Japan has world-class tech, deep engineering talent, and a fast-growing startup scene.
But turning local success into global growth is a different challenge - one the startups youโll meet at the J-StarX Silicon Valley Extended Program 2026 Virtual Demo Day came to the Bay Area to scale towards!
These 11 startups were selected from a larger cohort for their readiness for U.S. expansion. Meet the founders behind Allied Robotics Industries, Almaprism, Aurevia (Callara), Gigalogy, Lifiya, MiRNDa, LEAD Japan, TC3 (Tactna), Terravilla, Tokoshie, and YAOYO.
๐ November 18 | 4 โ 5 PM PST (U.S.)
๐ November 19 | 9 โ 10 AM (Japan)
๐ Virtual - join from where you are.
Request To Attend: https://t.co/1GWJGdkfPC
@JETRO_jgc@JETRO_info
Less than 0.23 seconds of downtime for model weight updates. That's what @humansand has achieved with weight teleportation, a new technique designed to make large-scale reinforcement learning more efficient.
Built with NVIDIA Blackwell hardware, the approach compresses and transfers model weight updates efficiently, even over internet-speed connections.
For AI researchers and developers, this means faster model updates and greater flexibility in where training infrastructure operates.
We see infrastructure efficiency as an important driver of AI progress, enabling researchers to make better use of compute and accelerate experimentation.
The team has open-sourced weight teleportation. Explore the technical details here: https://t.co/5W00FCFDUw
@christine_tsai@TonyW@dfect@vishalharnal@atxfrench@leoclee
Just a week after launching Craft, their first image-generation model, @canva has introduced Brush, a vision-language model built to make design editing faster and more intuitive.
Brush can understand a design's structure and make targeted changes, from updating individual elements to editing entire presentations.
For us, the significance lies in combining AI capabilities with efficiency and creative control. By making every element fully editable, Canva is showing how AI can enhance human creativity, giving people more freedom to experiment and less time spent on repetitive edits.
Read more here: https://t.co/dNOcPJP2T2
@christine_tsai@TonyW@vishalharnal@khailee@dfect@atxfrench@leoclee
What would it take to help the next billion build with AI?
Our first AI-for-All Learnathon gave us a starting point:
1) Access to tools
2) Experienced trainers
3) Projects learners care about
100+ working products submitted in a day!
Now, to make it work in more cities.
Introducing the next chapter of Otter. Today, we're launching Work Mode and Live Assist, two new capabilities designed to help you prepare for meetings, get real-time coaching, and turn conversations into action. Reach what's new: https://t.co/8zMDbWdO3h
Sakana AI won 1st place in the company category at the โRecursive Self Improvement in Mathematicsโ Hack (OpenMath 2026), held with @stephen_wolfram at Harvard Innovation Labs.
Competition: https://t.co/GIe4OW7GYq
Results: https://t.co/uTHtSIQVtH
Drawing mainly on our Sakana Fugu models and internal agentic tools, our team (Rachel Teo, Abderrahmene Hamdi, Wenyi Wang) contributed 14 mathematical families, including 6 complete resolutions, plus new bounds and structural advances.
Thanks to the organizers and everyone involved.
We plan to share more about our approach and learnings in a future report.
Congratulations to Hyemin Lee and the Finda team on making a significant move toward their vision of an AI-native bank!
The company has acquired a 60% stake in Daewon Savings Bank, marking a first for a South Korean fintech firm.
Having brokered over โฉ15 trillion in loans across 300+ products from 98 financial institutions, Finda brings substantial financial data and consumer insights to this new venture.
The company plans to use AI to develop deposit and lending products tailored to customers' financial needs, while deploying AI agents across banking operations to improve productivity.
We see significant potential in combining years of consumer borrowing insights with direct lending capabilities. Finda's approach reflects how AI and proprietary data can help financial institutions better understand underserved borrowers and develop products tailored to their needs.
Full story here: https://t.co/rGrn282xK5
@christine_tsai@TonyW@vishalharnal@khailee@dfect@atxfrench@leoclee
Canva has just launched its first AI image generation model, and it's already ranked #4 globally on DesignArena, making Canva Research the world's second-highest-ranked AI lab for image generation.
Built specifically for design, Canva Craft generates images at up to 2K resolution, with precise control over composition, typography, and individual elements that can be edited through Magic Layers.
For @canva's 250M+ users, this means greater creative flexibility, while its public API opens up new possibilities for developers building AI-powered design tools.
This is a compelling example of how frontier AI research can translate into practical creative tools, putting sophisticated design capabilities in the hands of millions.
Try Canva Craft: https://t.co/jjJ4NEEixu
Read more here: https://t.co/QV8iRccvKm
@christine_tsai@TonyW@vishalharnal@khailee@dfect@atxfrench@leoclee
Every image in this short was created w/ #CanvaCraft. Congrats @MelanieCanva Cliff Obrecht @themaninblue & @canva team on releasing Canva's 1st image generation model! Debuted as the world's 4th most advanced image generation model on @designarena, making Canva the 2nd most advanced AI lab for image gen.
AI has transformed software. Science is next, and science happens in a lab.
@warpsciences is out of stealth with $215M to build AI + robotics-powered chemistry labs that help researchers discover new medicines faster. Proud to back @kevinweil@dpatil@manohar_paluri and Sol Reisberg.
๐ฅ Introducing Warp Sciences, a new company applying AI and robotics to help the world develop new medicines faster.
Weโve raised a $215M Series A led by @lightspeedvp to bring frontier AI out of the browser and into the lab, and build scientific capacity here in America.
I co-founded Warp with my long-time friend @dpatil, FAIR co-founder @manohar_paluri, and drug hunter Sol Reisberg. The early team is full of incredible people like @bkspears9, who helped achieve Q>1 fusion at LLNL.
Huge congrats to @humansand on launching Ampersand.
A model trained to collaborate, and measured on what people actually get done together over months, not single-turn benchmarks. That's the right thing to optimize for.
From approximately 80MW committed in 2026 to a projected 580MW by 2027, Aolani is making a major push to expand its AI infrastructure footprint in Malaysia!
The addition of 500MW in AI data centre capacity is targeted to come online starting in the second half of 2027. These facilities will support its growing portfolio of AI factories and GPU deployments, serving global hyperscalers and AI-native companies.
We believe the real opportunity extends beyond building more data centres. By making high-performance compute more accessible to local enterprises, researchers, and startups, Malaysia has an opportunity to become not just a hub for AI infrastructure, but also a place where more AI innovation is built.
Full story here: https://t.co/ZcQPF5KHyz
@vishalharnal@khailee@AriefJohan@amirzakwan@atxfrench
Iโm so happy to be able to share some of our work with you. Our model Ampersand, we trained to help people come closer to their dreams.
I love the humans& team and am so glad I can share this moment with them and you.
AI keeps getting better at completing individual tasks. With Ampersand, @humansand is introducing a model designed to act as a collaborator, helping people work toward shared goals.
Trained through simulations with Persimmon, humans&'s model of human behavior, Ampersand learns to consider the longer-term impact of its actions.
In multi-month simulations, users working with Ampersand made greater progress toward their long-term goals than those using other frontier models evaluated.
We're particularly interested in AI that understands not just what people ask for, but what they're ultimately trying to achieve.
We look forward to seeing how developers bring this approach to new applications. Explore Ampersand here: https://t.co/OdZEm7OzWg
@christine_tsai@TonyW@vishalharnal@khailee@dfect@atxfrench@leoclee