AMD (AMD.US) Surpasses $1 Trillion Market Cap and Begins Competing for "Computing Power Definition Rights"! Officially Announces $8.2 Billion Acquisition of Fei-Fei Li's World Labs

According to Zhihui Finance APP, AMD (AMD.US), the most formidable rival to Nvidia in the AI GPU sector and a leading U.S. PC and AI data center high-performance chipmaker that recently surpassed a $1 trillion market capitalization with its stock repeatedly hitting record highs, suddenly announced on Monday U.S. Eastern Time that it would acquire “AI Godmother” Fei-Fei Li’s AI startup World Labs for $8.2 billion. Reportedly, AMD, led by Lisa Su, intends to acquire Fei-Fei Li’s World Labs in an all-stock transaction worth $8.2 billion, thus bringing one of the core AI research forces in world models and spatial intelligence into this chip giant’s integrated business portfolio.

This large-scale acquisition is expected to be completed by the end of 2026, pending approval from key regulatory authorities; according to statements released by both parties, upon closing the transaction, Fei-Fei Li will serve as AMD’s Executive Vice President and Chief Scientist, directly reporting to AMD’s CEO, Lisa Su.

Simply put, the core strategic value of this surprising major acquisition for investors is that AMD seeks to follow the path of Nvidia (NVDA.US), the world’s most valuable company and its strongest AI chip competitor, by making the development of cutting-edge AI large models and the design of chips, software, and AI systems even more deeply and closely integrated. Especially at a time when Meta’s Muse and OpenAI’s GPT-6 Astra are fueling a major wave in AI agents, and as AI technology shifts from answering questions to autonomously executing highly complex tasks, AMD wishes to understand the next generation of AI workloads in advance and thereby design supercomputing platforms best suited for these tasks. This means that AMD is beginning to compete for the “right to define computing power”—currently unique to Nvidia—by mastering the underlying computing requirements of cutting-edge models and participating earlier in the design of next-generation chip and system architectures.

Public information shows World Labs is an AI startup focused on spatial intelligence and world models, with a core business of enabling machines to perceive, generate, reconstruct, and simulate three-dimensional environments. Its Marble product can generate explorable 3D worlds from text, images, or video; the new-generation Atlas model launched in September further integrates 3D reconstruction, viewpoint generation, and spatiotemporal simulation capabilities. The company also enhanced its robotics simulation technology through the acquisition of SceniX, advancing a “real world–virtual training–real deployment” R&D workflow, with application areas spanning robotics models, cutting-edge physics, digital content, and industrial use cases.

From Nvidia GPU challenger to AI large model development, AMD plans to bring Fei-Fei Li’s AI startup World Labs under its umbrella with an $8.2 billion acquisition

The U.S. chip giant Advanced Micro Devices (AMD) has agreed to acquire World Labs for $8.2 billion, bringing the startup founded by AI pioneer and renowned researcher Fei-Fei Li into its group.

The two companies stated in a joint release on Monday that the all-stock deal is expected to close by year’s end, subject to regulatory approval.

As Nvidia’s biggest challenger in the AI chip market, AMD will gain a team of top researchers and the underlying technical framework of AI large models, as well as an in-depth insight into technological trends in the AI large model field. This should aid the company in planning future launches of key AI infrastructure hardware products.

“The more we understand the end-to-end process, the better systems we can build,” AMD CEO Lisa Su said in an interview broadcast Monday afternoon local time.

“That’s why we acquired World Labs,” Su said in the interview. “We want to bring in the world-class AI large model talent that Fei-Fei Li has gathered and fully integrate them with AMD’s capabilities in hardware, software, and systems.”

World Labs develops so-called “world models”—large AI software platforms capable of supporting the major trend of physical world applications. These models can quickly and precisely generate and reconstruct 3D environments. The AI startup says these capabilities can drive new, highly efficient workflows in robotics, fundamental scientific discovery, and factory equipment.

Fei-Fei Li, who previously served as World Labs’ CEO, is a distinguished Stanford University researcher and computer scientist. Under the agreement, she will join AMD as Executive Vice President and Chief Scientist, reporting directly to Lisa Su.

“As the company grows, our ambitions are also expanding,” Li said in an interview with the media on Monday, emphasizing the startup’s surging need for computing power.

“At this stage, working with AMD truly creates a significant growth opportunity for ourselves, AMD, and the entire AI ecosystem, accelerating the flywheel effect between software and hardware development,” she said in the interview.

This acquisition will also enable AMD to offer customers a variety of AI large models, just as Nvidia, led by Jensen Huang, does, thus boosting its product and service portfolio. The company has been expanding its lineup in hopes of selling more products and services to data center clients—currently its largest source of revenue. Its latest strategy is to provide all elements necessary for the rapid construction and deployment of these core AI facilities.

“There will be both open and proprietary models in the future,” Su said. “AMD has always contributed to this field. I believe we will continue to expand this contribution.”

Su added that the ultimate goal is to build better AI. “Our ambition is to shape the future pathway of AI computing.”

Nvidia, an even larger competitor in AI chips, was the first to bundle core AI infrastructure devices, AI cloud services and large models, along with AI chip hardware capabilities, helping clients swiftly launch and operate relevant AI large model systems/platforms. Now, Nvidia is leveraging this strategy to expand into new customer segments, far beyond cloud computing giants, although these giants still contribute the bulk of its growth.

Recently, Nvidia agreed to acquire another highly watched AI startup, Hugging Face, in a deal valued at about $13 billion. Hugging Face is like the “GitHub of the AI model world,” connecting models, datasets, development tools, enterprise customers, and developers. Post-acquisition, Nvidia can fill in the “model discovery–evaluation–optimization–deployment” chain atop its GPUs, NVLink networks, CUDA software stacks, and AI systems, allowing open models to naturally fit its hyperscale AI training and inference platforms and more rapidly identify which model architectures and workloads are driving the next wave of demand for computing power.

World Labs was founded in 2024 and currently has about 70 employees, including a robotics expert team from its July acquisition of SceniX. SceniX is a startup that develops software allowing robots to be trained in highly realistic virtual worlds.

During Lisa Su’s 12-year tenure at AMD, this U.S.-based chip giant has been striving for technological recognition, shedding its image as an underdog in Silicon Valley. In its core processor market, AMD has stepped out from Intel’s shadow and now boasts PC and data center CPU lines rated as the industry’s best.

In the AI accelerator market (AI chip market) for training and running AI models, AMD remains second in market share, still a significant distance behind Nvidia’s near-90% share. However, Su has kept launching a range of new AI infrastructure products, stating that these offerings can outperform the larger competitor’s AI GPU lineup.

Investors view AMD as a primary beneficiary of the unprecedented trillions spent on global AI capital, driving its stock price nearly threefold since the start of the year. AMD’s latest market capitalization is now about $1 trillion—a milestone reached by only 12 other major tech giants, including Nvidia and Apple.

Nvidia remains the most valuable company globally, with a market cap exceeding $5 trillion. But its stock price is up only about 22% this year, and its overall 2026-to-date performance is markedly behind nearly every other hot chip/semiconductor stock, significantly lagging the Philadelphia Semiconductor Index—known as the “AI compute infrastructure barometer.”

Helios seizes the present, while world model strategies define the future: AMD’s AI ambitions extend from selling computing power to participating in its definition

AMD surpassed the $1 trillion market capitalization milestone for the first time on September 21, fueled by a more than 200% surge in share price over the past year. Behind this was the market’s simultaneous reevaluation of the future growth potential of its AI accelerators and data center server CPUs.

Earlier this month, at the Global Technology Conference hosted by Wall Street’s Citigroup, AMD—Nvidia’s strongest competitor in the GPU ecosystem—announced that the market size for AI data center accelerated computing would reach $2 trillion by 2030 (a major upward revision from the previous $1 trillion estimate), noting that AI inference demand has become the main source of growth in AI computing resources, and AI agent workloads are driving both GPU and server CPU demand.

At the conference, AMD revealed that future procurement forecasts from Meta and two other core Helios AI lab clients all exceeded expectations when the strategic partnerships began. The company also projected over 80% year-on-year growth for its server CPU business in the latter half of this year, and over 70% next year. This latest forecast strongly supports ongoing expansion in compute demand, although higher customer expectations cannot all be considered as irrevocable infrastructure orders. Helios is AMD’s rack-level AI computing system, with Facebook parent company Meta as one of its key customers.

It’s important to clarify that AMD’s $2 trillion estimate for 2030 covers the entire computing market—including data center, PC, edge, and embedded fields—with approximately $1.4 trillion from data center AI accelerators, and about $220 billion from server CPUs.

Particularly noteworthy when it comes to fulfilling surging AI compute demand, AMD management said during the Citigroup conference that Meta, OpenAI, and Anthropic—all core clients—provided demand forecasts exceeding initial plans, and that projected demand for Helios in 2027 now surpasses AMD’s previous supply-side expectations. Management expects initial shipments of the MI450 and Helios to begin by the end of the third quarter, with ramping deliveries in Q4 and 2027; the company had previously forecasted over 100% year-on-year data center revenue growth in 2027, and over 80% and 70% server CPU revenue growth in the second half of 2026 and in 2027, respectively. The high-margin CPU business is expected to underpin overall profitability, while the collaboration with Cerebras will supplement low-latency inference capabilities.

Agent applications represented by Muse and Astra provide a clear workload foundation for this expansion: a single instruction can trigger browser actions, code execution, data retrieval, multi-step reasoning, and validation. GPUs handle model computation, CPUs support virtual machines, task orchestration, and tool execution, while memory and storage maintain context, files, and runtime state. As AI super-app platforms move from on-screen tasks into robotics and the physical world, the generation of 3D scenes, environment simulations, and evaluation of multiple action paths add even greater compute demand; the actual degree of growth depends on task scale, model efficiency, and deployment strategies.

Muse uses dedicated cloud VMs and browser automation to carry out cross-app tasks, even continuing work after users close the app. Astra strengthens computer operations, software engineering, and multi-step professional workflows. When AI can handle research, programming, shopping, and office processes, monetizable value extends from generating simple answers to delivering full task results. From a commercial perspective, higher success rates and lower execution costs are expected to expand usage among businesses and consumers, supporting growth in subscription, usage-based payments, and infrastructure revenue expectations.

This expansion of applications drives compute demand at every system layer: GPUs and other AI accelerators power model computation, CPUs run browsers, code, VMs, tools, and orchestrate tasks; increased model weights, context, and concurrent sessions boost high-speed memory demand, while task files, long-term memory, and tiered cache grow DRAM and enterprise SSD needs. Nvidia engineers have also noted that CPU tool execution speed affects agent workflow throughput, and that KV caching can be structured in layers across GPU memory, CPU memory, and storage, according to access needs.

This is why major Wall Street firms like Goldman Sachs and Jefferies are bullish on the sustained earnings growth of the U.S. market under the accelerating penetration of AI and the surge in compute demand. The adoption of AI agents could greatly boost demand in core AI infrastructure fields spanning compute, memory, storage, and optical interconnect networks.

Recently, Jefferies stated that, driven by the dual engines of the AI investment boom and better-than-expected AI-related corporate earnings, the S&P 500 Index is expected to soar to 8,000 points by the end of 2026 and further reach 9,000 points in 2027. The logic is clear: in a cycle where AI-driven earnings growth is more than twice the historical average, betting against earnings trends is risky. Jefferies’ baseline forecast for the S&P 500 at 8,000 in 2026 assumes EPS of $373 (35% year-on-year, much higher than the consensus 29%) and a 21.5x P/E ratio.

From both engineering and investment perspectives, this acquisition can, to some extent, be summarized as AMD’s effort to seize the “right to define foundational AI computing power”—currently led by Nvidia—by gaining earlier insight into what hardware models actually need. Atlas uses a multimodal, autoregressive diffusion architecture, which places distinct requirements on computational throughput, memory bandwidth, and scheduling efficiency through its matrix computation, spatial context management, and iterative generation. AMD can now directly feed these real-world demands back into the design of Instinct GPUs, EPYC CPUs, ROCm software, and rack systems. The two sides have already collaborated on Instinct optimization and workload adjustments; this acquisition will make that process more direct. For shareholders or investors, long-term value will depend on whether these R&D capabilities translate into lower task completion costs, faster customer deployments, and higher product adoption rates.