OpenAI Flags Its Top Chinese Rival. It's Not Who You Think
Zhipu is selling AI sovereignty. OpenAI wants the U.S. to see what’s coming
On June 26, OpenAI’s Global Affairs team published [a blog post titled “Chinese Progress at the Front”, identifying Zhipu AI as a key player in China’s push to expand its influence through artificial intelligence. The post highlights Zhipu’s government ties, overseas expansion, and efforts to export sovereign AI infrastructure to countries across Southeast Asia, the Middle East, and Africa.
It’s a striking move—one that elevates Zhipu from an industry competitor to something more strategic. And it raises an obvious question: why now?
This post looks at what Zhipu is actually doing, how it fits into China’s broader AI strategy, and what OpenAI might be signaling—intentionally or not—about its own evolving role in the global AI landscape. Because beneath the details lies a deeper shift: AI is no longer just a race between models. It’s a contest over who defines the infrastructure, the norms, and the future of digital power.
So what is Zhipu, really?
The real competition isn’t about whose chatbot scores higher on a benchmark. It’s about who shows up first with a full stack.
Zhipu AI is one of China’s so-called “AI tigers,” a term the Chinese press uses for the country’s new generation of large model startups. It was spun out of Tsinghua University in 2019 and has since become a flagship in Beijing’s push for AI sovereignty. Backed by over $1.4 billion in state-linked investment, Zhipu sits at the center of a coordinated effort to develop and export a Chinese AI stack, from chips to code to governance.
It’s also currently the most aggressive of China’s foundation model firms in going global.
Zhipu is opening regional offices in the UAE, Singapore, Malaysia, and the UK. It’s launching joint innovation centers in Indonesia and Vietnam. It’s marketing “sovereign LLMs” to governments looking to control their own models and infrastructure. It’s building those systems on Huawei Ascend hardware and has received capital from funds like Saudi Aramco’s Prosperity7. One of its Dubai labs is focused specifically on content moderation and “AI safety.”
In April, Zhipu filed IPO paperwork with China’s securities regulator, making it the first of China’s “Six Little Dragons” to formally start the listing process. It’s burning cash fast - losing about 2 billion yuan last year against 300 million in revenue - but the move is as much about strategic positioning as it is about funding. Zhipu wants to be first to market, and first to claim the title of China’s “first large model stock.”
Zhipu has strong ties to Beijing’s AI establishment. It incubated at the state-backed Beijing Academy of Artificial Intelligence (BAAI), which pioneered the WuDao and WuJie model families. Its leadership regularly meets with senior CCP officials. It’s part of the Digital Silk Road strategy, which is trying to build an end-to-end Chinese tech ecosystem that can be exported globally - especially to Belt and Road partner nations.
Zhipu isn’t just selling AI. It’s selling a framework: A self-contained system of infrastructure, governance, and values that countries can plug into without relying on the West.
Why is OpenAI suddenly talking about Zhipu
OpenAI didn’t have to name Zhipu. It could have focused on DeepSeek, or the state-backed Tongyi Qianwen models from Alibaba. But it didn’t. Instead, it framed Zhipu as a global actor exporting not just models, but standards. A builder of AI systems that are, in OpenAI’s words, “responsible, transparent, and audit-ready” - a clear signal that Zhipu is pitching itself as the OpenAI alternative to the Global South.
Why highlight this now? Because OpenAI is changing too.
The company recently launched OpenAI for Government, signed a $200 million deal with the U.S. Department of Defense, and is leading the $500 billion Stargate UAE project with Oracle, Cisco, Nvidia, and SoftBank. It’s becoming more than a model company. It’s becoming infrastructure. And with that comes a shift in posture—from innovator to geopolitical player.
In that context, Zhipu’s expansion isn’t just a foreign competitor but a foil- a narrative frame that helps OpenAI position itself to U.S. and allied governments as the necessary counterweight. The responsible stack. The democratic stack. The stack you can trust.
Framing Zhipu as a CCP-aligned soft-power tool doesn’t just raise alarm, it also answers a strategic question that gets at when and how the defense establishment will work its way into the AI race: Why should Washington and its partners back OpenAI, and not just let the market decide?
OpenAI is saying: Because the other side already is.
Is Zhipu really that threatening?
Zhipu’s implicit pitch to emerging markets: “You don’t have to rent your AI future from California. You can build your own—with us.”
In terms of model performance, customer traction, or economic scale? Not yet. In terms of reach? Possibly.
Zhipu is still early. Its IPO is months away. Its revenue is small. Its expansion into ASEAN and Gulf markets is promising but limited. And domestically, it’s under pressure. It’s fighting off competition from Alibaba, Baidu, and ByteDance - all of whom can undercut Zhipu on API pricing and bundle models with cloud services at massive scale.
There’s also the issue of stickiness. Many of the government clients Zhipu is courting - health ministries, education bureaus, local SOEs - may be eager to experiment with AI, but they’re also pragmatic. If U.S. or European providers can offer more powerful tools, or easier integration, or stronger diplomatic ties, loyalties can shift.
But that’s not the only risk.
The deeper risk, and the one OpenAI seems to be calling out, is that Zhipu is embedding systems before anyone else shows up. Infrastructure has a way of entrenching itself. Once you’ve trained your model on a Huawei box, integrated your content policies with a “safety lab,” and co-branded your national chatbot with a Chinese firm - you’re no longer just using their tools. You’re aligned with their stack.
That kind of alignment is hard to reverse.
The real contest: narrative vs. infrastructure
Still, what Zhipu is doing - and this is where it matters - is shifting the frame. It’s offering governments a story: You don’t have to rent your AI future from California. You can build your own - with us.
That’s the promise. And it comes bundled with hardware, model weights, safety protocols, and the political reassurance of co-ownership. Train your model on Huawei gear. Plug it into a national platform. Co-brand it. Regulate it your way. Zhipu isn’t just selling AI - it’s selling sovereignty, or at least a version of it that’s legible to countries caught between the U.S., China, and the demands of rapid digital modernization.
Is OpenAI scared of Zhipu? Doubtful. The positioning of the claim is more like “We see the game they’re playing - and we need to play it, too.”
Because underneath all the rhetoric about safety, hallucinations, and open models, the real competition isn’t about whose chatbot scores higher on a benchmark. It’s about who shows up first with a full stack. Who builds the default. Who embeds standards before anyone else gets a chance to.
That’s the world Zhipu is trying to build. And it may be the world OpenAI is trying to prevent. Either way, the fight is already on. And if we want to understand what’s at stake, let’s stop watching the models and start watching the infrastructure.