Apple and Google have just made an official announcement regarding a 'multi-year agreement': Apple is set to integrate Google's Gemini large - scale model and Google Cloud technology into its Apple Intelligence and Siri voice assistant. According to reports from Reuters and The Verge, Gemini will act as the fundamental large - scale model for AI features on Apple phones. Nevertheless, Apple will persist in developing its own large - scale models, with the specific division of labor between the two entities yet to be determined. Bloomberg has reported that Apple will pay Google $1 billion annually. However, in my opinion, this figure appears to be significantly low. If billions of Apple users across the globe truly develop the habit of accessing Gemini through Apple AI, the annual value of the consumed tokens would far surpass billions of dollars. The $1 billion may merely represent a minimum threshold.

The party that is undoubtedly the most disappointed at present is OpenAI, although it should have been mentally prepared. Since its initial announcement in 2023, OpenAI has been the primary partner of Apple AI. It has been using a combination of 'locally developed small - parameter models + GPT large - parameter models' to meet user demands. However, it must be acknowledged that users in most regions around the world have been largely dissatisfied with the performance of Apple AI so far. Whether the problem lies with Apple or OpenAI is inconsequential; what matters is that Apple has decided to bring in a more mature and ecologically sound partner to address the issue.

This implies that OpenAI's greatest potential on the consumer side has vanished. By the end of 2025, ChatGPT boasts 1.2 - 1.5 billion monthly active users (MAUs) and 40 - 50 million paying users (excluding those paying through third - party APIs). This is a substantial scale, but it is still overshadowed by Apple's user base. Theoretically, if Apple AI excels, GPT could become the 'AI brain' of Apple phones, with Apple's smart devices serving as ready - made distribution channels for GPT, generating significant and sustained cash flow. However, this opportunity now belongs to Google.

Before making this decision, Apple's management repeatedly weighed various options due to the long - term underperformance of Apple AI. In reality, there were only four viable choices:

Continue to place trust in OpenAI, believing that time will resolve the issues.

Accelerate the development of self - developed large - scale models, relying solely on internal solutions.

Collaborate with other AI startups such as Anthropic or xAI.

Form a partnership with Google.

The second option was undoubtedly the first to be eliminated. Over the past three years, Apple's investment in computing infrastructure and large - model training has been minimal. While all Silicon Valley giants have been burning money, only Apple has remained passive, waiting for the fruits of AI industry maturity to fall into its lap. In the realm of generative AI, Apple's accumulation is far weaker than that of Google, Amazon, and Microsoft. Without increased investment, the gap will only widen. Apple is not shying away from fundamental R&D investment; its focus lies elsewhere, such as on self - developed chips.

The third option may seem plausible but carries significant risks. Anthropic's Claude and xAI's Grok often outperform GPT in some large - model rankings. However, their foundations remain too weak, with incomplete ecosystems. Whether they can survive against OpenAI is a big question mark. If they are acquired by larger firms in a few years, Apple will face new challenges.

This leaves only the first and fourth options. In my view, what prompted Apple's decision was the strong performance of Gemini 3, which was released in November of the previous year. Gemini 3 Flash, released in December, achieved similar effects to Gemini 3 Pro at a lower cost, addressing Apple's primary concern about inference costs. Indeed, Google has been at the forefront of reducing inference costs globally. Its deep reasoning model, launched in early 2025, had costs similar to the groundbreaking DeepSeek R1, although its radiance (luster/spotlight) was completely overshadowed by the latter.

We can say that Gemini 1 (and its predecessor Bard) was a poor large - scale model, nearly leading Wall Street to issue Google's death sentence. Gemini 1.5 was a turnaround model, regaining Google's ticket to the AI era. Gemini 2 and Gemini 2.5 were models that continuously reclaimed lost ground, restoring Google to normalcy and boosting Wall Street's confidence. Gemini 3 is a decisive model, proving that Google's foundational large - model strength ranks first globally or ties with OpenAI.

Within two months of Gemini 3's release, two major events occurred. First, as mentioned in this article, Apple chose to partner with Google. Second, Google announced collaborations with retailers like Walmart to launch AI shopping guide services. This fully demonstrates that in the field of AI large - scale models, Google has entered a virtuous cycle of 'technology - users - revenue' and may be the first among major tech firms to generate substantial revenue from AI ToC (To Consumer) applications. Adding TPU sales to third parties, my estimate is that starting from 2026, direct generative AI revenue could account for a double - digit percentage of Google's total revenue.

Meanwhile, Google's competitors are under immense pressure. The launch of Gemini AI shopping guides immediately put pressure on Amazon. Apple's shift to Gemini has placed real pressure on OpenAI. Over the past few months, OpenAI has been signing computing infrastructure contracts in a high - profile manner, giving an impression of being 'all flash and no substance.' However, this is precisely an 'offensive defense' move after recognizing the danger. Google has caught up in foundational large - model strength and far surpasses OpenAI in terms of resources and ecosystems. OpenAI can only respond by heavily investing and signing partnerships to quickly bring more allies onto its platform. Even so, the delaying effect on Google's offensive is limited.

The relationship between Apple and Google is highly nuanced. They are fierce rivals in the smartphone arena but maintain close cooperative ties in other areas. Google pays Apple substantial fees annually to promote its search engine, while Apple rents significant Google Cloud services. This collaboration is even more nuanced. Gemini has already been integrated into Android, and Apple's move is tantamount to declaring that 'AI experiences for iPhone users will not be better than those for Android users and may even lag behind.' This will severely damage the premium image that iPhones have maintained over the past decade.

Despite such significant risks, Apple's decision to partner with Google indicates the urgency of the situation, leaving little room for consideration. Had OpenAI or other large - model developers provided Apple with the same confidence, Apple would likely not have relied on Google for such a crucial matter. This is a priceless vote of confidence in Google's AI technology. Google's investors are probably already celebrating, aren't they?

As for the 'Apple self - developed large - scale model' that is rumored to continue development, no one should hold high hopes at this moment. If Apple's CEO were still Steve Jobs, he would have undoubtedly decided to go all - in on self - developed large - scale models when ChatGPT emerged in November 2022. Had that been the case, Apple's computing power procurement momentum over the past few years would likely have been among the strongest globally (and would have propelled NVIDIA's market value to new historical records even faster). However, Tim Cook is not Steve Jobs. His strengths lie in supply chain and cost control, making it difficult for him to decide at critical moments to invest in an uncertain emerging technology. By the time he saw clearly, it was already too late to invest.

Everyone is human, not a god, with strengths and weaknesses. Over the past fifteen years, Apple has fully enjoyed the dividends brought by Cook's mastery of supply chain management and must now bear the side effects of his excessive caution. The reverse is also true. Only deified individuals are flawless, but such beings exist only in fantasy novels or so - called 'documentary' fantasy novels.

So, compared to Google, which has overcome nearly all challenges and possesses advantages in full - stack AI technologies, what cards does OpenAI still hold? Interestingly, its ace has become 'user habits.' Currently, ChatGPT still retains a comprehensive user advantage over Gemini:

MAU: GPT has 1.2 - 1.5 billion, while the Gemini app has 650 million. However, including core Google users utilizing Gemini functions may exceed 1 billion.

Paying users: GPT has 30 - 40 million, while Gemini's figures are unclear but generally believed to be lower.

API: GPT may have over 1 million developers, while Gemini may have 500,000 (420,000 in the second half of 2025).

OpenAI still wields significant influence in both consumer and developer communities. After all, it took Gemini over two years to largely close the gap, a period sufficient for GPT to cultivate user habits. Personally, I have been continuously paying for GPT for 21 months and would not immediately abandon it even if I become a Gemini paying user. The same goes for professional and enterprise - level users who have relied on GPT for extended periods.

Reality is indeed dramatic. Over three years ago, OpenAI was the disruptor shaking the world with technological innovation, while Google was the incumbent holding user resources. Now, their roles seem to have reversed. The situation has changed too rapidly. Who knows what it will look like in another three years?

This article has not received funding or endorsement from Google, OpenAI, or any of their affiliated parties.

The author of this article does not directly hold equity in any of the companies mentioned but may indirectly hold stakes through funds or trust plans.