OpenAI began a shift from exclusive distribution to multi-cloud distribution in one day. Microsoft ended exclusivity on April 27, and OpenAI models landed on Amazon Bedrock on April 28. In the same cycle, reports said OpenAI missed internal revenue and user-growth targets in early 2026. Anthropic also shifted on security and policy: India disclosed direct talks over Mythos, and Opus 4.7 became the first Claude model with production Glasswing safeguards.
Microsoft and OpenAI end exclusive licensing, opening OpenAI's models to any cloud provider
Microsoft and OpenAI amended their partnership on April 27 to end Microsoft's exclusive license on OpenAI intellectual property, with Microsoft retaining a nonexclusive license through 2032.
The amendment suggests OpenAI is shifting from a single-provider dependency toward multi-cloud distribution.
One day later, AWS put GPT-5.5, Codex, and Managed Agents on Amazon Bedrock in limited preview through existing AWS APIs and credentials. The amendment also removed the legal risk that Microsoft could challenge OpenAI's existing $50 billion deal with Amazon.
high confidence
Watch: whether Bedrock limited preview expands to GA and whether Azure keeps first-access advantages in subsequent OpenAI releases.verify ↗
OpenAI reportedly missed its own revenue and user growth targets in early 2026
The Wall Street Journal reported that OpenAI missed its internal revenue and user growth projections in early 2026.
The report indicates competitive pressure from Anthropic and Google may be slowing OpenAI's enterprise customer growth, though OpenAI disputed the characterisation.
CFO Sarah Friar warned colleagues the shortfall could make it harder to fund the company's five-year compute agreement with Oracle. OpenAI called the report 'ridiculous' and stated it is 'totally aligned on buying as much compute as we can.'
medium confidence
Watch: for direct disclosures in earnings, filings, or official OpenAI updates that confirm or contradict the reported miss.
Anthropic releases Claude Opus 4.7 with higher-resolution vision and production-deployed Glasswing safeguards
Anthropic released Claude Opus 4.7, describing it as an improvement over Opus 4.6 in advanced software engineering tasks, with new image support up to 2,576 pixels on the long edge.
Framing Opus 4.7 as the first model carrying production Glasswing filters suggests the Mythos breach accelerated Anthropic's security deployment timeline.
Pricing is unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens. The blog notes that prompts written for earlier Claude models may produce unexpected results.
high confidence
Watch: for external reports on Glasswing false-positive and false-negative rates as production usage scales.verify ↗
India's government discloses direct talks with Anthropic over Mythos cybersecurity concerns
developing · 4×India's Ministry of Electronics and Information Technology Secretary S. Krishnan disclosed on April 28 that the Indian government is in direct talks with Anthropic over the Mythos AI model.
The disclosure indicates India has become the third major government jurisdiction — alongside the US and UK — to formally engage Anthropic over Mythos.
India is the first major democracy in Asia to publicly disclose government-level engagement on Mythos. Krishnan described cybersecurity threats from the model as a 'very real threat.'
high confidence
Watch: for announced policy demands from India or formal cross-jurisdiction coordination with US/UK responses.
Claude Opus 4.7
worth attentionAnthropic added task budgets to the Opus 4.7 API in public beta, letting developers set per-task token spend limits in agentic loops.
Use this for production agentic pipelines where runaway token spend is a real cost risk — set a task budget instead of tracking usage manually.
Source: Anthropic Official Blog →Compute commitment exposure
OpenAI committed to a five-year $300 billion compute agreement with Oracle before the reported revenue miss. DRAM prices and H200 spot rates are rising. That combination increases infrastructure pressure while demand visibility stays uncertain heading into Q1 2026 earnings.
Source: CNBC →$9.71 per gigabyte
IDC forecasts DRAM will reach this price in 2026, up from $3.76 in 2025 — the rising hardware cost backdrop for AI infrastructure commitments made this quarter.
Source: CNBC →The distribution expansion and revenue miss point to the same pressure. OpenAI opened Bedrock access after ending Microsoft exclusivity, which may help offset slower enterprise growth.
A second pressure comes from costs and governance. DRAM input costs are rising while commercial returns are being tested. At the same time, Glasswing is now in Opus 4.7 and India joined US/UK government engagement on Mythos.