In the past week, the global AI market has seen two major mergers and acquisitions.

On August 17, Bloomberg reported that Stripe, a global payments behemoth, had sealed an acquisition deal with AI infrastructure startup OpenRouter, with the transaction value exceeding USD 7 billion (roughly RMB 47.3 billion).

Just three days prior, SpaceX finalized its merger with the programming tool Cursor, with the deal valued at a staggering USD 60 billion, marking the highest acquisition amount for a global startup. This figure even eclipses the USD 44 billion that Twitter fetched in its respective acquisition.

These two high-value acquisitions prompt us to ponder: Why is the AI sector witnessing large-scale acquisitions so frequently? Has the wave of AI mergers and acquisitions truly arrived?

▌1. AI M&A Has Been Active for a Year

These two high-value mergers and acquisitions were executed at a premium, suggesting that the industry is not heading into a downturn cycle.

The acquisition price for Cursor is notably high, with a USD 60 billion transaction value even surpassing the USD 44 billion that Elon Musk paid for Twitter.

Anysphere, Cursor's parent company, was established by four MIT students in 2022. This acquisition also represents a premium of over USD 10 billion compared to its valuation in the previous funding round.

Over the past six months, Anysphere has been aggressively raising funds. In November 2025, it secured a USD 2.3 billion Series D funding round led by Accel and Coatue, with participation from Google and Nvidia, valuing the company at USD 29.3 billion post-investment. In April 2026, reports indicated that it was in talks to raise over USD 2 billion at a valuation of USD 50 billion.

On June 16, just four days after SpaceX completed the largest IPO in U.S. stock market history (raising USD 75 billion and closing with a market capitalization of approximately USD 2.1 trillion on its debut day), SpaceX exercised its previously secured acquisition option and announced the all-stock acquisition of Anysphere for USD 60 billion.

Based on Cursor's annualized revenue of approximately USD 4 billion at the time, the USD 60 billion acquisition price corresponds to an estimated valuation multiple of about 15 times revenue, one of the highest acquisition multiples in the history of AI software companies and likely a record for startup acquisition amounts.

Moreover, despite Cursor's impressive annual recurring revenue (ARR) growth, according to Ramp's enterprise spending data, its market share among enterprise clients declined from approximately 41% in June 2025 to about 26% in May 2026.

The other acquisition also involved a substantial premium.

In May of this year, OpenRouter just completed its Series B funding round, valuing the company at approximately USD 1.3 billion. This means that in less than three months, OpenRouter's valuation has increased fivefold.

According to reports, OpenRouter has an annualized revenue of USD 140 million, translating to monthly revenue of USD 12 million.

OpenRouter, a model intermediary founded by Alex Atallah in New York in 2023, provides users with API access to multiple large model companies. This allows developers to freely switch and route on-demand among over 500 large language models without the need to separately interface with dozens of model providers such as OpenAI, Anthropic, Google, and Meta.

In addition to Cursor and OpenRouter, multiple M&A deals involving companies across different value chains in the AI sector are in the pipeline.

According to reports, Anthropic is in talks to acquire AI startup Decart AI for USD 6 billion. Decart AI, founded just three years ago, primarily operates in the real-time generative AI infrastructure sector.

Looking further back, in March 2025, ServiceNow (NYSE: NOW), a global enterprise-level SaaS giant, announced the full acquisition of AI company Moveworks for USD 2.85 billion.

In June 2023, big data giant Databricks announced the acquisition of AI startup MosaicML for USD 1.3 billion. MosaicML, founded in 2021, had only publicly disclosed one funding round and employed just 62 people. In its previous funding round, it was valued at USD 220 million.

In April 2025, NVIDIA acquired LeptonAI, founded just a year earlier by Yangqing Jia, as well as a series of inference layer companies such as OctoAI, Deci, and CentML.

▌2. The M&A Wave Driven by Three Intertwined Factors

We believe there are three primary factors driving the wave of M&A activities in the AI sector.

First, industry differentiation is narrowing as giants get involved in everything.

From the initial competition in general-purpose large models to niche vertical models, and then to competition in multiple sub-products such as video models and programming functions, the survival space for startups is shrinking.

This brings us back to the age-old question: when giants create products similar to yours, how can you survive? Being acquired by a giant may currently be the best option.

For example, Cursor, the programming tool mentioned above, saw its market share decline when faced with pressure from giants like Codex and Claude Code.

For giants, acquiring a company with mature products and a certain market scale is the most time- and cost-effective way to fill gaps in specific areas.

Grok's supplementation of its programming segment, in a sense, also provided Cursor with an opportunity to sell.

Second, industry costs have not decreased despite widespread adoption.

There are two core costs in the AI sector: talent and labor, and training and inference. Both of these costs have become more expensive in recent years due to intensified competition.

The capital expenditures (CAPEX) of multiple leading companies continue to rise, making AI competition a game that only giants can afford to play in the later stages.

Third, capital is increasingly concentrated among the leading players.

As mentioned above, with the closure of the differentiation window, the likelihood of startups standing out decreases, leading to fewer opportunities to secure funding.

Institutions are now betting their money on a few companies that have already proven their commercialization capabilities, causing the financing window for the entire industry to close again.

The same is true for the Chinese market. According to IT Juzi data, in the first half of the year, the top 20 companies raised a total of RMB 156.5 billion in financing, accounting for more than half of the total financing across the sector, with a portion also coming from robotics.

▌3. Implications for the Chinese Market

The sales of Cursor and OpenRouter, as well as the strategic layouts completed by Stripe, NVIDIA, and SpaceX through M&A, offer insights for the Chinese market but are not entirely applicable.

First, the Chinese market has a different structure. Computing power and models are largely concentrated in the hands of a few leading companies such as Alibaba, ByteDance, Tencent, and Baidu. Additionally, there are no new opportunities in payment or channel distribution in the Chinese market, meaning there are few opportunities for entrepreneurship and M&A in the middle layer.

There are also companies in the Chinese market similar to OpenRouter, such as SiliconFlow, which is currently recognized as the domestic player closest to OpenRouter's positioning.

Founded in August 2023 by Yuan Jinhui, the former founder of OneFlow, SiliconFlow's core product, SiliconCloud, is known as the "Token Factory," providing access to over 170 models and serving more than 13,000 enterprise clients.

After completing a Series B funding round exceeding RMB 2 billion in June this year, it was valued at RMB 7.74 billion (approximately USD 1.08 billion).

Second, the exit paths for model companies are also limited.

Frank Zhu once judged that selling model companies to large corporations would be the best outcome. However, as mentioned above, AI differs from the internet in that it has not yet reached the stage of competing for scenarios. Given that giants are involved in everything, the value that startups can bring to large corporations is not significant enough.

After Manus blocked the possibility of selling to overseas giants, going public remains the only option.

For example, SiliconFlow has chosen to go public, having submitted its listing application to the Hong Kong Stock Exchange at the end of June, aiming to become the "first AI Token factory" to go public.

Third, after analyzing the financing situation in the AI market in the first half of 2026, IT Juzi also concluded that the large model sector may witness its first round of "elimination" in the second half of the year.

Apart from Kimi, Deepseek, and StepFun, which have raised RMB 93 billion and are highly concentrated at the top, and companies like Zhipu and Minimax that have gone public, the funding windows for the remaining 200+ model companies are rapidly narrowing. We may see the first batch of layoffs, transformations, or M&A cases among large model startups.

Finally, from these acquisitions, we can also discern an underlying message in the current competition.

AI capabilities are no longer a complete moat; paying users are.

Whether a product or feature truly has a market in the future or is just "hype" depends on acquiring genuine paying users. Conversational general-purpose large models already largely meet the needs of everyday users, but this user base is also the most fluid in the future.

However, large models for video generation and AI office models triggered by certain innovations have now become the true battlegrounds in the market. The characteristic of these products is that users are genuinely willing to pay for them.

- The End -