Microsoft shifts AI strategy toward its own models as costs climb

Microsoft is moving to reduce its dependence on outside AI suppliers and is increasingly leaning on models it built itself, a sign that the economics of artificial intelligence are beginning to reshape how major technology companies deploy the tools inside their products.

According to reporting from Bloomberg, Microsoft has started using its in-house MAI models to handle some user prompts in two of its most important productivity apps, Excel and Word. The change does not mean Microsoft is abandoning partners such as OpenAI and Anthropic. But it does suggest the company is trying to control the cost of serving AI features at massive scale while also building more of the underlying technology it distributes to customers.

The shift lands at a moment when many large tech firms are reassessing their AI spending. The industry has spent the past two years racing to add generative AI to consumer and enterprise products, but the operational bill for inference, model hosting and third-party licensing has become increasingly difficult to ignore. Microsoft’s move is one of the clearest examples yet of a company trying to balance the promise of AI with the economics of delivering it.

Microsoft’s push to lower AI costs

For Microsoft, the decision to route some requests to its own models is part product strategy, part cost management. The company has long been one of OpenAI’s most prominent backers and has featured OpenAI-powered capabilities heavily across Office 365, Copilot and other software offerings. Now, though, Microsoft appears to be distributing some of that workload to models created internally under the MAI label.

The idea is straightforward: if Microsoft can serve certain queries with its own systems, it can potentially reduce fees paid to external providers and gain more flexibility over product design, latency and capacity planning. It also gives the company more control over how quickly it can iterate on features in its core applications.

This is not a full break with partners. Microsoft still uses third-party models and remains closely tied to the wider ecosystem that has helped fuel its AI ambitions. But the partial shift is notable because it reflects a broader realization across Silicon Valley: the most expensive part of AI is often not building the feature, but keeping it affordable once millions of people begin using it.

What changed in Word and Excel

Bloomberg reported that Microsoft has begun deploying its own MAI models to answer a portion of prompts inside Word and Excel, two tools that sit at the center of its productivity business. Those products are among the most widely used software applications in the company’s portfolio, which makes any change to their AI back end especially significant.

Office users increasingly expect built-in AI assistance to draft text, summarize information, interpret spreadsheets and automate repetitive work. Each of those interactions can carry a cost, especially when they are routed through powerful frontier models. Diverting even a percentage of that traffic to proprietary models can produce meaningful savings at Microsoft’s scale.

It also allows Microsoft to test where its own models are competitive enough to support everyday use cases. If the company can maintain quality while lowering costs, it could strengthen the case for a more vertically integrated AI stack inside Office and related products.

The MAI models become more visible

Microsoft’s internal AI program has become more prominent in recent months. At the company’s annual Build conference in May, it introduced seven new MAI models, including an agentic coding model and a text-to-image generator. The announcement underscored that Microsoft is no longer positioning itself only as a distributor of partner models; it is also trying to become a serious model builder in its own right.

The release of those models suggested a broader ambition than simply trimming expenses. Microsoft is evidently trying to establish a layered AI architecture: some tasks can still be handled by external systems, while others can be shifted to internal models optimized for specific workloads and products.

That approach mirrors a wider industry pattern. As the cost of generative AI becomes clearer, companies are increasingly trying to reserve the most expensive models for the hardest tasks and substitute cheaper, specialized or homegrown systems wherever possible.

Why in-house models matter

In-house models can offer several advantages beyond cost. They can reduce dependence on external suppliers, lower exposure to pricing changes, and provide tighter integration with a company’s own software stack. For enterprise software companies, those benefits can matter as much as raw performance.

They also give product teams more room to tune behavior for specific user experiences. A spreadsheet assistant, for example, may not need the same broad creative range as a consumer-facing chatbot. An internal model calibrated for office productivity could be faster, cheaper and easier to manage than a frontier model designed to solve a much wider set of problems.

Still, there are trade-offs. Building and maintaining strong in-house models requires talent, infrastructure and ongoing training investment. Microsoft’s choice suggests the company believes the long-term payoff justifies the extra effort.

The economics of AI are forcing a rethink

Microsoft’s move comes amid a larger correction in the AI market. For much of the past year, the industry was dominated by aggressive experimentation and rapid deployment. Companies rushed to add AI features wherever possible, often with little immediate concern for the cost structure underneath them. That phase, at least for some firms, appears to be ending.

Industry chatter has increasingly focused on efficiency, budget discipline and model selection. The shift has been especially visible among large firms that are paying both for their own infrastructure and for access to third-party models. When AI is embedded into every customer interaction, every prompt matters financially.

As a result, some companies are now scrutinizing which model should handle which task, and whether the most expensive option is really necessary for routine use. Microsoft’s decision to rely more on its own models fits squarely into that new mindset.

From rapid expansion to cost discipline

Earlier this year, the tech sector saw a wave of aggressive AI spending that some industry observers described as a race to maximize usage at almost any cost. More recently, however, stories across the sector have highlighted a very different mood: restraint.

Rather than treating model access as an endless expansion opportunity, companies are becoming more selective. They are looking for ways to reduce token consumption, optimize inference routes, and reserve premium models for the highest-value requests. Microsoft’s reported adjustment is one of the most prominent examples of this cost-conscious turn.

The change may also reflect a maturation of AI products. Once a company has proven that a feature is viable, the next challenge is often economic durability. What worked as a demonstration can become prohibitively expensive at enterprise scale unless the provider redesigns the system around efficiency.

How Microsoft compares with other tech giants

Microsoft is not alone in tightening its AI budget. Other major technology companies have reportedly been pursuing similar strategies, seeking to control costs as generative AI services become more deeply embedded in their product lines.

Amazon, Uber, Meta and Accenture have all reportedly taken steps to curb AI-related spending or reassess how aggressively they deploy the technology. The common thread is a recognition that AI features are no longer just flashy add-ons. They are becoming core product capabilities, and that means their costs can scale quickly.

In some cases, companies are also experimenting with alternative model providers that can deliver lower prices. Reports have suggested that some firms are looking at Chinese AI models for more affordable agentic functions, even while weighing security and governance concerns. That search for cheaper performance underscores just how sensitive the market has become to AI pricing.

A broader competitive shift

For years, the most important question in AI was which company had the best model. That question still matters, but it is increasingly joined by another one: which company can deliver acceptable performance most efficiently?

Microsoft’s internal model strategy shows how product leaders are starting to think differently. If two systems can produce good enough results, the one that is cheaper, easier to integrate and more controllable may now win the contract. In other words, the competition is shifting from pure capability to a mix of capability, economics and operational control.

That dynamic could have major implications for the broader AI ecosystem, especially for model vendors that depend on heavy enterprise usage. If more customers choose to route routine work to their own systems, the market for external inference services could become more competitive and more price-sensitive.

Why this matters for OpenAI and Anthropic

Microsoft’s relationship with OpenAI has been one of the defining partnerships of the modern AI boom. The company invested heavily in OpenAI and integrated its models across a wide range of products, helping bring generative AI into mainstream office software and enterprise workflows.

Anthropic has also been part of Microsoft’s broader model ecosystem, especially as the company has experimented with different providers to support various features. The reported move to use more of its own MAI models does not signal the end of those ties, but it does suggest that Microsoft wants a more balanced portfolio.

That balance may be important for negotiating power and cost control. If a platform company has credible internal alternatives, it is in a stronger position when discussing pricing, service levels and technical requirements with external vendors.

Microsoft told TechCrunch it had nothing further to add when asked about the reported changes. The company has not publicly elaborated on the scope of the shift, how many prompts are being routed to MAI models, or whether similar changes will reach other products.

In the absence of detailed disclosure, the move should be understood as an incremental reallocation rather than a dramatic strategic break. Even so, incremental shifts at Microsoft’s scale can have outsized consequences across the AI supply chain.

What the change could mean for Office users

For most customers, the immediate effect may be invisible. If the AI response in Word or Excel remains useful, users are unlikely to care whether the underlying model comes from Microsoft, OpenAI or another vendor. What they will notice are the outcomes: speed, accuracy, reliability and whether the assistant actually helps them work faster.

Behind the scenes, however, the change could influence product performance and feature design. Microsoft may prioritize tasks that are well suited to its own models and reserve external systems for more complex or specialized requests. Over time, that could lead to a more segmented AI experience inside Office products.

There is also a business implication. If Microsoft can reduce the cost of serving AI features without weakening the user experience, it may be able to preserve margins while expanding access. In a subscription-driven business, that matters just as much as technical ambition.

Possible benefits for customers

• Faster responses for common tasks if internal models are optimized for Office workflows
• Potentially more stable availability if Microsoft can manage capacity in-house
• Improved product-specific behavior tuned to documents and spreadsheets
• Greater chance that AI features remain bundled into existing subscriptions

Possible drawbacks or risks

• Internal models may be less capable than frontier systems on certain tasks
• Users could see uneven quality depending on which model handles a request
• Microsoft may be more selective about which features receive premium AI support
• Reduced reliance on third parties could slow access to best-in-class external breakthroughs in some cases

A timeline of Microsoft’s recent AI strategy

Microsoft’s move did not happen in isolation. It follows a series of steps that show the company steadily building a more independent AI foundation.

The timeline illustrates a clear progression: from partnership-led adoption to a hybrid model and now to a more assertive internal strategy. Microsoft is still part of the same AI ecosystem, but it is no longer content to depend on it entirely.

The bigger industry signal

Microsoft’s reported decision is about more than one company’s product roadmap. It reflects a turning point for the AI economy itself. During the early hype phase, many businesses focused on demonstrating that AI could be embedded everywhere. The next phase is about determining where those capabilities can be sustained profitably.

That shift matters for the entire market. Companies that can build efficient models, compress inference costs and deliver acceptable quality at lower prices will gain an advantage. The winners may not always be the firms with the biggest systems; increasingly, they may be the firms that can make AI economical enough to use at scale.

For Microsoft, this transition is especially significant because AI is no longer a side bet. It is woven into the company’s identity across cloud, productivity software and enterprise services. Any move to lower cost while maintaining capability has direct implications for both its competitive positioning and its financial performance.

What to watch next

There are several questions to monitor in the months ahead:

1. Will Microsoft expand MAI usage beyond Word and Excel into other Office and Copilot features?
2. How much of Microsoft’s AI traffic will remain with OpenAI and Anthropic?
3. Will the company disclose more about the performance of its internal models?
4. Will other major software vendors follow Microsoft’s lead and shift more requests to proprietary systems?

The answers will help determine whether Microsoft’s move is an isolated cost-management tactic or the beginning of a broader industry reset.

Bottom line

Microsoft’s reported decision to rely more heavily on its own AI models inside core productivity apps shows that the economics of generative AI are now shaping product architecture at the highest levels of the tech industry. The company still depends on outside partners, but it is clearly trying to reduce its exposure to the rising cost of AI delivery.

In the short term, the change may be nearly invisible to most users. In the long term, it could signal a more important shift: the companies that succeed in AI will not only need powerful models, but also a credible way to serve them cheaply enough to make them sustainable.