Reddit May Halt Licensing; Grok... Movie Maker; DC Picks Moonshots

Today's AI Outlook: ☁️

Reddit Reconsiders Google’s All-You-Can-Eat Plan

Reddit may restrict Google’s access to its posts when their $60M-a-year licensing agreement expires, according to reporting cited by AI Secret. The 2024 deal allows Google to use Reddit conversations for AI training, but Google’s expanding AI Overviews increasingly answer search queries without sending users to the websites that supplied the information. Reddit shares fell 8% following the report, adding pressure ahead of its July 30 earnings announcement.

Why it matters

Reddit’s value comes from millions of people producing candid, searchable conversations. When Google licenses those discussions, converts them into AI answers and keeps users inside its own products, Reddit risks helping build a substitute for the traffic it once received.

The Deets

• Reddit receives about $60M annually under its current Google agreement.
• The company is reportedly considering limiting Google’s AI access when the deal expires.
• Reddit shares are down roughly 25% this year.
• Google-driven traffic reportedly fell 23% at Politico, about 25% at CNN and more than 85% at Business Insider over the past year.
• Reddit’s next earnings report is scheduled for July 30.

Key takeaway

AI licensing deals are becoming harder to justify when the buyer uses a publisher’s own content to reduce visits to the publisher.

🧩 Jargon Buster - AI Overviews: Google-generated summaries that appear above traditional search results and answer a query using information collected from multiple websites.

The AI Tab Comes Due

Alphabet and Tesla exceeded revenue expectations, increased their AI spending plans and still watched their shares fall after earnings. Investors appear increasingly focused on the cash required to build AI infrastructure, particularly when promised products and revenue remain several years away.

Why it matters

AI investment once served as a reliable signal that a company had an ambitious future. Wall Street is now asking how quickly that spending can become revenue, especially as data centers, chips, robotics programs and custom hardware consume cash at historic levels.

The Deets

• Alphabet raised its 2026 capital expenditure guidance to as much as $205B.
• Alphabet reported negative $5.9B in free cash flow, compared with nearly $25B a year earlier.
• Google Cloud revenue climbed 82%, but Alphabet shares still declined.
• Tesla’s capital spending rose 142% to $5.79B.
• Tesla reported negative $1.1B in free cash flow while investing in Cybercabs, Optimus and chip production.

Key takeaway

Investors still believe AI can create enormous businesses, but belief now comes with a demand for clearer timelines, stronger margins and visible returns.

🧩 Jargon Buster - Capital expenditure: Money a company spends on long-term assets such as data centers, factories, servers and specialized chips.

♟️ Power Plays

Kimi K3 Lands in a Distillation Dispute

White House science and technology chief Michael Kratsios accused China’s Moonshot AI of using distillation to train its Kimi K3 model with capabilities extracted from Anthropic’s Fable 5. The accusation arrived shortly before Moonshot’s planned publication of K3’s model weights on July 27, raising questions about whether one of China’s strongest open models was developed through unauthorized access to American systems.

Why it matters

Kimi K3 reportedly performs near the American frontier while preparing for an open-weight release. Evidence that its capabilities came from prohibited extraction could strengthen calls for tighter model access, chip controls and restrictions on how overseas companies interact with U.S. AI systems.

The Deets

• Kratsios said Moonshot frequently changed how it connected to U.S. models to conceal its activity.
• He also alleged the company obtained advanced Nvidia chips through Thailand.
• Anthropic previously accused Moonshot, DeepSeek and MiniMax of conducting “industrial-scale” capability extraction campaigns.
• Kimi K3 was released last week, with its full weights scheduled for publication on July 27.
• Anthropic policy executive Sarah Heck described the alleged activity as intellectual property theft with national security implications.
• The reporting did not include detailed public evidence proving the accusation.

Key takeaway

The Kimi K3 dispute adds intellectual property enforcement to the growing conflict over chips, open models and national AI competitiveness.

🧩 Jargon Buster - Distillation: A training process in which a smaller or newer model learns by studying the outputs of another model rather than relying only on original training data.

🛠️ Tools & Products

Grok Gives Hollywood a December Deadline

A three-minute adaptation of The Odyssey created with Grok Imagine spread widely on X after viewers struggled to distinguish it from conventionally filmed footage. Creator Heavy Pulp reportedly used two anchor frames and detailed prompts to generate a dramatic cave scene featuring Calypso and Odysseus. Elon Musk reposted the clip and said Grok would produce a full-length version of the story by the end of the year.

Why it matters

The clip demonstrated convincing faces, lighting, wind, shadows, performances and camera changes across a sustained scene. That level of visual consistency could lower production costs for independent creators while increasing pressure on studios, visual effects companies and entertainment unions.

The Deets

• The viral scene runs for about three minutes.
• Heavy Pulp used two anchor frames and detailed prompting to maintain characters and setting.
• Musk said a historically accurate, full-length Grok adaptation could arrive by year-end.
• The announcement followed the opening of Christopher Nolan’s $250M IMAX adaptation.
• Nolan’s film reportedly earned $264M worldwide after opening.

Key takeaway

Generative video is beginning to produce scenes that viewers can experience as entertainment rather than primarily as technical demonstrations.

🧩 Jargon Buster - Anchor frame: A reference image used to help an AI video system preserve a character, environment or visual style across multiple generated shots.

🚀 Funding & Startups

AMD Buys a Seat at Claude’s Compute Table

AMD agreed to support as much as 2 gigawatts of computing infrastructure for Anthropic, a chip agreement reportedly worth tens of billions of dollars. AMD will also invest up to $5B in the AI company, while Anthropic plans to use Claude to help improve AMD’s chip development.

Why it matters

The deal gives Anthropic another large source of computing capacity beyond the dominant Nvidia ecosystem. AMD gains a major frontier AI customer, a strategic investment and direct exposure to how advanced models can support semiconductor design.

The Deets

• Anthropic could deploy as much as 2 gigawatts of AMD-powered AI infrastructure.
• The chip portion of the agreement could be worth tens of billions of dollars.
• AMD plans to invest up to $5B in Anthropic.
• Anthropic will use Claude in work aimed at improving AMD’s chips.
• The partnership expands Anthropic’s access to computing capacity while strengthening AMD’s position in the AI accelerator market.

Key takeaway

Frontier AI companies are becoming strategic partners, anchor customers and investment targets for the chipmakers competing to power them.

🧩 Jargon Buster - Gigawatt: A measure of power equal to 1 billion watts, increasingly used to describe the immense electricity requirements of large AI data center projects.

🧪 Research & Models

Washington Picks 278 AI Moonshots

The Department of Energy selected 278 projects across 342 institutions for the first round of its Genesis Mission, a federal initiative designed to connect scientists with advanced AI models, agent frameworks, federal data and supercomputers. The projects will target nuclear energy, fusion, semiconductor design, critical minerals, drug research, biosecurity, power grids and advanced materials.

Why it matters

Genesis places AI inside the country’s scientific infrastructure, giving researchers access to computing and models that many institutions could not obtain independently. The program also signals that Washington increasingly views frontier AI as a tool of national research, industrial policy and energy strategy.

The Deets

• The Department of Energy received more than 5,000 proposals.
• Universities lead 168 teams, while national laboratories lead 87.
• The largest proposed award is $60M over three years for AI-assisted nuclear plant construction.
• Participants will receive access to DOE supercomputers, advanced models and agentic tools.
• Industry commitments include more than $500M, including $40M in Microsoft computing credits.
• All selections remain subject to award negotiations.
• DOE says the wider mission is intended to make American science twice as productive.

Key takeaway

The federal government is assembling AI models, computing infrastructure and scientific institutions around a coordinated national research agenda.

🧩 Jargon Buster - Agentic workflow: A process in which AI systems can plan tasks, use software, analyze results and take multiple actions with limited human guidance.

⚡ Quick Hits

• OpenAI and Anthropic are backing Australia’s push for AI regulation as legal certainty becomes another factor in global model competition.
• Anthropic contributed another $20M to an AI policy organization, expanding frontier-model regulation’s role in U.S. midterm politics.
• Meta introduced Content Seal for identifying AI-generated images, although detection currently requires a limited standalone tool.
• IBM said enthusiasm for AI hardware diverted enterprise spending away from mainframes, contributing to a 42% decline in that business.
• Monday.com is cutting about 20% of its workforce as it reorganizes around an AI Work Platform.
• Amazon eliminated positions within its AGI organization while concentrating spending on AI projects with clearer customer applications.
• OpenAI raised its planned infrastructure investment to $750B through 2030, beginning with a major data center campus in Georgia.

🔧 Tools of the Day

• Gumloop: Build a group of AI agents that audit website pages, review backlinks, identify content updates and organize the findings inside Google Sheets.
• Cursor Router: Automatically selects the least expensive model capable of completing each coding task.
• Claude Security: Anthropic’s AI security product is now available in public beta.
• Presence: OpenAI’s managed enterprise service helps companies deploy customer-facing chat and voice agents.
• Antares: Cisco’s small, open-weight security models run locally and scan code for vulnerabilities.
• Dana: Applied Intuition’s agentic development platform is designed to accelerate work on autonomous vehicles and robotics.
• Glean: A no-code agent platform focused on enterprise context, governance and workplace automation.

Today’s Sources: The Internet AI Secret, The Rundown AI