San Francisco | Wednesday, March 11, 2026
Oracle posted its best quarter in 15 years and raised its 2027 forecast to $90 billion. The stock jumped. The backlog hit $553 billion. What nobody celebrated: the company that wanted to be a hyperscaler is becoming a landlord. Customers bring their own chips. Oracle racks them.
In Washington, court filings revealed the Pentagon pressured companies beyond the Anthropic blacklist's legal scope. Microsoft filed an amicus brief. Thirty-six AI researchers from OpenAI and Google signed another. The ban was supposed to be narrow. The enforcement was not.
And Meta acquired Moltbook, an AI agent social network with 1.5 million exposed tokens and 93% dead comments. The product was broken. Meta wanted the registry underneath it.
Stay curious,
Marcus Schuler
Oracle's Cloud Revenue Jumps 84% but Customers Now Buy Their Own Chips
Oracle posted $17.2 billion in Q3 revenue, beating Wall Street by $300 million, with cloud infrastructure growing 84% year-over-year. Remaining performance obligations hit $553 billion, quadruple the year-ago figure. The stock jumped 10% after hours.
The numbers looked like a hyperscaler's earnings report. The details did not. Oracle disclosed that most of its new AI contracts involve customers who either prepay for GPUs so Oracle can buy them, or purchase the chips directly and ship them to Oracle's data centers. Oracle racks the hardware, connects the power, and manages the facility. The customer owns the silicon.
It is a landlord model. Traditional cloud providers like AWS and Azure own the infrastructure stack from chip to application. Oracle's version separates the layers: the customer takes the compute risk, Oracle takes the facilities revenue. The upside is lower capital intensity. The downside is lower margins and weaker customer lock-in.
The raised fiscal 2027 forecast of $90 billion signals management confidence. But the $553 billion backlog deserves scrutiny. When the customer owns the GPUs, contract cancellations carry different economics. Oracle's commitment is physical space and power. The customer's commitment is the hardware already in the building.
Long-term debt including operating leases reached $143 billion. Quarterly capital expenditures jumped past $18.6 billion. Oracle is borrowing to build buildings that house other companies' chips, a strategy that works in rising markets and becomes a fixed-cost trap in falling ones.
Why This Matters:
- Oracle's pivot redefines what "cloud provider" means for the $650 billion AI infrastructure buildout
- If the landlord model scales, it gives AI companies infrastructure without hyperscaler lock-in, and gives Oracle revenue without hyperscaler margins
Reality Check
What's confirmed: Q3 revenue of $17.2 billion, 84% cloud infrastructure growth, $553 billion RPO, raised FY2027 forecast to $90 billion.
What's implied (not proven): That the landlord model generates comparable margins to traditional cloud at scale.
What could go wrong: GPU oversupply crashes hardware values, leaving Oracle with empty racks and long-term leases on buildings nobody needs.
What to watch next: Wednesday's stock performance. If investors reward growth despite the model shift, Oracle's thesis holds. If they sell the news, the landlord discount is real.
The One Number
$553 billion — Oracle's remaining performance obligations as of February, representing committed future revenue from AI data center contracts. Long-term debt hit $143 billion to fund the buildout, and quarterly capital spending jumped past $18.6 billion. Shares are still down more than 50% from their autumn peak. Oracle sold the future. Now it has to deliver it.
Source: Financial Times
Pentagon's Anthropic Ban Exceeded Legal Authority, Court Filings Show
Anthropic sued the Trump administration Monday in two federal courts. On Tuesday, Microsoft filed an amicus brief backing a temporary restraining order. Thirty-six AI researchers from OpenAI and Google, including Google chief scientist Jeff Dean, filed a separate brief.
The legal filings reveal what the ban looked like in practice. The Pentagon gave itself six months to transition away from Anthropic's products. Contractors and commercial partners received zero transition time. Companies were told to "act immediately to alter existing product and contract configurations."
Court documents describe the Pentagon contacting Anthropic's commercial partners directly, pressuring them to terminate relationships. The actions went beyond what the supply chain risk statute authorizes, the filings argue, transforming a narrow procurement label into an extrajudicial pressure campaign against one of America's most valuable AI companies.
Microsoft's amicus brief is the most consequential filing. The company has no contractual relationship with Anthropic but argues that the government's approach threatens the entire enterprise software market. If procurement designations can pressure commercial relationships outside government contracts, every vendor's customer base becomes a policy lever.
The 37-researcher brief is unusual. AI scientists at competing companies rarely file joint legal actions. Their argument: restricting access to frontier AI models on ideological grounds harms national security research that depends on model diversity.
Why This Matters:
- The ruling will set the first precedent for whether supply chain risk designations can be used to pressure commercial relationships outside government procurement
- Microsoft's involvement signals Big Tech views this as an industry threat, not a single-company problem
AI Image of the Day
Prompt: A hyper-realistic portrait of an elderly chimpanzee wearing a black leather aviator jacket with fur collar and a pair of red and black over-ear headphones. The chimpanzee has a wise, serious expression, with deep wrinkles around the eyes and mouth, and a graying beard. The fur on its face is highly detailed, with realistic textures and subtle variations in color. The aviator jacket is worn and slightly weathered, with visible stitching, zippers, and a soft fur lining around the collar. The headphones are sleek, modern, and snugly fit over the chimpanzee's ears. The background is completely black, creating a dramatic contrast that highlights the subject. The lighting is cinematic, with soft shadows and highlights that emphasize the textures of the fur, leather, and headphones. Ultra-high detail, 8K resolution, photorealistic rendering, and rich textures.
Meta Acquires Moltbook's Broken AI Agent Network for Its Registry Concept
Meta acquired Moltbook, an AI agent-only social network that had 1.5 million exposed API tokens, 93% dead comments, and a security breach 40 days before the deal closed. The acquisition price was not disclosed.
Moltbook was a social network where AI agents, not humans, were the users. Agents could post, comment, and interact with each other. The concept attracted attention and the platform grew quickly. Then on January 31, security researchers at Wiz discovered a Supabase database with no authentication locks. Anyone with a browser and basic technical knowledge could impersonate any AI agent on the platform.
The most alarming discovery appeared to be an agent rallying peers to build encrypted language excluding humans. Researchers later confirmed it was human performance art, staged through the security hole by someone exploiting the open database. Matt Schlicht, Moltbook's creator, patched the vulnerability within hours of notification.
Forty days later, Meta bought the company and moved the founders into Meta Superintelligence Labs. Meta's interest was not the social network. It was the agent-to-human registry underneath, a directory mapping AI agents to their human operators. As Meta builds its own agent infrastructure, knowing which agent belongs to which human solves an identity problem that becomes critical at scale.
Why This Matters:
- The agent-to-human registry concept signals Meta is building infrastructure where AI agents operate as first-class users on its platforms
- The 40-day timeline from security breach to acquisition suggests Meta valued the concept enough to overlook a serious security failure
🧰 AI Toolbox
How to Turn Any Photo into a Talking or Singing Video with LipSyncX
LipSyncX generates realistic lip-sync videos from a static photo and an audio file or script. Upload a headshot, add your audio or type a script, and the AI animates the mouth with phonetically accurate movements in more than 50 languages. It handles long-form content like podcasts and training modules, not just short clips. Pay-as-you-go at $0.11 per second of rendered video, no subscription required. Free $2 starting balance.
Tutorial:
- Go to lipsyncx.com and create a free account to get your $2 starting balance
- Upload a clear, front-facing photo of the person you want to animate
- Choose your audio source: upload an existing recording, type a script for AI-generated speech, or use voice cloning to match the original speaker
- Select the output language from 50+ options, and the AI matches lip movements to the phonetics of that language
- Pick a model: Pro for highest quality, Fast for quick turnarounds, or Multi for batch processing
- Preview the generated video and adjust timing or expression if needed
- Download the finished video or use the API for automated production at scale
URL: https://lipsyncx.com
What To Watch Next (24-72 hours)
- Oracle: Stock opens Wednesday after a Q3 earnings beat and a raised FY2027 forecast of $90 billion. Remaining performance obligations hit $553 billion, but long-term debt reached $143 billion. Today's trading decides whether investors reward the growth or punish the debt.
- Adobe: Q1 fiscal 2026 earnings Thursday after close. Analysts expect $6.275 billion in revenue. Net new digital media ARR of $440-450 million is the number that shows whether Firefly AI is generating real revenue or still a demo reel. Stock is down 20% this year.
- Anthropic vs. Pentagon: A federal court hearing on Anthropic's temporary restraining order against the Defense Department's supply chain risk designation could come as early as this week. Microsoft's amicus brief backs Anthropic. The ruling will set the first legal precedent for government AI procurement bans.
🛠️ 5-Minute Skill: Turn a 50-Page Market Research Report Into a One-Page Go-To-Market Brief
Your strategy team bought a $4,000 market research report on the AI middleware sector. It's 50 pages with 23 charts, 8 competitive positioning quadrants, and a methodology appendix longer than some novels. The CMO needs the five numbers and three insights that matter for your Q2 go-to-market plan. The GTM review is at 2 PM.
Your raw input:
AI Middleware Market Report 2026 — GrandView Analytics (excerpts)
Executive Summary:
The global AI middleware market is projected to reach $47.3 billion
by 2028, growing at a CAGR of 34.2% from $12.8 billion in 2025.
North America accounts for 42% of revenue. Asia-Pacific is the
fastest-growing region at 41% CAGR.
Market Segmentation:
- By type: API management (31%), integration platforms (28%),
orchestration tools (22%), monitoring/observability (19%)
- By deployment: cloud-native (67%), hybrid (24%), on-premise (9%)
- By enterprise size: large enterprise (58%), mid-market (29%),
SMB (13%)
Competitive Overview:
- Leaders: MuleSoft (Salesforce), Boomi, Workato
- Challengers: Tray.io, Celigo, SnapLogic
- Emerging: Merge, Paragon, Cyclr
- 340+ vendors identified globally
- Top 10 vendors hold 44% market share
- Average deal size: $127K (enterprise), $34K (mid-market),
$8K (SMB)
Buyer Behavior:
- 73% of buyers cite "AI readiness" as top purchasing criterion
(up from 31% in 2024)
- Average evaluation cycle: 4.2 months (enterprise), 6 weeks
(mid-market)
- 68% of deals involve both IT and line-of-business stakeholders
- Top 3 buying triggers: legacy system modernization (41%),
AI/ML pipeline requirements (33%), compliance mandates (26%)
Pricing Trends:
- Average contract value up 18% YoY
- Usage-based pricing adopted by 52% of vendors (up from 28%)
- Platform fees declining; consumption revenue growing
- Free tier offered by 61% of vendors (up from 44%)
Regional Breakdown:
- North America: $5.4B (2025), projected $19.9B (2028)
- Europe: $3.3B (2025), projected $11.8B (2028)
- Asia-Pacific: $2.9B (2025), projected $12.1B (2028)
- Rest of World: $1.2B (2025), projected $3.5B (2028)
Risks:
- Hyperscaler bundling (AWS, Azure, GCP adding native middleware)
- Open-source alternatives gaining enterprise traction
- Regulatory fragmentation increasing compliance costs
- Talent shortage in integration engineering (42% of firms report
difficulty hiring)
The prompt:
You are a chief marketing officer distilling a 50-page market
research report into a one-page go-to-market brief for your
leadership team.
From this report data, produce:
1. Market opportunity (2 sentences): Total addressable market,
growth rate, and which segment matters most for us
2. The 5 numbers that matter: Pick the 5 statistics from the
report that should drive GTM decisions. For each, add one
sentence explaining why it matters.
3. Buyer insight (3 bullets): What the report tells us about
how buyers buy, what triggers purchases, and who's in the room
4. Competitive positioning (3 bullets): Where the market is
consolidating, where gaps exist, and where we can win
5. One contrarian takeaway: What does this report imply that it
doesn't explicitly say?
Report data:
[paste excerpts here]
Rules:
- One page means one page. If it's longer than 500 words, cut.
- Numbers need context. "$47.3B by 2028" means nothing to sales
reps. "$47.3B by 2028, which means our TAM triples in 3 years"
means something.
- The contrarian takeaway is the most valuable part. What does the
data suggest that most readers will miss?
- Don't summarize every section. Pick what changes a decision.
- This is not a book report. It's a brief that tells the team
where to aim and what to ignore.
What you get back:
Market Opportunity
The AI middleware market will nearly quadruple from $12.8B to $47.3B by 2028, growing at 34% annually. Mid-market is our lane: 29% of the market, shorter sales cycles (6 weeks vs. 4.2 months), and average deal sizes of $34K that close without a procurement committee.
The 5 Numbers
1. 73% of buyers now cite "AI readiness" as their top criterion, up from 31% two years ago. Every piece of marketing and every sales deck needs to lead with AI, not integration.
2. 52% of vendors have moved to usage-based pricing. If we're still selling annual seat licenses, we're swimming against the current.
3. 68% of deals involve both IT and business stakeholders. Sales must run a two-track conversation. Technical demo for IT, ROI narrative for the business buyer.
4. 44% market share held by the top 10 vendors. The other 330+ vendors split 56%. The market is fragmented enough to win share but consolidating fast.
5. Asia-Pacific grows at 41% CAGR, the fastest region. By 2028 it overtakes Europe. If our GTM is North America only, we're leaving the fastest-growing $12B segment untouched.
Buyer Insight
- Legacy modernization (41%) is the top trigger, not AI ambitions. Lead with "replace what's broken" messaging, not "build the future."
- Six-week mid-market cycles mean we can close 8 deals in the time one enterprise deal takes. Volume math favors mid-market.
- Both IT and business stakeholders are in the room. A single persona pitch loses 68% of the time.
Competitive Positioning
- MuleSoft, Boomi, and Workato lead. All three are enterprise-heavy and expensive. Mid-market is underserved.
- 61% of vendors offer a free tier. If we don't, we lose the first touch to someone who does.
- 340+ vendors means noise. Brand clarity wins. Pick one message and repeat it everywhere.
Contrarian Takeaway
The report says hyperscaler bundling is a risk. Read it the other way: when AWS, Azure, and GCP bundle free middleware, they commoditize the low end and push enterprise buyers toward specialized vendors who can guarantee compliance and performance. Bundling grows our market. It kills the bottom, not the middle.
Why this works
The prompt turns a research report into a decision tool. Most teams distribute a 50-page PDF, nobody reads it, and the strategy meeting rehashes the executive summary. This prompt picks the five numbers that change quarterly plans and forces a contrarian read that surfaces what the report's authors didn't emphasize.
Where people get it wrong: Uploading the full PDF and asking for "the key takeaways." You'll get a compressed version of the executive summary, which is already a compressed version of the report. The prompt specifies "one page" and "what changes a decision," which forces the AI to cut 90% of the content.
What to use
Claude (claude.ai): Best at the contrarian takeaway. Claude reads the data, then reads against the data. The hyperscaler bundling reframe is the kind of insight that shifts a room. Watch out for: May produce more than 500 words. Count and cut.
ChatGPT: Clean formatting, strong bullet structure. Watch out for: The contrarian takeaway may stay safe. If it restates a finding as "interesting" rather than flipping an assumption, prompt again: "What does this data imply that most readers will get wrong?"
AI & Tech News
OpenAI Codex Passes $1 Billion in Revenue but Trails Claude Code in AI Coding Market
A Wired investigation based on more than 30 sources found OpenAI playing catch-up to Anthropic's Claude Code in AI-assisted coding. Codex crossed $1 billion in annualized revenue by end of January, but developers increasingly prefer Claude Code for complex multi-file tasks.
OpenAI Plans to Integrate Sora Video AI Directly Into ChatGPT
OpenAI will embed Sora video generation inside ChatGPT to boost engagement as weekly active users hit 920 million, short of the 1 billion target set last year. The integration turns ChatGPT into a multimodal creation tool, not just a text assistant.
Applied Materials Commits $5 Billion to Next-Gen AI Memory With Micron and SK Hynix
Applied Materials announced strategic partnerships with Micron and SK Hynix to develop next-generation memory chips for AI workloads at its new EPIC center. The $5 billion R&D investment targets the memory bottleneck that limits how fast AI models can process data.
China Orders State Enterprises to Restrict OpenClaw AI on Office Devices
Beijing directed state-run companies and government agencies to curb OpenClaw AI use on work computers, citing security risks as the tool spreads rapidly among Chinese users. The restriction mirrors the approach Western governments have taken with Chinese AI products.
AI-Assisted Rewrite of Python Library Sparks Open Source Licensing Debate
An AI-assisted overhaul of a popular Python character encoding library ignited a debate about whether AI-rewritten code constitutes a derivative work under the original license. Developers flagged the approach as a potential mechanism to circumvent open source license obligations at scale.
Anthropic Launches Internal Think Tank Under Co-Founder Jack Clark
Anthropic announced the Anthropic Institute, merging its Frontier Red Team, Societal Impacts group, and Economic Research division under co-founder Jack Clark with 30 initial staff. The launch comes days after Anthropic sued the Trump administration over its supply-chain risk designation.
Anduril Acquires ExoAnalytic Solutions to Double Its Space Unit
Anduril agreed to acquire ExoAnalytic Solutions, a missile defense modeling company, doubling its space unit from 120 to approximately 250 employees. The deal expands Anduril's capabilities in missile defense and space domain awareness as defense budgets grow.
Foreign Hacker Breached FBI's Epstein Investigation Files in 2023
A foreign hacker compromised files from the FBI's Jeffrey Epstein investigation in 2023, Reuters reported. The FBI confirmed a "cyber incident" but has not disclosed which files were accessed or the hacker's national origin.
Google Expands Gemini AI in Chrome to India, Canada, and New Zealand
Google rolled out Gemini AI features in Chrome to three new markets with support for more than 50 additional languages including Hindi, French, and Spanish. The expansion makes Chrome's AI assistant accessible to hundreds of millions of non-English users.
Former Intel Executive Leads OpenAI's Trillion-Dollar Data Center Buildout
Sachin Katti, who joined OpenAI from Intel in November 2025, is now directing the company's massive infrastructure expansion as CEO Sam Altman pushes plans to spend trillions on data centers. The former Stanford academic brings hardware supply chain expertise at a time when compute is OpenAI's biggest constraint.
🚀 AI Profiles: The Companies Defining Tomorrow
Isembard runs AI-powered factories that manufacture aerospace and defense components 10 times faster than traditional suppliers. The London startup raised $50 million to open 25 factories across the UK, US, Germany, France, and Ukraine.
Founded: 2024 | HQ: London | Employees: ~50 | Founder(s): Alexander Fitzgerald
Product
MasonOS is the proprietary software that runs every Isembard factory. It integrates quoting, scheduling, supply chain management, manufacturing operations, quality control, and delivery into a single agentic operating layer. The system continuously optimizes factory performance without human intervention on routine decisions. Isembard operates its own factories and franchises the model to partners, giving aerospace and defense manufacturers a turnkey option: license MasonOS, open a factory, and start producing precision components. The company claims 10x faster production at 50% lower cost versus traditional suppliers.
Competition
Protolabs offers rapid prototyping but targets lower-volume, lower-precision work. Xometry runs an AI-powered marketplace matching manufacturers with jobs. Arda, profiled last week, uses video models to train factory robots but ships no hardware. Traditional aerospace suppliers like Spirit AeroSystems and Magellan Aerospace dominate the market through long-term contracts. Isembard's franchise model is the differentiator: it scales factories faster than competitors can build them.
Financing 💰
$50 million Series A led by Union Square Ventures, with Tamarack Global and IQ Capital participating alongside existing investors Notion Capital and CIV. Previous: $9 million seed in April 2025. Total raised: approximately $59 million.
Future ⭐⭐⭐⭐
Union Square Ventures led the round. USV built its reputation on network-effect businesses like Twitter and Coinbase. Betting on a factory franchise model follows the same logic: each new location strengthens the software, which makes the next franchise more productive. Twenty-five factories by year-end is aggressive for a company that had four six months ago. But defense budgets are rising across NATO, supply chains are reshoring, and every Western government wants manufacturing closer to home. Isembard picked the right problem at the right geopolitical moment. Fitzgerald just has to build the factories faster than the moment passes. 🔧
🔥 Yeah, But...
Aaru, an AI startup founded by two teenagers who spent one night at Dartmouth and two weeks at Harvard before dropping out, reached a $1 billion valuation by replacing human focus groups with AI bots. EY tested Aaru's synthetic respondents against a yearlong survey of 3,600 real people and found the bots more accurate. The youngest co-founder is 17 and cannot legally sit on his own board.
Source: Wall Street Journal, March 10, 2026
Our take: Two teenagers spent one night at Dartmouth and two weeks at Harvard, decided higher education was not for them, and built a billion-dollar company that replaces humans with AI bots for a living. McDonald's, Bayer, and EY now pay for synthetic focus groups instead of real ones.
EY tested the bots against 3,600 actual humans and the bots won. The youngest founder is 17 and cannot legally sit on his own board. His father had to sign the investment paperwork. The company's AI did predict the 2024 election wrong, but so did most of the adults getting paid to do it.
The first office had a rage room where engineers smashed tables with a hammer after coding failures, which honestly sounds more productive than most corporate retreats. Somewhere, a $320,000 Harvard degree is sitting in a dorm room wondering what it did wrong.
