“The most complex corporate partnership in technology history.” That is how Dan Ives of Wedbush Securities described the Microsoft-OpenAI relationship this week. He was being diplomatic. A more precise description would be: two companies that need each other, are competing with each other, are invested in each other, and are simultaneously about to answer to separate sets of public shareholders who will have strong opinions about all of the above.”
Ground Truth · Episode 06 · June 2026On June 2nd 2026, Mustafa Suleyman — the DeepMind co-founder who became Microsoft’s CEO of AI in 2023 — walked on stage at Build 2026 in San Francisco and announced seven new AI models built entirely in-house by Microsoft. The headline model was MAI-Thinking-1: a sparse Mixture-of-Experts reasoning model with 35 billion active parameters, a 256,000-token context window, and — this is the detail that matters — trained end-to-end on commercially licensed data with zero distillation from OpenAI, Anthropic, or any other third-party model. Microsoft had spent $13 billion backing OpenAI. It had spent $5 billion backing Anthropic. It had built its entire enterprise AI strategy around selling other companies’ models through Azure. And now it had built a frontier reasoning model from scratch, specifically designed to be independent of everyone it had invested in. Welcome to Episode 06 of Ground Truth. Today we unpack the most consequential corporate realignment in AI since Google invested in Anthropic.
Seven Models, One Strategic Signal, and a Partnership That Just Changed Shape
The MAI model family. Microsoft launched seven models at Build 2026 covering the full stack of enterprise AI use cases: MAI-Thinking-1 (frontier reasoning), MAI-Code-1-Flash (5 billion parameter coding, rolled out to every GitHub Copilot subscriber on launch day), MAI-Image-2.5 and its Flash variant (image generation), MAI-Transcribe-1.5 (speech transcription), and MAI-Voice-2 (voice synthesis). Together they give Microsoft a first-party offering across every category where it has previously depended on OpenAI or other partners.
The zero-distillation claim is strategically significant. Every MAI model was trained on commercially licensed data with no distillation from any third-party model. This is not a technical flex. It is a legal and commercial positioning. Enterprise procurement increasingly demands clean training data provenance — legal teams will not sign off on deploying models whose training data lineage is uncertain or whose outputs might be tainted by third-party model distillation. Microsoft has built a model family that its legal team can defend and its enterprise customers can deploy without IP exposure. That is a procurement argument, not a benchmark argument. And in enterprise sales, procurement arguments often matter more than benchmarks.
Azure now has three model tiers. Microsoft’s Azure customers can now choose from: first-party MAI models, OpenAI models (GPT-5.5 and its variants), and over 11,000 models in the open catalog including Claude, Llama, Mistral, and Gemini. The era of Azure as an OpenAI reseller is formally over. Azure is now a multi-model marketplace where OpenAI is one option among many — an option that Microsoft also competes with directly.
$13 billion: Microsoft’s total investment in OpenAI. $5 billion: Microsoft’s investment in Anthropic. 45%: OpenAI’s share of Microsoft’s total cloud backlog. 11,000+: models available on Azure, of which 7 are now Microsoft’s own. $50 billion: OpenAI’s deal with Amazon announced March 2026, which Microsoft considered a violation of exclusivity. $110 billion: OpenAI’s March 2026 funding round that gave it enough capital to pursue infrastructure independence. Both Anthropic and OpenAI are heading to IPO answerable to public shareholders. Microsoft holds equity in both. Microsoft now competes with both. The tension is not theoretical.
The Microsoft-OpenAI Relationship: A Timeline of Entanglement
The Microsoft-OpenAI partnership began in 2019 with a $1 billion investment and an exclusive cloud agreement. Microsoft would provide OpenAI with the compute infrastructure to train its models; OpenAI would give Microsoft exclusive access to its models for productisation. The arrangement was transformative for both companies. OpenAI got the compute it needed to build GPT-3, GPT-4, and the ChatGPT product that changed the industry. Microsoft got an AI capability that it embedded across its entire product portfolio — Copilot in Office, Bing AI, Azure OpenAI Service — and became the most AI-forward of the major cloud providers as a direct result.
The partnership held through the GPT-4 era. It began showing stress in 2025 and 2026 as both companies grew large enough to have genuinely divergent interests. OpenAI raised $110 billion in March 2026, giving it the capital to pursue infrastructure independence from Azure. The company signed a $50 billion “chips-for-equity” deal with Amazon in which Amazon’s AWS becomes a primary compute provider and distribution channel — a direct encroachment on what Microsoft had understood to be its exclusive territory. Microsoft weighed a breach-of-contract lawsuit. The renegotiated 2025 agreement ultimately gave both parties more independence, but the relationship had visibly shifted from strategic alliance to managed interdependence with competitive edges.
The MAI announcement is the clearest signal yet of where that trajectory leads. Microsoft is no longer content to be the company that packages and distributes other people’s frontier AI. It wants to own the model layer, not merely the distribution layer. That ambition puts it in direct competition with the companies it has funded and the companies whose models it sells. Dan Ives called it the most complex corporate partnership in technology history. He was not exaggerating.
Microsoft invested $13 billion in OpenAI. It invested $5 billion in Anthropic. It sells both companies’ models on Azure. It just built models that compete with both. Both companies are about to go public. Their IPO prospectuses will need to address what it means that their biggest infrastructure partner is also their direct competitor. That is a genuinely novel disclosure challenge and the S-1 lawyers are earning their fees right now.Neal Lloyd · Ground Truth, Episode 06
The Power Is Concentrating Further. And Moving in a New Direction.
For enterprise buyers, more choice is real. Azure customers who previously defaulted to GPT because it was the obvious option on Azure now have genuine alternatives at every capability tier: MAI-Code-1-Flash for cost-efficient coding tasks already inside GitHub Copilot, MAI-Thinking-1 for reasoning workloads with clean IP provenance, and a 11,000-model catalog for everything else. The competitive pressure this creates will drive pricing down across the board. For the finance director signing the AI infrastructure contract, this week’s announcement is good news.
For OpenAI, the dependency asymmetry has shifted. OpenAI still represents 45% of Microsoft’s cloud backlog — that is a significant revenue relationship that Microsoft cannot afford to disrupt carelessly. But Microsoft providing that compute is a far larger percentage of OpenAI’s operational capability. OpenAI’s Amazon deal reduces that dependency, but slowly and partially. The company is in the middle of a public market process in which its single largest infrastructure partner is simultaneously its investor and its new direct competitor. The S-1 risk factor disclosure for that situation is going to be interesting reading.
For Anthropic, the position is more nuanced. Microsoft invested $5 billion in Anthropic and runs Claude on Azure. The MAI announcement does not directly threaten Claude’s Azure presence — Microsoft is explicitly running a multi-model strategy and Claude remains a premium option for enterprise customers who prefer it. But the context has changed: Anthropic is now a partner of a company that is actively building competitors to its core product, sold through the same channel, to the same customers. The Anthropic IPO prospectus will need to address the Microsoft relationship — both the $5 billion investment and the competitive dynamics — in its risk factors.
For the broader industry, the message is consolidation. The AI model layer is not a competitive market of many players. It is increasingly a competition among a small number of very large organisations — Microsoft, Google, Amazon, Meta, and the frontier labs they have funded — each of which is simultaneously partner, investor, customer, and competitor to the others. The interlocking relationships are becoming more complex, not less, as the market matures. The antitrust implications of this structure are beginning to attract serious regulatory attention in both the US and EU.
While Microsoft Had the Spotlight, Everyone Else Kept Moving
The White House issued a new AI executive order on June 2nd formally abandoning the Biden-era AI safety framework and directing federal agencies to prioritise AI adoption over AI caution. The order explicitly frames the regulatory approach of the previous administration as an obstacle to American AI leadership and directs a review of all AI-related executive actions from 2021-2025. The geopolitical framing is direct: the US must lead in AI to maintain strategic advantage over China, and over-regulation is a national security risk. The order has no immediate operational effect but sets the direction of federal AI policy for the remainder of the administration.
OpenAI expanded Codex with Sites, Annotations, and enterprise plugins, moving the AI agent beyond coding tasks into broader business workflows. The expansion positions Codex as a direct competitor to Microsoft’s GitHub Copilot — built on OpenAI models, sold through Microsoft’s platform, now competing with Microsoft’s own Copilot product. The competitive geometry is genuinely unusual.
Google’s Gemini M5 model rolled out in beta this week featuring a 128 billion parameter architecture with NPU acceleration that reduces thermal throttling by 40% compared to its predecessor — a meaningful efficiency gain for data centre deployment at scale. The efficiency story is becoming the primary battleground in 2026: not raw capability, which is now table stakes, but the cost and energy efficiency of delivering that capability at scale.
Anthropic’s IPO filing momentum continued, with the company on track for first operating profit of approximately $559 million in Q2 2026 — a milestone that meaningfully de-risks its public market story relative to OpenAI, which is not projecting profitability until 2029. Anthropic reaching operating profit before going public is a materially different proposition for institutional investors than OpenAI’s loss-funded growth story.
The AI industry in June 2026 is not a market. It is an entanglement — of capital, infrastructure, competition, and partnership so thoroughly intertwined that the standard frameworks for understanding corporate relationships barely apply. Microsoft competing with its own investees through its own platform while selling their products is not a normal market structure. It is what happens when the money is big enough and the stakes are high enough that everyone decides they cannot afford to depend on anyone else.Neal Lloyd · Ground Truth, Episode 06
Ground Truth, Episode 06 · June 2026
Neal Lloyd covers the real-world impact of AI — money, power, geopolitics, and the stories behind the headlines. Ground Truth is his daily AI news and analysis series on emdexter.blogspot.com.
- Episode 01The Gold Rush$3.6T to Wall Street
- Episode 02ChatGPT Knows EverythingDreaming V3 and the 53-day deadline
- Episode 03Siri Is Now GoogleWWDC 2026 and the EU lockout
- Episode 04America’s AI Law Is a MessThe 50-state patchwork crisis
- Episode 05Is AI Taking Your Job?The real data on displacement
- Episode 06Microsoft vs EveryoneYou are here
