The "Coopetition" Between OpenAI and Microsoft

“OpenAI is going to eat Microsoft alive,” says Elon Musk.

Sep 12, 2025

Welcome to The Median, DataCamp’s newsletter for September 12, 2025.

In this edition: OpenAI and Microsoft sign a new partnership memorandum and plan a PBC restructuring, Alibaba launches Qwen3-Next and Qwen3-ASR, Perplexity joins the race to power government AI with a $0.25 agency offer, French startup Mistral AI raises €1.7B to scale frontier models, and OpenAI inks a $300B cloud deal with Oracle that reshapes its partnership with Microsoft.

This Week in 60 Seconds

OpenAI and Microsoft Sign New Partnership Memorandum

OpenAI and Microsoft signed a non-binding memorandum of understanding (an agreement that outlines intent but isn’t legally enforceable) for the next stage of their collaboration. Alongside this, OpenAI plans to convert its for-profit arm into a Public Benefit Corporation (PBC), while its nonprofit will retain control and hold an equity stake worth over $100 billion. We explained in a previous issue what a PBC is and why OpenAI wants (and needs) this structure, and we’ll unpack the evolving Microsoft–OpenAI partnership in the Deeper Look section.

Alibaba Launches Qwen3-Next and Qwen3-ASR

Alibaba released two new models under its Qwen3 family. Qwen3-Next is a large language model built with a new hybrid design that helps it handle longer context more efficiently while keeping quality high. Qwen3-ASR is an upgraded speech recognition system that works better in noisy environments and comes in a “Flash” version for real-time transcription.

Perplexity Joins the Race to Power Government AI

Perplexity announced Perplexity for Government, giving U.S. federal agencies secure-by-default access to its most advanced AI models, with protections that prevent any request data from being stored. It also launched a government-tailored version of its Enterprise Pro platform at a cost of just $0.25 per agency for the first 15 months. OpenAI and Anthropic previously made similar $1 offers, part of what we explained in a past issue as a broader race to lock in government clients by running “loss leader” deals.

Mistral AI Raises €1.7B to Scale Frontier Models

French startup Mistral AI announced a €1.7B Series C round at an €11.7B valuation, led by Dutch semiconductor giant ASML. The funding will support custom frontier AI solutions for industries and government, reinforcing Mistral’s independence as a European AI player. With this, Mistral joins U.S. and Asian rivals in securing mega-rounds, a reminder that the AI race is accelerating worldwide, not just in Silicon Valley.

OpenAI Signs $300B Cloud Deal With Oracle

OpenAI agreed to purchase $300B in computing power from Oracle over five years—one of the largest cloud contracts ever signed. The deal secures long-term capacity for OpenAI’s models, underscoring the enormous infrastructure demands behind frontier AI. As a side effect, Oracle’s stock spiked more than 35%, briefly making founder Larry Ellison the world’s richest person. The move also reshapes the dynamics of OpenAI’s relationship with Microsoft—we’ll unpack what this means for the partnership in the Deeper Look section.

New Course: Software Development with Windsurf

A Deeper Look at This Week’s News

Microsoft and OpenAI: From Partnership to Power Struggle

The Microsoft–OpenAI partnership is one of the most important alliances in tech. It began with billion-dollar bets and exclusive deals, delivered breakthroughs like GPT-3 and Copilot, and nearly collapsed during the Altman ouster. Today, the two remain bound together, but their relationship looks more like “coopetition” than pure collaboration.

Milestones that cemented the alliance

OpenAI began in 2015 as a nonprofit lab, but by 2019 it needed serious compute and capital. Microsoft stepped in with a $1B investment and exclusive cloud partnership, positioning Azure as OpenAI’s home and making Microsoft the preferred channel for OpenAI’s breakthroughs.

This gave OpenAI the resources to train massive models and gave Microsoft a seat at the table in the race to AGI. On fast-forward, here are the most important milestones:

Negotiations, standoffs, and a redrawn deal

By mid-2025, Microsoft had poured $13–14B into OpenAI and held a near-half stake, but OpenAI was pushing to restructure into a conventional for-profit model and pursue an IPO.

Reports described Microsoft as “holding out” with veto power, while OpenAI considered extreme options—even declaring one of its models as AGI to break free of exclusivity clauses.

In September 2025, both sides announced a non-binding agreement to redefine their partnership, giving OpenAI more freedom to raise capital and use multiple clouds while keeping Microsoft tied in as a major investor and distribution channel.

Coopetition: Partners and rivals at the same time

The new structure shifts the relationship into coopetition. Microsoft continues to embed GPT-5 across its platforms, while OpenAI increasingly sells directly to enterprises and consumers. Microsoft invests in its own models to reduce reliance, and OpenAI diversifies its compute across Oracle, Google, and others.

Notably, OpenAI signed a massive $300 billion cloud deal with Oracle to secure long-term compute power starting in 2027. In parallel, Microsoft struck a deal to bring Anthropic’s Claude models into Office 365 apps, underlining its intent to lessen reliance on OpenAI and source AI from multiple partners.

Each depends on the other, but both are preparing for a future where they might compete head-to-head.

Musk’s provocation: “OpenAI will eat Microsoft alive”

In August, reacting to Microsoft’s rollout of GPT-5, Musk wrote on X that “OpenAI is going to eat Microsoft alive.”

Musk hints at the fact that Microsoft has become deeply dependent on OpenAI’s models to power core products like Bing Chat, Office Copilot, and Azure AI services. While Microsoft invested billions and secured preferred access, it is now locked into OpenAI’s technology roadmap. If OpenAI accelerates further, Musk sees Microsoft as a customer vulnerable to the supplier that fuels its most strategic features.

He also hints at OpenAI’s restructuring plans. By converting its for-profit arm into a Public Benefit Corporation with the nonprofit still in control—and holding an equity stake exceeding $100B—OpenAI is building financial and governance strength that could give it independence from Microsoft.

What do you think?

The future

Regulators still need to approve OpenAI’s restructuring. If successful, OpenAI’s nonprofit board could see over $100B in value, making it one of the best-funded philanthropic entities in history.

In the meantime, Microsoft and OpenAI remain deeply entwined: one provides the compute and customer base, the other the models that anchor Microsoft’s AI strategy. The new agreement will determine whether this is a long-term partnership or the start of a gradual uncoupling by 2030.

Industry Use Cases

Google DeepMind Uses AI to Sharpen Gravitational Wave Detection

Google DeepMind partnered with Caltech and the Gran Sasso Science Institute to improve the sensitivity of the LIGO gravitational-wave observatory. Their new “Deep Loop Shaping” method reduces noise in LIGO’s feedback system by 30–100 times, allowing scientists to detect more black hole and neutron star collisions. This advance could expand our ability to study cosmic events and deepen our understanding of galaxy evolution.

Educators Use Claude to Redesign Teaching and Research

Anthropic analyzed over 74,000 educator conversations on Claude.ai and surveyed Northeastern University faculty. The findings show professors use Claude most often for curriculum development, academic research, and assessing student performance. Many build custom interactive tools, from chemistry simulations to gamified quizzes, while staying cautious about automating grading. For most, Claude serves as a thought partner, augmenting creativity rather than replacing it.

Instituto PROA Uses Llama to Scale Career Prep in Brazil

In Brazil, nonprofit Instituto PROA is using Llama models hosted on Oracle Cloud to help young people prepare for job interviews. The AI assistant generates tailored reports in Portuguese on employers and openings, cutting prep time from 30 minutes to just five. PROA scaled its program by 60x, now reaching 35,000 students per year, and is exploring multimodal features to bring even richer insights to candidates.

Tokens of Wisdom

We believe in hybrid teams, and I think that's the way that things are going to grow. Hybrid teams are both humans, supercharged with AIs, and agents that need to work together with humans and other agents.

—Karen Ng, Head of Product at HubSpot

Learn more about building and managing hybrid teams made of humans and agents in our latest DataFramed podcast.