Microsoft, OpenAI Shed Exclusivity, Revenue.

Mazen Al Sabbagh

— 7 min read
Share

TL;DR

  • Microsoft and OpenAI redefined their partnership: Microsoft relinquishes its revenue share, rescinds the Azure exclusivity clause, and eliminated the AGI dissolution trigger. This strategic repositioning precedes both companies' earnings reports. (Bloomberg, CNBC)
  • Jury selection commences today for Musk v. Altman: The Oakland federal court begins proceedings for a refocused four-count case seeking $150 billion in damages, with opening arguments anticipated Tuesday. (CNN, NPR)
  • Beijing vetoed Meta's $2 billion acquisition of Manus: China's government mandated the withdrawal of Meta's bid for the Singaporean AI agent startup, marking the first high-profile Chinese interdiction against a US company acquiring a Chinese-linked AI firm. (Bloomberg, CNBC)
  • OpenAI's Sora consumer application ceased operations Sunday: The app's shutdown, attributed to an estimated $1 million daily in compute costs and a halved user base, follows the dissolution of a projected $150 million Disney partnership. The API remains available through September. (The Decoder, TechCrunch)
  • Major tech earnings are due Tuesday: Alphabet, Microsoft, Meta, and Amazon report results. Analysts are scrutinizing whether over $135 billion in combined 2026 AI capital expenditure is yielding commensurate revenue, or if the market's willingness to reward speculative AI investment is nearing an inflection point. (Invezz)

Lead Story: Microsoft and OpenAI Rework Core Agreement

The foundational partnership of commercial AI just underwent a significant structural recalibration. On Sunday, Microsoft and OpenAI jointly published revised terms, dismantling three core tenets of their original 2023 agreement: Azure exclusivity, Microsoft's revenue share from OpenAI's commercial offerings, and the contentious AGI clause that permitted partnership dissolution based on self-determined AGI achievement.

Under the updated framework, OpenAI gains multi-cloud flexibility, a concession acknowledging its prior testing of workloads on Google Cloud and Oracle. Microsoft will cease collecting its share of OpenAI's revenue, previously estimated by Bloomberg at $2-3 billion annually. In exchange, OpenAI's existing payments to Microsoft will continue through 2030, capped, and Microsoft retains its IP license through 2032, now under non-exclusive terms.

The removal of the AGI clause carries significant symbolic weight. The original stipulation allowed OpenAI's board to dissolve the commercial partnership upon achieving artificial general intelligence, ostensibly to prevent a for-profit entity from controlling AGI. Its deletion signals both parties view the clause as a liability rather than a safeguard, particularly following OpenAI's internal governance turbulence in 2023-2024.

The timing is strategically precise. Microsoft reports earnings Tuesday evening, where Satya Nadella will address Azure AI growth, guided at 37-38% for Q3. This restructuring enables Microsoft to frame the partnership as mature and commercially robust, rather than dependent or uncertain. For OpenAI, multi-cloud freedom arrives as GPT-5.5's API rollout scales, a launch delayed since April 23.

Market interpretation will oscillate between this being a healthy evolution or a quiet de-coupling. Tuesday's financial disclosures will likely clarify that assessment.

In Other News

Musk v. Altman: The AI Era's Landmark Trial Commences. Jury selection begins this morning in Oakland's federal courthouse, presided over by Judge Yvonne Gonzalez Rogers, noted for Epic v. Apple. This trial could prove to be the most consequential technology litigation since the Microsoft antitrust case. Elon Musk's claims have been condensed from 26 to four: unjust enrichment, fraud, constructive fraud, and breach of charitable trust. He seeks Sam Altman's removal as OpenAI CEO and $150 billion in restitution. The jury pool size is reportedly triple the norm, reflecting the challenge of impartiality in the Bay Area. Nine jurors, no alternates. The witness roster includes Musk, Altman, Greg Brockman, and Satya Nadella, whose testimony now carries additional weight given Sunday's deal restructuring. Opening arguments are scheduled for Tuesday. The court has allocated four weeks, with deliberations anticipated around May 12. This trial directly overlaps with earnings week, creating a unique media collision where key executives will simultaneously testify under oath and present quarterly results.

Beijing Intervenes: Meta's Manus Acquisition Blocked. China's National Development and Reform Commission ordered a halt to Meta's planned $2 billion acquisition of Manus, the Singaporean AI agent startup recognized for multi-step autonomous task completion. Meta had reportedly initiated the integration of the Manus team into Reality Labs. The NDRC cited technology transfer concerns, mirroring Washington's CFIUS reviews which have routinely blocked Chinese acquisitions of US tech firms for years. This marks the first publicly confirmed Chinese governmental interdiction of a US company acquiring a Chinese-linked AI firm. The move coincides with the White House's NSTM-4 memo, which continues to reverberate, accusing China of "industrial-scale" AI distillation theft. This symmetry underscores the accelerating decoupling of US and Chinese AI ecosystems, with both governments now actively impeding cross-border AI deals.

Google-Anthropic Bet Divides Analysts. The Google-Anthropic mega-investment, disclosed late Friday, generated sharply contrasting analyses over the weekend. JPMorgan and Bank of America issued bullish notes, asserting the deal de-risks Google's TPU hardware roadmap and represents a relative bargain compared to Anthropic's $1 trillion secondary market valuation. Conversely, the bear case, articulated by several sell-side analysts, highlights that a substantial portion of Anthropic's revenue recirculates through Google Cloud, implying Google partially subsidizes its customer's expansion. Motley Fool described the $350 billion primary valuation as a "screaming bargain" against secondary prices. The deal's 5 GW compute commitment over five years is its most structurally significant component, locking Anthropic into Google Cloud infrastructure at a scale rivaling Amazon's own Anthropic compute deal. This unprecedented competitive dynamic emerges ahead of Anthropic's projected October IPO.

Sora's Closure: A $365 Million Economic Calculus for AI Video. OpenAI's Sora consumer video application ceased operations Sunday as scheduled, concluding an experiment deemed "completely unsustainable" by sources. Both the app and web interface are defunct, though the API will persist until September 24. At its peak, Sora attracted approximately one million users, a figure that halved by its shutdown. A projected $150 million partnership with Disney dissolved over quality and cost considerations. Accruing an estimated $1 million daily in compute costs—roughly $365 million annualized—Sora demonstrated that generative video at consumer scale presents an economic challenge, not merely a technological one. TechCrunch characterized the shutdown as "a reality check for AI video," noting that Runway, Pika, and other video startups are closely observing, given that Sora's economics were presumably superior due to OpenAI's scale advantages.

X / Social Pulse

The Musk-Altman proceedings drive current AI discourse, yet developer sentiment regarding GPT-5.5's initial testing phase presents a more salient indicator. CodeRabbit data shows GPT-5.5 identifying 79.2% of expected code issues against its predecessor's 58.3%, marking a notable capability improvement for code review workflows. More striking, MRCR v2 long-context evaluation at one million tokens recorded GPT-5.5 scoring 74.0% versus 36.6% for GPT-5.4, confirming the unified omnimodal architecture delivers substantial advancements in needle-in-haystack retrieval. Developer consensus identifies GPT-5.5 as a robust upgrade, not a foundational paradigm shift, closing the gap with Opus 4.7 on coding tasks without definitively surpassing it, while its long-context performance now leads the field. The API delay, however, continues to frustrate builders awaiting scaled testing.

DeepSeek V4 elicits guarded reactions following a weekend of community testing. LMSYS integrated SGLang Day-0 support, making V4 accessible on their leaderboard infrastructure, yet initial real-world reports suggest a disparity between benchmark claims and actual Huawei hardware performance, particularly regarding latency and throughput. DeepSeek responded Monday by implementing a 75% price reduction on V4-Pro through May 5, a move suggesting an intent to incentivize adoption with margin while independent benchmark verification remains pending.

One to Watch

Earnings Week Will Test Whether AI Spending Has a Revenue Story. Tuesday evening ushers in the quarter's most consequential earnings cluster, with Alphabet, Microsoft, Meta, and Amazon reporting within hours of each other. These four entities collectively allocated over $135 billion to AI capital expenditure for 2026. Microsoft guided Azure AI growth at 37-38%. Alphabet is projected to report $92.2 billion in revenue. Meta forecasts $55.5 billion (+31% YoY) while simultaneously executing 8,000 job cuts and aggressively investing in AI infrastructure. Saxo Bank's preview note articulates the market's evolving posture: investors are discontinuing the practice of rewarding AI expenditure based solely on speculative belief. The critical inquiry remains whether cloud AI revenue, representing tangible enterprise adoption of models such as Claude, GPT-5.5, and Gemini, is scaling sufficiently to rationalize the extensive capital outlay. A market disappointment would expose the AI sector, currently accounting for 45% of S&P 500 market capitalization, to its inaugural significant stress test. Conversely, positive results would intensify the investment race.

Quick Hits

  • Cognition AI (Devin) is reportedly negotiating a $25 billion valuation round, more than doubling its prior $10.2 billion assessment. Annual Recurring Revenue (ARR) is said to have surged from $1 million in September 2024 to $73 million by mid-2025. (Bloomberg)
  • FISA Section 702's 10-day extension expired today, leaving the surveillance authority in legislative limbo ahead of a potential Wednesday vote. Speaker Johnson advocates for a three-year extension without warrant requirements. (NPR)
  • DeepSeek slashed V4-Pro prices 75% through May 5, with cache hit fees cut 90%. This directly responds to lukewarm community reception on Huawei hardware, positioning V4-Pro at approximately one-tenth the cost of GPT-5.5 Pro. (Reuters, TNW)
  • GPT-5.5 API access remains delayed, with OpenAI providing no updated timeline. Enterprise customers report being informed of "weeks, not days," dampening enthusiasm following Wednesday's launch.
  • The DOJ Anthropic appeal pause awaits formal action from Judge Rita Lin. Legal observers anticipate a ruling this week that could effectively resolve the Pentagon blacklist dispute.

The forthcoming week encapsulates the most concentrated convergence of AI developments since the industry's inception. A $150 billion legal proceeding commences in Oakland, juxtaposed with the lead defendant's primary partner recalibrating their foundational agreement days before critical earnings announcements. Entities investing a cumulative $135 billion in AI infrastructure confront investors demanding quantifiable returns. A surveillance statute, critical to classified AI deployments, has lapsed. Beijing and Washington are now symmetrically impeding cross-border AI acquisitions. Concurrently, three frontier models introduced within the last five days undergo rigorous developer evaluation, an assessment that will ultimately determine their market traction. While model capabilities exhibit convergence, the broader market and geopolitical vectors show increasing divergence this week.

Sources

Lock in. M. mazen@thorterminal.com

Read more