Microsoft Sells OpenAI Models in China, While Amazon Aims to Catch-Up

MSFT has quietly built one of the most consequential AI businesses in China, reportedly selling OpenAI's GPT-series models to major Chinese technology companies through its Azure cloud platform. ByteDance, the parent of TikTok, is on track to spend more than $1 billion a year on Microsoft AI and cloud services, making it one of Microsoft's largest global AI customers. Ant Group, Meituan, and Tencent are also reported to be significant spenders. The arrangement works because Microsoft, under its unique licensing agreements with OpenAI, sets its own resale policies and is permitted to offer these models in markets where OpenAI itself will not operate.

The mechanics matter for the regulatory picture. Microsoft does not host the models inside China. Instead, Chinese customers access the GPT APIs over the internet from Microsoft data centers in other jurisdictions, primarily Singapore, which keeps the intellectual property outside Chinese borders. This structure currently sits in a gap in US export-control law, which restricts the shipment of advanced AI chips but does not yet broadly restrict cloud-based remote access to AI capabilities. The US House of Representatives passed the Remote Access Security Act by a wide bipartisan margin, which would extend export-control rules to cover exactly this kind of remote cloud access. The bill still requires Senate passage and a presidential signature, so its ultimate fate remains uncertain. Microsoft has also recently begun cutting Azure R&D positions in Beijing and Shanghai, signaling that geopolitical pressure is beginning to reshape even its most profitable growth territory in the region.

The second thread of this story concerns AMZN and the race to reach the AI frontier. Peter DeSantis, Amazon's senior vice president overseeing semiconductor, AI, and quantum efforts, told CNBC that Amazon has not been at the frontier for the largest, most demanding AI workloads and that this is a fair criticism. DeSantis said he expects Amazon to close that gap in the "coming year," pointing to the company's Nova2 model, which has attracted roughly 50,000 customers, and to its investment in custom silicon. Amazon has also committed up to an additional $25 billion in Anthropic, its primary external AI partner, as part of a broader infrastructure deal. The combined picture is one of a hyperscaler caught between an honest admission of current limitations and a substantial bet that its cloud scale, chip investment, and Anthropic partnership can compress the gap faster than the market expects. For investors, the key question is whether DeSantis's timeline holds and whether Amazon's frontier model progress becomes visible in enterprise win rates before the next major wave of model releases from OpenAI.