microsoft says Maia 300 should deliver about 40% more AI output per megawatt than leading GPU-based systems, with deployments starting in 2027. Nadella described it publicly as the company's biggest bet yet on its own silicon.
The reasoning is cost. Microsoft has told investors that running OpenAI and in-house models on Maia 200 lowers operating costs by roughly 30% to 40% compared with top-end Nvidia hardware, according to reporting this summer. Maia 200 already supports Copilot, and GitHub Copilot and Microsoft Foundry are expected to follow by year-end.
But Nadella has been clear that this does not mean Microsoft is leaving Nvidia or AMD. The company's fleet is deliberately mixed. Analysts also note that Maia is expected to represent only a modest slice of Microsoft's total AI chip spending in early 2027, even though capital spending is reported to run above $175 billion a year. So the practical story is leverage. Having a credible in-house option improves Microsoft's bargaining position and its margins, without betting everything on one design.
Nadella has also raised an issue many CEOs avoid: trust. He has suggested treating powerful AI models as potential insider risks, assuming they could make mistakes or be compromised, and building containment from the start. For businesses deploying AI agents, that is a useful and slightly uncomfortable piece of advice.
Dylan Field: The Layer Where Products Get Their Personality
Dylan Field sits at the opposite end of the stack. Figma makes no chips. It makes the software where teams design and now increasingly build products. That is exactly why his view is valuable. If AI makes code cheap and fast to produce, what separates one product from another?
Field's answer is design, taste and a clear point of view. Speaking at Config India in Bengaluru this month, he said design is increasingly what sets products apart as code becomes commoditized, and that he does not expect AI to cause designer job losses. He also noted that India is now Figma's second-largest market by daily usage.
The company's numbers suggest AI is helping rather than hurting. Figma reported second-quarter revenue of $370 million, up 48% from a year earlier, with growth accelerating for a third straight quarter. Its prompt-to-app tool, Figma Make, can now work directly inside production codebases, and the company reports that more than 80% of larger paid customers were using its AI credits weekly by the end of June. Figma has also been opening its platform to outside AI agents through MCP, with new partners added this month.
It is not all smooth. Figma said it is searching for a new CTO, with its AI and editor engineering teams reporting directly to Field in the meantime. Still, his central bet is worth taking seriously: when everyone can generate software, the winners will be the teams with better judgment about what to build.
“ When anyone can generate code, the scarce skill becomes knowing what is worth building. — The Empire Insight Editorial Board
Where These Five Leaders Agree and Disagree
Put side by side, a few patterns stand out.
- Efficiency is the new currency. Haas talks about performance per watt, Nadella about output per megawatt, Jassy about cost per inference. Even Huang's pitch rests on getting more work out of every unit of power.
- Nobody wants a single supplier. Amazon and Microsoft are building chips while still buying from Nvidia. Arm sits underneath many of them. The market is becoming layered, not winner-takes-all.
- Agents change the hardware story. Always-on AI agents lean heavily on CPUs, memory and networking, which is why Arm, Amazon and Nvidia are all talking about them.
- The money has to show up. Huang says demand converts to revenue. Jassy points to customer commitments. Field points to paying customers using AI credits. The skeptics want proof, and so far each leader is offering numbers rather than promises.
Where they differ is on who captures the value. Huang believes it flows to integrated systems. Jassy and Nadella believe owning more of the stack improves margins. Haas wants to be the common foundation under everyone. Field believes value drifts up to design and product thinking. All four can be partly right at once, and that is what makes the next two years interesting.
What This Means for Businesses and Small Teams
You do not need a data center to benefit from any of this. The competition among these companies tends to push the price of AI down and the quality up. A few practical takeaways:
- Expect cheaper AI over time. Custom chips and heavy competition should reduce the cost of running AI tools, but plan for price changes rather than assuming they are fixed.
- Avoid locking yourself in too early. With so many platforms racing, flexible contracts and portable workflows are safer than a long commitment to one vendor.
- Invest in judgment, not just tools. If software gets easier to produce, your edge comes from understanding your customers and having a clear point of view.
- Treat AI agents with care. Give them limited access, review what they do and keep humans in charge of sensitive decisions.
- Watch energy and capacity news. Power availability is becoming a real limit on AI growth, and it will shape pricing and availability.
The Bottom Line
Nobody can say for certain how this plays out. Spending is enormous, forecasts are bold, and some of the numbers come from the companies themselves, so a healthy dose of skepticism is reasonable. But the direction is clear. The AI economy is being built layer by layer, from silicon to cloud to the design tools that shape the final product, and each of these five leaders is pulling hard on a different layer. The smartest move for everyone else is to keep learning, stay flexible and pay attention to the economics underneath the headlines.
Frequently Asked Questions
Who are the key leaders shaping AI infrastructure in 2026?
Jensen Huang of Nvidia, Rene Haas of Arm, Andy Jassy of Amazon, Satya Nadella of Microsoft and Dylan Field of Figma each influence a different layer, from chips and cloud to design software.
Why are Amazon and Microsoft building their own AI chips?
Mainly to lower costs, reduce dependence on a single supplier and improve margins. Both still buy chips from other vendors, so the goal is more choice rather than full replacement.
What is Arm's role in the AI boom?
Arm designs the processor architecture that many cloud and device chips are based on. It is expanding into data centers and has launched its own AGI CPU as AI agents increase demand for efficient processors.
How is Jensen Huang shaping the AI economy?
Through Nvidia's complete AI systems, including the Vera Rubin platform, and by arguing that AI spending is a redirection of existing infrastructure budgets toward “AI factories.”
Where does Figma fit into AI strategy?
Figma focuses on the product layer. Its tools help teams design and build with AI, and CEO Dylan Field argues that design and taste become the main differentiators as code gets easier to generate.
Key Takeaways • AI Leaders in 2026
AI has shifted from a software contest to an infrastructure race built on chips, power and cloud capacity.
Amazon and Microsoft are building custom chips to cut costs, while still buying from Nvidia and AMD.
Arm is moving from phones into data centers as AI agents increase demand for efficient CPUs.
Figma's bet is that design and judgment matter more as code becomes cheaper to produce.
Efficiency, not hype, will decide who wins the next phase of the AI economy.
AI Infrastructure at a Glance
Figures reported by the companies and the press in 2026Bottom Line: “The next AI winners will be the ones who make intelligence cheaper to run.”