AI & Tech Digest: Tokyo's Hub, DeepSeek's Leap, Apple's AI Mandate, and Meta's Talent Exodus
Stay updated with the latest in AI and tech: Tokyo emerges as a key innovation hub, DeepSeek challenges frontier AI models, Apple's new CEO faces a critical AI product mandate, and Meta sees top talent migrate to surging AI startups.
The tech world is buzzing with significant shifts and groundbreaking announcements this week. From a major global tech event highlighting cutting-edge AI to a Chinese AI lab making strides that could reshape the LLM landscape, innovation is clearly accelerating. Meanwhile, leadership changes at tech giants underscore the increasing imperative of AI, even as talent flows towards dynamic startups.
TL;DR
- SusHi Tech Tokyo 2026 is set to define the future of technology with focused domains, including a significant presence from TechCrunch Startup Battlefield.
- Meta loses key AI talent, Weiyao Wang, to Thinking Machines Lab, which recently secured a multibillion-dollar Google Cloud deal for Nvidia GB300 chips.
- Apple's incoming CEO, John Ternus, is under pressure to deliver a 'killer AI product' to solidify the company's position in the AI race, following Tim Cook's departure announcement.
- DeepSeek has unveiled its new V4 Flash and V4 Pro AI models, aiming to 'close the gap' with leading frontier models with their 1 million token context windows.
- DeepSeek's V4 models are also highlighted for their compatibility with Huawei technology, positioning China's AI advancements to challenge major US rivals like Google, OpenAI, and Anthropic.
Tokyo Emerges as a Global Tech Nexus with SusHi Tech Tokyo 2026
Tokyo is rapidly establishing itself as a pivotal destination in the global tech landscape, especially with the upcoming SusHi Tech Tokyo 2026 event. This conference is distinguishing itself from typical tech gatherings by focusing on four tightly defined technology domains, complete with live demonstrations, dedicated exhibit floors, and sessions featuring global leaders in technology and funding. This structured approach aims to cut through the usual industry hype, particularly in areas like AI.
TechCrunch is participating as an official media partner, and its Startup Battlefield team will be on-site. They plan to select one standout semifinalist from the SusHi Tech Challenge to advance directly to the prestigious TechCrunch Disrupt Startup Battlefield 200. This collaboration underscores the event's importance as a launchpad for innovative startups.
Tokyo's SusHi Tech Tokyo 2026 is poised to be a critical event for defining future technology, moving beyond vague themes to focus on tangible advancements and fostering global collaboration.
Sessions will feature prominent figures such as Howard Wright (Nvidia), Rob Chu (AWS), and Eric Benhamou (Benhamou Global Ventures), signaling a deep dive into the infrastructure and practical applications of AI. The event highlights Tokyo's ambition to become a central hub for technological innovation and investment in 2026.
Meta's AI Talent Shifts to Thinking Machines Lab After Major Google Cloud Deal
Meta has experienced a notable talent departure as Weiyao Wang, an expert in multimodal perception systems and a contributor to open-world segmentation projects like SAM3D, has left the company after eight years. Wang has joined Thinking Machines Lab (TML), an AI startup that is rapidly expanding its operations.
This move coincides with TML's recent multibillion-dollar cloud deal with Google, announced this past Tuesday at Google Cloud Next. This significant agreement grants Thinking Machines Lab early access to Nvidia's latest GB300 chips, positioning it as one of the first startups to leverage this cutting-edge hardware. The deal places TML in the same infrastructure tier as major AI players like Anthropic and Meta itself, highlighting its growing influence in the AI sector.
The departure of a key AI talent from Meta to Thinking Machines Lab, fresh off a multibillion-dollar Google Cloud deal for Nvidia GB300 chips, signifies a dynamic shift in the competitive AI landscape.
Reports indicate that Meta had previously held talks to acquire Thinking Machines around last year, suggesting that TML's continued independence and recent infrastructure advancements make it a formidable player in AI development. The recruitment of top talent like Wang further strengthens TML's capabilities in advanced AI research and development.
Apple's Next CEO, John Ternus, Faces Crucial Mandate: A Killer AI Product
Apple's current CEO, Tim Cook, has announced his departure in September to become the company board's executive chairman. His successor, John Ternus, the current SVP of hardware engineering, will inherit a paramount task: to launch a truly revolutionary AI product. While Cook successfully guided the company post-Steve Jobs, he did not "crack AI," leaving this critical challenge for Ternus.
The expectation is that Ternus will unveil a product that translates the complex power of AI into an intuitive and delightful experience for the masses, much like Apple did for other technologies. This isn't just an aspiration; it's considered essential. Current powerful AI agent technologies, such as Claude Code and OpenClaw, remain too technical or risky for widespread adoption. If Apple doesn't simplify AI for consumers, a competitor will.
Apple's incoming CEO, John Ternus, faces the immediate and critical challenge of developing a 'killer AI product' to democratize AI for the masses and secure Apple's future in the rapidly evolving tech landscape.
Ternus has historically maintained a low public profile, described as a methodical operator rather than a visionary. However, with the CEO role, he may be "liberated to reach for the skies." The success of his tenure will largely depend on his ability to deliver on this significant AI mandate, especially after Apple Intelligence, rolled out in 2024, was perceived as underwhelming.
DeepSeek Previews V4 Models, Aiming to Rival Frontier AI
Chinese AI lab DeepSeek has officially launched preview versions of its highly anticipated next-generation large language model, DeepSeek V4. This update follows last year's V3.2 model and the accompanying R1 reasoning model, which generated considerable buzz in the AI community. The new releases, DeepSeek V4 Flash and V4 Pro, are mixture-of-experts models, each boasting impressive 1 million token context windows.
This extensive context window allows for processing large codebases or documents within prompts, a significant advancement for practical AI applications. The mixture-of-experts approach is designed to activate a select number of parameters per task, aiming to reduce inference costs. Notably, the V4 Pro model features a massive 1.6 trillion total parameters, with 49 billion active, making it the largest open-weight model currently available, surpassing competitors like Moons.
DeepSeek's V4 Flash and V4 Pro models, with their 1 million token context windows and 1.6 trillion total parameters, mark a significant stride in open-weight AI, aiming to bridge the performance gap with frontier models.
DeepSeek asserts that these new models are designed to "close the gap" with leading frontier models. The company's previous R1 model had already made waves by claiming to achieve its training at a fraction of the cost of leading US systems. The V4 release reinforces DeepSeek's position as a serious contender in the global AI race.
China's DeepSeek V4 Model Challenges US AI Dominance and Highlights Huawei Compatibility
Chinese AI company DeepSeek has launched a preview of its much-anticipated V4 AI model, claiming it can compete "toe-to-toe" with leading closed-source systems from major US rivals like Anthropic, Google, and OpenAI. This announcement comes a year after DeepSeek's R1 model rattled the US AI industry with claims of significantly lower training costs.
The V4 model represents a major improvement over its predecessors, particularly in coding capabilities. This focus on coding is crucial, as it has become central to the development and success of AI agents and tools such as ChatGPT Codex and Claude Code. Furthermore, DeepSeek has explicitly highlighted the V4 model's compatibility with domestic Huawei technology, marking a significant milestone for China's chip industry.
DeepSeek's V4 model challenges US AI leadership with its advanced coding capabilities and explicit compatibility with Huawei hardware, signaling China's growing self-reliance and competitiveness in the global AI landscape.
While DeepSeek has not disclosed the training costs or hardware used for V4, past controversies include US officials accusing the company of utilizing banned Nvidia chips and Anthropic alleging misuse of Claude to enhance DeepSeek's own products. Despite these claims, the V4 release underscores DeepSeek's ambition to advance China's AI capabilities and compete effectively on the global stage.