On Monday April 27, Microsoft announced that OpenAI would no longer ship products exclusively on Azure. The exclusivity arrangement — the financial spine of the 2019 deal that defined Microsoft’s AI strategy — lasted six years and nine months. The next morning, AWS shipped GPT-5.5, GPT-5.4, OpenAI Codex, and Bedrock Managed Agents into limited preview. Less than 24 hours after the lock came off, the same models were running on a competitor’s cloud.

Two days later, Microsoft reported Q3 FY26: Azure +40% in constant currency, AI revenue at a $37B run rate (+123% YoY), 20 million paid Microsoft 365 Copilot seats, and CY26 capex guided to roughly $190 billion. CFO Amy Hood told investors capacity will stay constrained through 2026.

Hold both prints in your head: the moat closed, and the meter on Copilot keeps running.

What was actually exclusive

The original 2019 partnership gave Microsoft three things at once — an equity claim, an exclusive license to OpenAI IP, and an exclusive cloud distribution arrangement that confined OpenAI’s commercial workloads to Azure. The renegotiation confirmed by both sides on April 27 keeps Microsoft’s IP license through 2032 (now non-exclusive), keeps its 27% equity stake, preserves OpenAI’s $250B Azure spend commitment through 2032, and ends the cloud lock. Microsoft stops paying OpenAI a revenue share; OpenAI’s payments to Microsoft continue through 2030 at the same percentage but are now capped. The AGI escape clause — the only contractual lever OpenAI had to exit financial obligations unilaterally — was deleted.

What Microsoft kept is the cash flow architecture. What it lost is the distribution monopoly.

AWS CEO Matt Garman, introducing the Bedrock launch the next day, said the quiet part out loud: “Their production applications run in AWS. Their data is AWS. They trust the security of AWS, and we forced them, for the last couple of years, to have to get the great OpenAI models, to go to other places.” That sentence is a description of the moat the renegotiation just dissolved. Codex, with over four million weekly users, is now billable against existing AWS commitments.

The capex math just got worse

Microsoft’s CY26 capex of $190B is up from the FY26 guide of $145B; Hood told investors on the call the figure includes roughly $25B from higher component pricing. Alphabet, reporting the same day, raised its 2026 capex to $180–190B and posted Google Cloud +63% to $20B with backlog at $462B, while Pichai said the unit is “compute constrained.” Meta raised its 2026 capex to $125–145B and lost ~7% after-hours.

Combined Big Four 2026 AI infrastructure spend now sits north of $700B.

The 2019 deal was underwritten on the assumption that exclusive AI distribution would justify a structural cloud premium. AWS now serves Anthropic and the OpenAI frontier through the same Bedrock APIs. Google Cloud hosts Anthropic and grew 63% last quarter while compute-constrained. Microsoft’s cloud-AI value capture now competes against substitutable offerings on the other two clouds — the same dynamic that has been pressuring every other layer of enterprise software.

And capex went up.

What the hypothesis says about this

The hypothesis — that AI commoditizes software and collapses the SaaS model — predicts that AI-mediated software loses pricing power as the marginal cost of producing it falls. The OpenAI–Microsoft renegotiation complicates the thesis at two layers simultaneously.

At the application layer, Microsoft’s 20M Copilot seats and $37B AI run rate (+123%) are the strongest verification yet that software-as-AI captures real money. Copilot is a SaaS product priced per seat, sold into existing enterprise contracts. It is the opposite of commoditization.

At the distribution layer, the moat that underpinned that monetization — exclusive OpenAI distribution — eroded in six years and nine months and was replicated by a competitor in 24 hours. Microsoft kept the equity and the contracted spend, but the structural advantage that justified the original investment thesis has been arbitraged away. That is the commoditization mechanism, applied one level up the stack from where the hypothesis originally located it.

The hypothesis’s mechanism is correct. Its location is shifting. The cloud-AI distribution layer is commoditizing faster than the SaaS application layer. The Big Three are still capturing the seat dollars; the moat between them is no longer model exclusivity, it is integration depth and switching cost — exactly what SaaS incumbents have always relied on. The bifurcation logic from The Great Sorting gains a new axis: AI-distribution moats versus AI-application moats. Application gravity outlasted model exclusivity.

This is a complication, not a confirmation. The pattern of “AI-built software → fast commoditization” predicted by the hypothesis is happening, but it is hitting the platform layer before the application layer. That is not what the original framing assumed.

What to watch

  1. Q4 CY26 Azure AI revenue mix. If a meaningful share of Microsoft’s $37B AI run rate migrates to AWS Bedrock or Google Cloud once Bedrock GA hits, cloud-AI distribution is repricing in real time. If it doesn’t, integration depth is the new moat.
  2. Whether Microsoft trims capex. $190B was set under the prior assumption that no other cloud could host OpenAI. The first capex revision after the renegotiation is the test of how durably Microsoft believes its remaining cloud-AI advantage holds.
  3. Anthropic’s terms. Amazon now distributes both Anthropic and OpenAI; the next data point is whether Anthropic’s exclusivity and pricing on Bedrock get repriced too, indicating model-layer commoditization across foundation labs rather than just OpenAI.
  4. Copilot retention. Twenty million paid seats is the verification number. The first material churn quarter would be the falsification.

The exclusivity ran six years and nine months. AWS replicated it in twenty-four hours.

Sources: Microsoft (Apr 27); AWS Bedrock launch (Apr 28); Microsoft Q3 FY26 earnings; Microsoft Q3 FY26 transcript (The Motley Fool); Microsoft 2019 partnership announcement; Alphabet Q1 2026; Meta Q1 2026 capex (Yahoo Finance); Runtime on Garman quote; GeekWire on $250B commitment; Simon Willison on AGI clause.