As we enter a new era based on artificial intelligence, we believe this is the best time to articulate principles that will govern how we will operate our AI datacenter infrastructure and other important AI assets around the world. We are announcing and publishing these principles – our “AI Access Principles” – today at the Mobile World Congress in Barcelona in part to address Microsoft’s growing role and responsibility as an AI innovator and a market leader.
Like other general-purpose technologies in the past, AI is creating a new sector of the economy. This new AI economy is creating not just new opportunities for existing enterprises, but new companies and entirely new business categories. The principles we’re announcing today commit Microsoft to bigger investments, more business partnerships, and broader programs to promote innovation and competition than any prior initiative in the company’s 49-year history. By publishing these principles, we are committing ourselves to providing the broad technology access needed to empower organizations and individuals around the world to develop and use AI in ways that will serve the public good.
These new principles help put in context the new investments and programs we’ve announced and launched across Europe over the past two weeks, including $5.6 billion in new AI datacenter investments and new AI skilling programs that will reach more than a million people. We’ve also launched new public-private partnerships to advance responsible AI adoption and protect cybersecurity, new AI technology services to support network operators, and a new partnership with France’s leading AI company, Mistral AI. As much as anything, these investments and programs make clear how we will put these principles into practice, not just in Europe, but in the United States and around the world.
These principles also reflect the responsible and important role we must play as a company. They build in part on the lessons we have learned from our experiences with previous technology developments. In 2006, after more than 15 years of controversies and litigation relating to Microsoft Windows and the company’s market position in the PC operating system market, we published a set of “Windows Principles.” Their purpose was to govern the company’s practices in a manner that would both promote continued software innovation and foster free and open competition.
I’ll never forget the reaction of an FTC Commissioner who came up to me after I concluded the speech I gave in Washington, D.C. to launch these principles. He said, “If you had done this 10 years ago, I think you all probably would have avoided a lot of problems.”
Close to two decades have gone by since that moment, and both the world of technology and the AI era we are entering are radically different. Then, Windows was the computing platform of the moment. Today, mobile platforms are the most popular gateway to consumers, and exponential advances in generative AI are driving a tectonic shift in digital markets and beyond. But there is wisdom in that FTC Commissioner’s reaction that has stood the test of time: As a leading IT company, we do our best work when we govern our business in a principled manner that provides broad opportunities for others.
The new AI era requires enormous computational power to train, build, and deploy the most advanced AI models. Historically, such power could only be found in a handful of government-funded national laboratories and research institutions, and it was available only to a select few. But the advent of the public cloud has changed that. Much like steel did for skyscrapers, the public cloud enables generative AI.
Today, datacenters around the world house millions of servers and make vast computing power broadly available to organizations large and small and even to individuals as well. Already, many thousands of AI developers – in startups, enterprises, government agencies, research labs, and non-profit organizations around the world – are using the technology in these datacenters to create new AI foundation models and applications.
These datacenters are owned and operated by cloud providers, which include larger established firms such as Microsoft, Amazon, Google, Oracle, and IBM, as well as large firms from China like Alibaba, Huawei, Tencent, and Baidu. There are also smaller specialized entrants such as Coreweave, OVH, Aruba, and Denvr Dataworks Corporation, just to mention a few. And government-funded computing centers clearly will play a role as well, including with support for academic research. But building and operating those datacenters is expensive. And the semiconductors – or graphical processing units (GPUs) – that are essential to power the servers for AI workloads remain costly and in short supply. Although governments and companies are working hard to fill the gap, doing so will take some time.
With this reality in mind, regulators around the world are asking important questions about who can compete in the AI era. Will it create new opportunities and lead to the emergence of new companies? Or will it simply reinforce existing positions and leaders in digital markets?
I am optimistic that the changes driven by the new AI era will extend into the technology industry itself. After all, how many readers of this paragraph had, two years ago, even heard of OpenAI and many other new AI entrants like Anthropic, Cohere, Aleph Alpha, and Mistral AI? In addition, Microsoft, along with other large technology firms are dynamically pivoting to meet the AI era. The competitive pressure is fierce, and the pace of innovation is dizzying. As a leading cloud provider and an innovator in AI models ourselves and through our partnership with OpenAI, we are mindful of our role and responsibilities in the evolution of this AI era.
Throughout the past decade, we’ve typically found it helpful to define the tenets – in effect, the goals that guide our thinking and drive our actions as we navigate a complex topic. We then apply these tenets by articulating the principles we will apply as we make the decisions needed to govern the development and use of technology. I share below the new tenets on which we are basing our thinking on this topic, followed by our 11 AI Access Principles.
Our AI Access Tenets
Fundamentally, there are five tenets that define Microsoft’s goals as we focus on AI access, including our role as an infrastructure and platforms provider.
First, we have a responsibility to enable innovation and foster competition. We believe that AI is a foundational technology with a transformative capability to help solve societal problems, improve human productivity, and make companies and countries more competitive. As with prior general-purpose technologies, from the printing press to electricity, railroads, and the internet itself, the AI era is not based on a single technology component or advance. We have a responsibility to help spur innovation and competition across the new AI economy that is rapidly emerging.
AI is a dynamic field, with many active participants based on a technology stack that starts with electricity and connectivity and the world’s most advanced semiconductor chips at the base. It then runs up through the compute power of the public cloud, public and proprietary data for training foundation models, the foundation models themselves, tooling to manage and orchestrate the models, and AI-powered software applications. In short, the success of an AI-based economy requires the success of many different participants across numerous interconnected markets.
You can see here the technology stack that defines the new AI era. While one company currently produces and supplies most of the GPUs being used for AI today, as one moves incrementally up the stack, the number of participants expands. And each layer enables and facilitates innovation and competition in the layers above. In multiple ways, to succeed, participants at every layer of the technology stack need to move forward together. This means, for Microsoft, that we need to stay focused not just on our own success, but on enabling the success of others.
Second, our responsibilities begin by meeting our obligations under the law. While the principles we are launching today represent a self-regulatory initiative, they in no way are meant to suggest a lack of respect for the rule of law or the role of regulators. We fully appreciate that legislators, competition authorities, regulators, enforcers, and judges will continue to evolve the competition rules and other laws and regulations relevant to AI. That’s the way it should be.
Technology laws and rules are changing rapidly. The European Union is implementing its Digital Markets Act and completing its AI Act, while the United States is moving quickly with a new AI Executive Order. Similar laws and initiatives are moving forward in the United Kingdom, Canada, Japan, India, and many other countries. We recognize that we, like all participants in this new AI market, have a responsibility to live up to our obligations under the law, to engage constructively with regulators when obligations are not yet clear, and to contribute to the public dialogue around policy. We take these obligations seriously.
Third, we need to advance a broad array of AI partnerships. Today, only one company is vertically integrated in a manner that includes every AI layer from chips to a thriving mobile app store. As noted at a recent meeting of tech leaders and government officials, “The rest of us, Microsoft included, live in the land of partnerships.”
People today are benefiting from the AI advances that the partnership between OpenAI and Microsoft has created. Since 2019, Microsoft has collaborated with OpenAI on the research and development of OpenAI’s generative AI models, developing the unique supercomputers needed to train those models. The ground-breaking technology ushered in by our partnership has unleashed a groundswell of innovation across the industry. And over the past five years, OpenAI has become a significant new competitor in the technology industry. It has expanded its focus, commercializing its technologies with the launch of ChatGPT and the GPT Store and providing its models for commercial use by third-party developers.
Innovation and competition will require an extensive array of similar support for proprietary and open-source AI models, large and small, including the type of partnership we are announcing today with Mistral AI, the leading open-source AI developer based in France. We have also invested in a broad range of other diverse generative AI startups. In some instances, those investments have provided seed funding to finance day-to-day operations. In other instances, those investments have been more focused on paying the expenses for the use of the computational infrastructure needed to train and deploy generative AI models and applications. We are committed to partnering well with market participants around the world and in ways that will accelerate local AI innovations.
Fourth, our commitment to partnership extends to customers, communities, and countries. More than for prior generations of digital technology, our investments in AI and data centres must sustain the competitive strengths of customers and national economies and address broad societal needs. This has been at the core of the multi-billion-dollar investments we recently have announced in Australia, the United Kingdom, Germany, and Spain. We need constantly to be mindful of the community needs AI advances must support, and we must pursue a spirit of partnership not only with others in our industry, but with customers, governments, and civil society. We are building the infrastructure that will support the AI economy, and we need the opportunities provided by that infrastructure to be widely available.
Fifth, we need to be proactive and constructive, as a matter of process, in working with governments and the IT industry in the design and release of new versions of AI infrastructure and platforms. We believe it is critical for companies and regulators to engage in open dialogue, with a goal of resolving issues as quickly as possible – ideally, while a new product is still under development. For our part, we understand that Microsoft must respond fully and cooperatively to regulatory inquiries so that we can have an informed discussion with regulators about the virtues of various approaches. We need to be good listeners and constructive problem solvers in sorting through issues of concern and identifying practical steps and solutions before a new product is completed and launched.
Our AI Access Principles
The foregoing tenets come together to shape the new principles we are announcing below. It’s important to note that, given the safety, security, privacy, and other issues relating to responsible AI, we need to apply all these principles subject to objective and effective standards to comply with our legal obligations and protect the public. These are discussed further below. Subject to these requirements, we are committed to the following 11 principles:
Provide access and support for AI developers who create models and applications
We are committed to enabling AI innovation and fostering competition by making our cloud computing and AI infrastructure, platforms, tools, and services broadly available and accessible to software developers around the world. We want Microsoft Azure to be the best place for developers to train, build, and deploy AI models and to use those models safely and securely in applications and solutions. This means:
- As we grow chip capacity, we are expanding Microsoft’s cloud computing AI infrastructure to enable the training and deployment of more foundation models, both proprietary and open source, and large and small
Today, our partnership with OpenAI is supporting the training of the next generation of OpenAI models and increasingly enabling customers to access and use these models and Microsoft’s CoPilot applications in local data centres. At the same time, we are committed to supporting other developers, training, and deploying proprietary and open-source AI models, both large and small.
Today’s important announcement with Mistral AI launches a new generation of Microsoft’s support for technology development in Europe. It enables Mistral AI to accelerate the development and deployment of its next-generation Large Language Models (LLMs) with access to Azure’s cutting-edge AI infrastructure. It also makes the deployment of Mistral AI’s premium models available to customers through our Models-as-a-Service (MaaS) offering on Microsoft Azure, which model developers can use to publish and monetize their AI models. By providing a unified platform for AI model management, we aim to lower the barriers and costs of AI model development around the world for both open source and proprietary development. In addition to Mistral AI, this service is already hosting more than 1,600 open-source and proprietary models from companies and organizations such as Meta, Nvidia, Deci, and Hugging Face, with more models coming soon from Cohere and G42.
We are committed to expanding this type of support for additional models in the months and years ahead.
- We are making AI models and development tools broadly available to software application developers around the world, so every nation can build its own AI economy
As reflected in Microsoft’s Copilots and OpenAI’s ChatGPT itself, the world is rapidly benefiting from the use of a new generation of software applications that access and use the power of AI models. But our applications will represent just a small percentage of the AI-powered applications the world will need and create. For this reason, we’re committed to ongoing and innovative steps to make the AI models we host and the development tools we create broadly available to AI software application developers around the world in ways that are consistent with responsible AI principles.
This includes the Azure OpenAI service, which enables software developers who work at start-ups, established IT companies, and in-house IT departments to build software applications that call on and make use of OpenAI’s most powerful models. It extends through Models as a Service to the use of other open source and proprietary AI models from other companies, including Mistral AI, Meta, and others.
We are also committed to empowering developers to build customized AI solutions by enabling them to fine-tune existing models based on their own unique data sets and for their specific needs and scenarios. With Azure Machine Learning, developers can easily access state-of-the-art pre-trained models and customize them with their own data and parameters, using a simple drag-and-drop interface or code-based notebooks. This helps companies, governments, and non-profits create AI applications that help advance their goals and solve their challenges, such as improving customer service, enhancing public safety, or promoting social good. This is rapidly democratizing AI and fostering a culture of even broader innovation and collaboration among developers.
We are also providing developers with tools and repositories on GitHub that enable them to create, share, and learn from AI solutions. GitHub is the world’s largest and most trusted platform for software development, hosting over 100 million repositories and supporting more than 40 million developers. We are committed to supporting the AI developer community by making our AI tools and resources available on GitHub, giving developers access to the latest innovations and best practices in AI development, as well as the opportunity to collaborate with other developers and contribute to the open source community. As one example, just last week we made available an open automation framework to help red team generative AI systems.
Ensure choice and fairness across the AI economy
We understand that AI innovation and competition require choice and fair dealing. We are committed to providing organizations, AI developers, and data scientists with the flexibility to choose which AI models to use wherever they are building solutions. For developers who choose to use Microsoft Azure, we want to make sure they are confident we will not tilt the playing field to our advantage. This means:
- We are making available public APIs to enable developers to access and use AI models we host on Microsoft Azure
The AI models that we host on Azure, including the Microsoft Azure OpenAI API service, are all accessible via public APIs. Microsoft publishes documentation on its website explaining how developers can call these APIs and use the underlying models. This enables any application, whether it is built and deployed on Azure or other private and public clouds, to call these APIs and access the underlying models.
- We are supporting a common public API to enable network operators to support software developers
Network operators are playing a vital role in accelerating the AI transformation of customers around the world, including for many national and regional governments. This is one reason we are supporting a common public API through the Open Gateway initiative driven by the GSM Association, which advances innovation in the mobile ecosystem. The initiative is aligning all operators with a common API for exposing advanced capabilities provided by their networks, including authentication, location, and quality of service. It’s an indispensable step forward in enabling network operators to offer their advanced capabilities to a new generation of AI-enabled software developers. We have believed in the potential of this initiative since its inception at GSMA, and we have partnered with operators around the world to help bring it to life.
Today at Mobile World Congress, we are launching the Public Preview of Azure Programmable Connectivity (APC). This is a first-class service in Azure, completely integrated with the rest of our services, that seamlessly provides access to Open Gateway for developers. It means software developers can use the capabilities provided by the operator network directly from Azure, like any other service, without requiring specific work for each operator.
- Developers may choose how to distribute and sell their AI models, tools and applications for deployment and use on Microsoft Azure, whether via the Azure Marketplace or directly to customers
We are committed to maintaining Microsoft Azure as an open cloud platform, much as Windows has been for decades and continues to be. That means in part ensuring that developers can choose how they want to distribute and sell their AI software to customers for deployment and use on Microsoft Azure. We provide a marketplace on Azure through which developers can list and sell their AI software to Azure customers under a variety of supported business models. Developers who choose to use the Azure Marketplace are also free to decide whether to use the transaction capabilities offered by the marketplace (at a modest fee) or whether to sell licenses to customers outside of the marketplace (at no fee). And, of course, developers remain free to sell and distribute AI software to Azure customers however they choose, and those customers can then upload, deploy, and use that software on Azure.
- We respect the needs of developers by ensuring we do not use any non-public information or data from the training, building, deployment, or use of developers’ AI models in Microsoft Azure to compete with those models
We believe that trust is central to the success of Microsoft Azure. We build this trust by serving the interests of AI developers and customers who choose Microsoft Azure to train, build, and deploy foundation models. In practice, this also means that we avoid using any non-public information or data from the training, building, deployment, or use of developers’ AI models to compete against them.
- We enable customers using Microsoft Azure to switch to another cloud provider by enabling them to easily export and transfer their data
We know that customers can and do use multiple cloud providers to meet their AI and other computing needs. And we understand that the data our customers store on Microsoft Azure is their data. So, we are committed to enabling customers to easily export and transfer their data if they choose to switch to another cloud provider. We recognize that different countries are considering or have enacted laws limiting the extent to which we can pass along the costs of such export and transfer. We will comply with those laws.
Meet our societal responsibilities
We recognize that new AI technologies raise an extraordinary array of critical questions. These involve important societal issues such as privacy, safety, security, the protection of children, and the safeguarding of elections from deepfake manipulation, to name just a few. These and other issues require that tech companies create guardrails for their AI services, adapt to new legal and regulatory requirements, and work proactively in multistakeholder efforts to meet broad societal needs. We’re committed to fulfilling these responsibilities, including through the following priorities:
- We are supporting the physical and cybersecurity needs of all the AI models and applications that run in our AI datacenters
We are committed to safeguarding the physical security of our AI data centres, as they host the infrastructure and data that power AI solutions. We follow strict security protocols and standards to ensure that our data centres are protected from unauthorized access, theft, vandalism, fire, or natural disasters. We monitor and audit our data centres to detect and prevent any potential threats or breaches. Our data centre staff are trained and certified in security best practices and are required to adhere to a code of conduct that respects the privacy and confidentiality of our customers’ data.
We are also committed to safeguarding the cybersecurity of our AI models and applications, as they process and generate sensitive information for our customers and society. We use state-of-the-art encryption, authentication, and authorization mechanisms to protect data in transit and at rest, as well as the integrity and confidentiality of AI models and applications. We also use AI to enhance our cybersecurity capabilities, such as detecting and mitigating cyberattacks, identifying and resolving vulnerabilities, and improving our security posture and resilience.
We’re building on these efforts with our new Secure Future Initiative (SFI). This brings together every part of Microsoft and has three pillars. It focuses on AI-based cyber defences, advances in fundamental software engineering, and advocacy for stronger application of international norms to protect civilians from cyber threats.
- We are applying a strong Responsible AI Standard to keep people at the center of AI design decisions and respect enduring values, including fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability
As AI becomes more pervasive and impactful, we recognize the need to ensure that our technology is developed and deployed in a way that is ethical, trustworthy, and aligned with human values. That is why we have created the Microsoft Responsible AI Standard, a comprehensive framework that guides our teams on how to build and use AI responsibly.
The standard covers six key dimensions of responsible AI: fairness; reliability and safety; privacy and security; inclusiveness; transparency; and accountability. For each dimension, we define what these values mean and how to achieve our goals in practice. We also provide tools, processes, and best practices to help our teams implement the standard throughout the AI lifecycle, from design and development to deployment and monitoring. The approach that the standard establishes is not static, but instead evolves and improves based on the latest research, feedback, and learnings.
- We are investing in initiatives to spread AI skilling broadly around the world
We recognize that countries need more than advanced AI chips and datacenters to sustain their competitive edge and unlock economic growth. AI is changing jobs and the way people work, requiring that people master new skills to advance their careers. That’s why we’re committed to marrying AI infrastructure capacity with AI skilling capability, combining the two to advance innovation.
In just the past few months, we’ve combined billions of dollars of infrastructure investments with new programs to bring AI skills to millions of people in countries like Australia, the United Kingdom, Germany, and Spain. We’re launching training programs focused on building AI fluency, developing AI technical skills, supporting AI business transformation, and promoting safe and responsible AI development. Our work includes the first Professional Certificate on Generative AI.
Typically, our skilling programs involve a professional network of Microsoft certified training services partners and multiple industry partners, universities, and nonprofit organizations. Increasingly, we find that major employers want to launch new AI skilling programs for their employees, and we are working with them actively to provide curricular materials and support these efforts.
One of our most recent and important partnerships is with the AFL-CIO, the largest federation of labor unions in the United States. It’s the first of its kind between a labor organization and a technology company to focus on AI and will deliver on three goals: (1) sharing in-depth information with labor leaders and workers on AI technology trends; (2) incorporating worker perspectives and expertise in the development of AI technology; and (3) helping shape public policy that supports the technology skills and needs of frontline workers.
We’ve learned that government institutions and associations can typically bring AI skilling programs to scale. At the national and regional levels, government employment and educational agencies have the personnel, programs, and expertise to reach hundreds of thousands or even millions of people. We’re committed to working with and supporting these efforts.
Through these and other initiatives, we aim to democratize access to AI education and enable everyone to harness the potential of AI for their own lives and careers.
- We are managing our AI data centres in an environmentally sensitive manner and using AI to advance environmental sustainability needs
In 2020, Microsoft set ambitious goals to be carbon-negative, water-positive and zero waste by 2030. We recognize that our data centres play a key part in achieving these goals. Being responsible and sustainable by design also has led us to take a first-mover approach, making long-term investments to bring as much or more carbon-free electricity as we will consume onto the grids where we build data centres and operate.
We also apply a holistic approach to the Scope 3 emissions relating to our investments in AI infrastructure, from the construction of our data centres to engaging our supply chain. This includes supporting innovation to reduce the embodied carbon in our supply chain and advancing our water positive and zero-waste goals throughout our operations.
At the same time, we recognize that AI can be a vital tool to help accelerate the deployment of sustainability solutions from the discovery of new materials to better predicting and responding to extreme weather events. This is why we continue to partner with others to use AI to help advance breakthroughs that previously would have taken decades, underscoring the important role AI technology can play in addressing some of our most critical challenges to realizing a more sustainable future.***We know that the principles governing our approach are only a first step. We expect that we will need to evolve these principles and our approach as AI technology and industry moves forward and the applicable laws and regulations change. We look forward to continuing dialogue with the many stakeholders that are now playing critical roles in building the new AI economy. If experience teaches us anything, it’s that we’ll all need to succeed together.