Alibaba Aims for Independence with New AI Chips, Model
The chip highlights the vendor’s move toward a full-stack AI strategy and its efforts to wean itself off Nvidia AI chips.
The chip highlights the vendor’s move toward a full-stack AI strategy and its efforts to wean itself off Nvidia AI chips.
DeepSeek is now chasing a valuation above $20 billion as Tencent Holdings and Alibaba Group discuss possible investments in the Chinese AI startup.
The Information reported that on Wednesday, citing four people who knew about the talks. DeepSeek, which is owned by hedge fund High-Flyer Capital Management, had only just started talking to outside investors for the first time.
By Friday, the reported target was at least $300 million at a valuation of at least $10 billion. Now the asking price has climbed fast as interest builds around DeepSeek.
The talks are still going on, so the final number could still change. The amount DeepSeek wants to raise could also change. Some U.S. venture capital companies may be cautious because DeepSeek is a Chinese startup.
Earlier this year, Cryptopolitan reported that DeepSeek did not show U.S. chipmakers its flagship model for performance tuning. We also reported that one of DeepSeek’s newer models was trained on Nvidia’s most advanced banned chip.
Back in January 2025, the first big DeepSeek release helped trigger a global tech selloff and pushed Chinese rivals to upgrade their own models.
Meanwhile, on the Dwarkesh Podcast on Wednesday, Nvidia chief executive Jensen Huang said it would be “a horrible outcome” for the United States if DeepSeek optimized its new AI models to run on Huawei chips instead of American hardware.
Jensen said, “If future AI models are optimised in a very different way than the American tech stack,” and as “AI diffuses out into the rest of the world” with Chinese standards and technology, China “will become superior to” the United States.
On chip performance alone, Huawei still trails Nvidia. The Ascend 910C, which came before the 950PR, delivers about 60% of the inference performance of Nvidia’s H100. That H100 is already two generations behind Nvidia’s current best chip.
American chips are about five times more powerful than Chinese rivals today, and that gap is expected to widen to 17 times by 2027. Huawei is targeting 750,000 AI chip shipments in 2026, but its total production amounts to only about 3% to 5% of Nvidia’s combined computing power.
“A lot of work has to go into it to change. But go to the global south, go to the Middle East. Coming out of the box, if all of the AI models run best on somebody else’s tech stack, you’ve got to be arguing some ridiculous claim right now that that’s a good thing for the United States,” said Jensen.
Jensen said his real concern is not just the gap in chip strength. He said China could still catch up in AI because it has “abundant energy” and a “large pool of AI researchers.”
If DeepSeek V4 runs well on Ascend chips, that would give China another route in AI development that does not depend on Nvidia across the supply chain.
That same funding rush showed up elsewhere on Wednesday. Vast Data announced a $1 billion funding round at a $30 billion valuation, and Nvidia was one of the backers.
The company says it supports projects that power millions of GPUs. Its customers include CoreWeave, Mistral, the U.S. Air Force, and Cursor. The new round more than tripled Vast’s $9.1 billion valuation from 2023. Drive Capital and Access Industries led the Series F. Fidelity Management and Research Co., NEA, and Nvidia also joined.
The financing included both primary and secondary capital. Dealroom said AI companies globally have already raised $280.5 billion this year, with more than $170 billion going to OpenAI, Anthropic, and xAI.
Chris Olsen of Drive Capital said, “The scale and speed of AI adoption are creating a new class of infrastructure company.” Chris added, “VAST is emerging as the clear leader in this category, with the architecture and momentum to support the world’s most demanding AI environments.”
The smartest crypto minds already read our newsletter. Want in? Join them.
Alibaba Group Holding Ltd. introduced QwenVLo, its newest multimodal AI model, as the Chinese tech giant continues to bolster its footing in the growingly competitive global AI battle.
The model allows users to create and edit images with text prompts and visual inputs, capabilities meant to be on par, or even compete with leading AI developers like OpenAI and DeepSeek.
QwenVLo is an improvement on Alibaba’s previous Qwen2. 5-VL baseline model and extends it to support new tasks such as text-to-image and image-to-image generation.
One standout feature is that of progressive generation, which means that users can see an image as it’s being built — a level of transparency and interactivity rare in most AI tools today.
“This newly upgraded model not only ‘understands’ the world but also generates high-quality recreations based on that understanding,” the company said in a blog post.
The push into AI is no longer a glam-up girl for Alibaba’s e-commerce dominance. The company has been aggressively working its way into AI since early 2024.
In February, Alibaba CEO Eddie Wu said that the e-commerce and IT behemoth would prioritize its work on artificial general intelligence (AGI). He stressed that developing AGI had now become Alibaba’s top priority, highlighting the rising significance of AI to the company’s growth and competitive edge in the future.
Alibaba has already launched a range of models under the Qwen brand, each tailored to a specific modality — text, image, video, and audio. QwenVLo replaces that mix of things and is built to perform on lower-end machines, including mobiles and personal laptops; however, unlike most state-of-the-art AI models, which still need loads of cloud computing power to run efficiently.
In March 2025, Quark released a new Quark app, which included the new Qwen functionality. Now, the app operates as an AI-powered smart assistant with search, summarization, and creative tools. It demonstrates Alibaba’s desire to build AI models and weave them into consumer-facing experiences.
The release of QwenVLo by Alibaba coincides with a wave of AI-related initiatives worldwide. US-founded OpenAI has released GPT-4o, its most advanced multimodal model yet, which can understand and respond with text, images, and audio.
Over in China, all eyes were on DeepSeek, a homegrown startup for whom cost efficiency is fast becoming a make-or-break challenge in AI development, as it claimed to have created a large competitive language model for just a few million dollars.
Tech titans all over China have since rushed out similar AI services that are faster, cheaper, and more versatile. One of its first serious entries is Alibaba’s Qwen Vlo, which has the potential to take on both its Western and Chinese competitors by delivering high-quality and low-weight multimodal functionality.
The race is now more about who can build the smartest model, but who can make that intelligence usable at scale.
Alibaba is betting that the types of people who use an app every day- students, designers, business owners, and developers- will flock to tools that are quick, simple to use, and optimized for various devices.
However, whether Qwen VLo is something marquee names dominate the users in the market is yet to be seen. But Alibaba has hammered the message: China is not just keeping pace with the world in the race for AI supremacy — it is determined to lead it.
KEY Difference Wire helps crypto brands break through and dominate headlines fast