The Tech‑Reader AI Digest

Wednesday, June 17, 2026

#AI #TechNews #Digest

Story 1: OpenAI's 2025 Financials Leak — $38.5 Billion Net Loss, $17.2 Billion Paid to Microsoft

What happened: Audited financial documents for OpenAI's 2025 fiscal year were obtained by independent tech journalist Ed Zitron and independently verified by the Financial Times, providing the clearest public look at the company's books ahead of its anticipated IPO. The documents show OpenAI recorded $13.07 billion in revenue in 2025 — more than triple the $3.7 billion it posted in 2024, and ahead of its internal $10 billion target. Total costs and expenses reached $34 billion, producing an operating loss of $20.92 billion. After accounting for charges related to its 2025 conversion from a nonprofit to a for-profit entity — which generated a $41.55 billion non-cash loss tied to changes in the fair value of convertible interests and warrant liabilities — the net loss attributable to OpenAI came to $38.53 billion. OpenAI declined to comment.

The documents detail the scale of OpenAI's financial dependence on Microsoft. Of the $34 billion in total spending, OpenAI paid Microsoft $17.2 billion in 2025 — $10.59 billion for research and development expenses, widely understood to represent training compute costs, and $6.047 billion in cost-of-revenue charges covering inference. Microsoft, in turn, paid OpenAI $303 million. SoftBank paid OpenAI $867 million. At year-end, OpenAI held just over $50 billion in assets, with roughly half in cash.

The relationship between OpenAI's revenue growth and its cost structure is the most important line in the documents. Revenue tripled; gross margin declined from approximately 40% in 2024 to 33% in 2025, because more capable models consume more compute per query. OpenAI and Microsoft renegotiated their partnership in April 2026, capping total future revenue-share payments at $38 billion through 2030, down from a prior trajectory of approximately $135 billion — a restructuring that could represent $97 billion in savings over five years.

Why it matters: OpenAI filed a confidential S-1 with the SEC on or around May 22, 2026. Under SEC rules, the full public prospectus must be filed at least 15 days before the investor roadshow, meaning these figures would eventually become public regardless. What has changed today is the timing: prospective investors now have access to the audited 2025 numbers weeks or months before the official filing forces disclosure. The picture those numbers present is not a simple one. At an $852 billion private-market valuation — established in a February 2026 funding round — OpenAI trades at roughly 65 times its 2025 revenue and recorded a $20.92 billion operating loss in the same year it is seeking to go public. The non-cash restructuring charges account for the gap between the operating loss and the $38.5 billion net figure, but the operating loss is the number an IPO roadshow will have to explain.

Aaron's take — Revenue tripled. Gross margin compressed. The operating loss more than doubled. The company paid its primary landlord $17.2 billion and received $303 million back. That is the financial structure of a company making a very large bet on a cost curve that has not yet turned in its favor. The Microsoft renegotiation — capping future revenue share at $38 billion versus a prior trajectory of $135 billion — is the most important balance sheet event in the documents, and it happened after the fiscal year these numbers describe. Whether the renegotiated structure is enough to make the IPO math work at $852 billion is the question Goldman Sachs and Morgan Stanley are being paid to answer. The documents are out now. The roadshow will be when the market answers it.

Story 2: Microsoft Taps AWS to Keep GitHub Online — AI Agents Are Breaking the World's Largest Code Forge

What happened: Microsoft confirmed Tuesday that GitHub is using capacity from Amazon Web Services as a temporary measure to handle AI-driven infrastructure strain that has pushed the platform below its enterprise service-level agreement thresholds. A Microsoft spokesperson told Business Insider: "The incredible spike in agentic development that began late last year has tested our infrastructure's limits." The company added that it is "both accelerating our move to Azure and continuing to explore a multi-cloud strategy" to support continued growth. Amazon declined to name individual customers but confirmed it provides infrastructure to parties requiring global scale.

The numbers behind the crisis are precise. GitHub COO Kyle Daigle confirmed in April that the platform was processing 275 million commits per week, on pace for 14 billion in all of 2026, against 1 billion in all of 2025. AI agent-opened pull requests surged from 4 million in September 2025 to more than 17 million by March 2026 — a more than fourfold increase in six months. GitHub Actions compute minutes grew from 500 million per week in 2023 to 2.1 billion in a single week in early 2026. The platform logged nine service-degrading incidents in May 2026 and availability fell to approximately 88.4 percent in June — well below the 99.9 percent threshold written into enterprise SLAs. GitHub CTO Vlad Fedorov had already revised the infrastructure expansion target from 10x in October 2025 to 30x by February 2026, after actual agentic usage outran every forecast.

The root cause is structural. GitHub's core platform was built in 2008 on Ruby on Rails and still runs a near-two-million-line monolithic application. A human developer opens a pull request after hours or days of work. An AI coding agent opens pull requests continuously, in parallel, across multiple repositories, around the clock — with no weekends, no sleep, and no usage pattern that capacity planning models were built to anticipate. Microsoft had planned to complete GitHub's full migration to Azure by 2027. The AWS arrangement is a stopgap while that migration and a deeper architectural redesign proceed in parallel.

Why it matters: The GitHub situation is the most concrete illustration yet of what AI agent adoption looks like at infrastructure scale. The platform that Microsoft has spent years positioning as the control plane for AI-driven software development is now running on a competitor's cloud because the AI tools it hosts have outpaced its own capacity planning by an order of magnitude. Every major cloud provider is simultaneously managing demand that exceeds its build cycle. Microsoft uses AWS for GitHub. Google pays SpaceX $920 million per month for Colossus compute. Anthropic pays SpaceX $1.25 billion per month. NVIDIA Trainium chips at AWS are sold out through 2028. Goldman Sachs projects $7.6 trillion in cumulative AI capex will be required from 2026 to 2031. The demand curve is outrunning the construction cycle at every company simultaneously.

Aaron's take — Microsoft paid $7.5 billion for GitHub in 2018. It is now routing GitHub traffic through Amazon because the AI coding agents it helped create — Copilot, Claude Code, Codex — generate more commits in a week than the entire platform handled in a year in 2025. That is not a planning failure. No infrastructure team in 2023 could have forecast 14 billion commits in a single year. What it is, is the first clear measurement of what the agentic software development era looks like when it hits a monolithic platform built eighteen years ago. The question Microsoft has to answer is whether the AWS stopgap buys enough time for the Azure migration and architectural rebuild to catch up — or whether the 30x expansion target gets revised again before the year is out.

Story 3: G7 Closes in Évian — Amodei and Lutnick in the Same Room as Fable 5 Talks Continue

What happened: The 52nd G7 Summit concludes today in Évian-les-Bains, France, with AI governance among its headline agenda items. The centerpiece AI event — a dedicated working lunch Wednesday bringing G7 political leaders and approximately a dozen senior technology executives together — included Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, Cohere CEO Aidan Gomez, Mistral AI CEO Arthur Mensch, and Salesforce CEO Marc Benioff, among others.

Commerce Secretary Howard Lutnick, whose department issued the export control directive that took Fable 5 and Mythos 5 offline on June 12, was also present at the summit. Amodei and Lutnick are confirmed to have been in Évian simultaneously as working-level negotiations between Anthropic's technical staff and Commerce Department officials continued in Washington. Reuters confirmed that both men attended the summit; neither Anthropic nor the Commerce Department commented on whether they spoke directly. No deal to restore Fable 5 or Mythos 5 access was announced as of Wednesday morning.

The G7's Hiroshima AI Process has produced principles and codes of conduct since 2023 but no enforceable international regulation. The Évian summit was designed by the French presidency to move closer to concrete multilateral commitments, with youth safety, frontier AI risks in the cyber and biological domains, and AI infrastructure as the primary discussion areas. OpenAI's chief global affairs officer Chris Lehane had stated before the summit that the company expected participating firms to leave having agreed to a package of voluntary commitments.

Why it matters: The Fable 5 export control order — the most significant unilateral U.S. government action in frontier AI model distribution in history — arrived forty-eight hours before the summit opened and remained in force throughout. G7 leaders whose citizens cannot access Anthropic's newest models sat across from Amodei at a working lunch designed to discuss safe and effective AI deployment. Whether the summit produces the voluntary commitments Lehane projected, and what those commitments look like given the backdrop of a live export control dispute, is the most consequential open question as Évian closes today.

Aaron's take — Macron designed this summit to be a showcase for multilateral AI cooperation. It became, instead, a real-time case study in what unilateral AI governance looks like when a U.S. government directive lands on a Friday night and is still in force when the G7 opens Monday morning. Amodei spent three days in Évian as the CEO of a company negotiating with the Commerce Department of one of the G7 member states for permission to sell his flagship products. Lutnick spent three days in Évian as the official who sent that letter. Whatever they discussed — or didn't — will shape the regulatory precedent the industry is watching more closely than any voluntary commitment signed in France.

Quick Hits — The Rest of Today's AI World

Anthropic / Claude

Fable 5 and Mythos 5 remain offline. Amodei at G7 Évian; Lutnick also present — no public comment on direct talks. Washington-level negotiations continuing. Refund deadline for June 9–14 subscribers: June 20 at 11:59 p.m. ET.
Kalshi: 58% odds Fable 5 restored for U.S. users by July 1. Polymarket: 67% — standing news from Tuesday.
Anthropic reportedly close to operating profitability in the current quarter, per Motley Fool citing analyst estimates.

OpenAI

2025 audited financials leaked — $38.5B net loss, $13.07B revenue, $17.2B paid to Microsoft — see Story 1. Sam Altman at G7 Évian working lunch Wednesday. GPT-4.5 retirement June 27 — standing news.

xAI / SpaceX

SPCX trading Wednesday. $60B Cursor acquisition announced Tuesday — Q3 close pending regulatory approval — standing news. MSCI index inclusion mechanics ongoing.

Gemini (Google)

Demis Hassabis at G7 Évian working lunch. No new Google AI product announcements Wednesday.

Microsoft / GitHub Copilot

GitHub tapping AWS capacity to handle AI agent infrastructure surge — see Story 2. Availability at 88.4% in June, below 99.9% enterprise SLA. Azure migration target 2027.

Apple

No new announcements. WWDC26 concluded June 13 — standing news.

Nvidia

No new announcements. GB200 racks at Stargate Abilene — standing news. Vera Rubin Q3 ramp — standing news. NVIDIA Trainium chips at AWS reported sold out through 2028.

Perplexity

No new announcements today.

Ollama / LM Studio

No new announcements. Ollama 0.30 and LM Studio mlx-engine v1.8.5 — standing news from June 5.

DeepSeek / Alibaba Qwen / Z.ai

No new announcements. Chinese models at 61% of global OpenRouter developer API traffic — standing news.

Cohere / Aleph Alpha

Aidan Gomez at G7 Évian working lunch Wednesday. $20B merger pending regulatory approval — standing news.

That's your AI world for Wednesday. Back tomorrow. — Aaron

Aaron Rose is a software engineer and technology writer at tech-reader.blog.

Catch up on the latest explainer videos, podcasts, and industry discussions below.

Search This Blog

Tech-Reader.blog