Microsoft is accelerating its push into the AI development tools market with a new product that dramatically lowers the price of its code-generation artificial intelligence model while improving performance. Simultaneously, the company is fleshing out its custom chip strategy to fundamentally reduce AI infrastructure costs.
On the 11th (local time), Microsoft announced the release of ‘MAI Code 1.1 Flash’, a successor to its proprietary coding AI model, and confirmed its deployment on GitHub Copilot, the company’s code-generation service. This model improves upon the ‘MAI Code 1 Flash’ first introduced in June, focusing on performance enhancements for command-line interface (CLI) tasks and the .NET development environment.
The most striking change is the price. The new model is priced at $0.20 per million tokens for input and $1.20 per million tokens for output. Cached inputs are charged at $0.02 per million tokens. This represents a roughly 73% reduction compared to the previous model. Microsoft explained that it lowered the price to about a quarter of the prior model’s cost by refining its training and service operation methods.
Performance has also improved. According to Microsoft’s internal evaluations, the new model’s ‘Terminal Bench 2.1’ performance increased by 22%, while .NET task performance improved by 15%. Token generation speed is 25% faster, and the number of tokens required to complete tasks has decreased by 25%, significantly boosting efficiency. In real-world GitHub Copilot usage, the rate at which generated code is retained increased by 4%, and user return rates rose by 9%.
The model also newly incorporates image understanding capabilities. Paid Copilot users can directly select this model within Visual Studio Code and Copilot CLI. Free and student users can access the model when Copilot automatically selects the best model for a given task. Support for the existing MAI Code 1 Flash will end on September 10th.
Microsoft’s software-level cost reductions are intertwined with a larger hardware strategy. On the same day, Microsoft’s stock closed down about 0.6% at $502.61, a move interpreted as a reaction to news that the company’s custom-developed AI chip, the ‘Maia 200’, is nearing launch. Barron’s reported that Microsoft could release the next-generation Maia processor as early as September. While Microsoft has not officially confirmed this timeline or production scale, the move to fundamentally control AI infrastructure costs through custom silicon is becoming clear.
The Maia 200 is an inference chip manufactured using TSMC’s 3-nanometer process, featuring over 140 billion transistors, 216 gigabytes (GB) of high-bandwidth memory (HBM), and performance exceeding 10 petaflops (PFLOPS) at 4-bit computation. According to Microsoft, the chip delivers 30% better performance per dollar compared to the latest alternative hardware previously in use. Deployment has already begun at data centers in Iowa, with expansion planned for Arizona and other regions. This carries significant weight given that Copilot, Azure, and OpenAI workloads consume massive computational resources. If Microsoft can design more of its infrastructure in-house and increase output relative to cost, the impact could be substantial.
From an investment perspective, Microsoft’s valuation sits at an interesting point. The closing price of $502.61 on the 11th is approximately 12.69% below the GF Value estimate of $575.66 projected by some analysts. However, whether Maia translates into tangible financial results remains to be seen. Microsoft anticipates roughly $50 billion in capital expenditures this quarter alone—an enormous sum even by its standards. The bullish case rests on custom chips lowering AI costs, Azure continuing to expand, and margins improving. The bearish case, conversely, points to the risk of pouring massive funds into silicon development without achieving cost savings sufficient to justify the investment.
Ultimately, the key factor is not the chip’s technical specifications but its economics. If Maia can lower the processing cost of every AI token, Microsoft stands to build one of the most powerful weapons in the AI race.