PART 01 A Gentle Acquisition and Realistic Considerations

'It is never too late to seek a newer world.'

On September 28, AMD officially announced the acquisition of World Labs, founded by Fei-Fei Li, in an US$8.2 billion all-stock transaction. Subsequently, Li quoted this line from Alfred, Lord Tennyson's Ulysses in her personal letter.

At 48, she started her entrepreneurial journey and, two years later, sold her company to AMD for US$8.2 billion, marking the second-largest acquisition in AMD's history. Following the deal, Li was appointed Executive Vice President and Chief Scientist at AMD, reporting directly to Chairman and CEO Lisa Su.

This quote is not just a middle-aged person's encouragement to herself but also a pledge to Lisa Su and the AI industry.

Lisa Su was an early investor in World Labs. Li referred to her as a 'great friend' in her announcement letter. At CES this January, Su invited Li on stage for a keynote speech, where they demonstrated Li's product, Marble, achieving a fourfold acceleration on AMD chips. Even earlier, the two companies had begun jointly training models.

Looking back, their relationship has steadily warmed, making the acquisition seem like a natural progression.

Silicon Valley has witnessed countless fierce acquisition battles: Oracle vs. PeopleSoft, Microsoft vs. Yahoo, Broadcom vs. Qualcomm—each dramatic enough to fill a book. In contrast, AMD's acquisition of World Labs was remarkably gentle. Given that both Lisa Su and Fei-Fei Li are Chinese-American women, this could be considered Silicon Valley's top-tier 'Girls Help Girls' moment.

The two women at the helm of this acquisition are now pivotal figures in Silicon Valley.

Lisa Su took over AMD in 2014, reviving the then-struggling company with a market value of just over US$2 billion and heavy losses. Twelve years later, its market value once surpassed US$1 trillion. The key lay in Su's bets on the Zen architecture and AI technology as an engineer. Facing complex and grand challenges, she excels at narrowing the battlefront and rebuilding around a certain technological pivot.

Now, physical AI is emerging as a critical exploration direction for the next phase of the AI industry.

The competition around large language models (LLMs) seems unable to reignite the industry's passion. More exploration is shifting toward physical AI: robotics, autonomous driving, and industrial simulation.

NVIDIA, which has reaped substantial rewards during the LLM era, has already established a new Layout (layout) around physical AI: Cosmos provides foundational world models, Isaac offers robotics development and simulation tools, and Omniverse organizes and operates the digital world. Thus, NVIDIA is no longer just a card vendor; it can now define how robotics companies generate data, train models, and deploy applications—directly corresponding to industry influence.

AMD wants a piece of this pie. It has launched relevant hardware and software solutions and participated in robotics, industrial simulation, and world model ecosystems.

But Lisa Su is clearly not satisfied. She aims to transform AMD from a supplier into an ecosystem provider. NVIDIA has proven that the most substantial profits in the AI era come from controlling the entire technology stack.

Acquisitions have become one of AMD's most efficient ways to catch up. In recent years, AMD has acquired Xilinx, Pensando, Silo AI, ZT, and others, with a clear logic: to complete the ecosystem spanning chips, networks, server systems, software, and models. World Labs fits into the 'models' segment.

AMD stated in its acquisition announcement that building a computing platform for next-gen AI requires a deep understanding of how models evolve.

In simpler terms: Chip sellers must know before chip buyers what kind of chips future AI will need. Traditionally, chip companies waited for software demands and then met them with more advanced processes, architectures, and packaging. However, in the new AI paradigm, models iterate too quickly, making model teams the 'prophets' for hardware teams.

NVIDIA's success has proven that software can define hardware. Now, Lisa Su wants to take it a step further: using models to define hardware.

The evolution of world models is precisely Li's exploration direction.

This Chinese-American girl, who immigrated to the U.S. with her parents as a child, worked in her parents' dry-cleaning store in New Jersey during her youth, entered Princeton on a scholarship, and then created ImageNet. She brought large-scale, annotated visual data into computer vision research, enabling breakthroughs in deep learning with neural network algorithms and GPU computing.

Without ImageNet, the computer vision wave might not have surged in 2012.

In 2024, Fei-Fei Li founded World Labs, continuing this academic trajectory: enabling AI to acquire spatial intelligence. While LLMs excel at processing symbols and text, World Labs aims to further handle objects, scales, perspectives, motion, and physical relationships. Li often says, 'The universe is made of real things.'

Capital quickly followed: In 2024, World Labs secured US$230 million in funding. By February 2026, the company completed another US$1 billion funding round, with investors including AMD, NVIDIA, and Autodesk. According to media reports, its valuation reached US$5 billion just 16 months after its founding.

Behind the soaring valuation lies World Labs' continuous technological breakthroughs: Marble, launched in November 2025, can create editable 3D worlds based on text and image inputs. Atlas, released in September 2026, further processes text, images, videos, and 3D information, aiming for reconstruction, generation, and simulation.

Additionally, in July 2026, World Labs acquired robotics company SceniX, signaling its expansion from a single 3D generation tool to spatial intelligence infrastructure.

This 'expansion' ambition ultimately brought Li and Su to the acquisition negotiation table. The US$8.2 billion deal is expected to close by the end of 2026, pending regulatory approval.

PART 02 A Certain Deal, an Uncertain Future

In late 2025, Mark Zuckerberg also made a nearly equally urgent deal.

On December 30, Meta announced the acquisition of AI agent company Manus for over US$2 billion—Meta's third-largest merger after WhatsApp and Scale AI. From contact to signing took just over ten days, with Zuckerberg, a heavy user of Manus, making the final decision.

Manus develops general-purpose AI agents: autonomous Agents capable of searching, writing code, and operating software. While not in the same track (track) as World Labs' spatial intelligence, the underlying anxiety is the same: Social platform traffic has peaked, and the AI era's gateway might not be a chatbox but intelligent agents that can work on behalf of humans. This is what Meta lacks.

However, this deal did not end well. It was ultimately halted by China's National Development and Reform Commission, and the transaction was canceled. The Manus team announced its return to independent operations on September 1 and, on September 28—the same day AMD announced its acquisition of World Labs—released Manus 2.0 for overseas users while announcing plans to build a team for the domestic market.

Of course, Li is unlikely to face similar troubles. Her challenge lies in whether the technology can transition from the lab to the real commercial world and deliver sufficient value.

A domestic AI industry insider shared their views on the acquisition: It highlights the bright prospects of world models but also indicates a long road ahead in the short term.

World models are computational black holes. Generating an interactive 3D world consumes far more training and inference computational power than text models. This is a limitation for startups. Li lamented, 'Without dedicated hardware investment, AI will remain trapped in the digital world.'""From a practical standpoint, selling to AMD is indeed one of the best choices for World Labs. AMD can provide more abundant chips, computational power, and industrial resources while helping World Labs reach a broader developer and enterprise client base. More importantly, there is Lisa Su's trust: After the acquisition, Li will report directly to her.

The question that will continue to draw attention is: Can this acquisition enable AMD to 'recoup its investment' and even yield substantial returns? Moreover, can the two Chinese-American leaders continue the 'Girls Help Girls' narrative after truly working together?

Silicon Valley has seen precedents.

In 2014, Google acquired four-year-old DeepMind for a reported £400 million. Demis Hassabis negotiated terms with Larry Page: maintain independence and avoid absorption by Google Brain. However, many conditions became obsolete after a decade. In 2023, cornered by ChatGPT, Google forcibly merged the two, leading to ongoing cultural clashes and an exodus of core employees. In June this year, Nobel laureate John Jumper left Google for Anthropic. In August, Hassabis stepped down from DeepMind's daily management to become Chairman of Google DeepMind and Chief Scientist of parent company Alphabet.

This 12-year integration revealed the irreconcilable tension between independent labs and corporate commercial rhythms.

Similar risks exist in AMD's acquisition. Chip companies' roadmaps are measured in quarters, product generations, and customer orders. After World Labs joins AMD, its research will gain faster access to industrial resources but will also face more internal and external questions about 'budgets, delivery, and revenue.'""Another issue is open ecosystems.

World Labs' past investors span AMD, NVIDIA, Intel, Cisco, and Autodesk. This shareholder structure aligns with a foundational model company's ideal position: maintaining platform neutrality to enable products to operate in diverse hardware and software environments. This neutrality now faces scrutiny. Although AMD emphasized in its announcement that the team would continue advancing research and an open ecosystem, the true answer lies in product interfaces, licensing methods, performance optimization, and commercial partnerships.

Behind this lies real and subtle conflicts of interest. AMD hopes World Labs will become a differentiated advantage for its chips; for World Labs to become industry infrastructure, it needs as many platforms, clients, and developers as possible.

Fortunately, this is an era of rapid AI iteration. Many problems encountered along the way are solved through development.

There is no pre-existing roadmap for the new world.

Two decades ago, Li pushed machine learning to 'see' the world with ImageNet. Today, she faces more complex challenges: How can AI understand a 3D environment capable of motion, collision, occlusion, and change, and how can it integrate into the commercial world's operations?""Lisa Su leads a 'reborn' AMD, no longer plagued by survival crises but now focused on catching up with NVIDIA.

Faster chips, more usable models—these individual technological capabilities must ultimately form a smoothly functioning ecosystem to secure greater influence in the next phase of AI development (whether physical AI or otherwise).""Of course, for AMD, whose market value is nearing US$1 trillion, an US$8.2 billion acquisition is just a small sum. This is an advantage of large companies: They can 'spend' to fill capability gaps, with controllable risks and vast potential upside if successful.

But for Li, who transitioned from academia, selling her company to AMD truly means entering a 'newer world'—or even an 'entirely new world.' It holds captivating new stories but also many 'traps' she has never experienced.

Regardless, at least at the moment of signing, Silicon Valley witnessed its most expensive 'Girls Help Girls' scenario imaginable.