Top Stories
Georgia Power Is Forcing a Family Off Its Land to Power AI Data Centers
According to CBS News (Skyler Henry reporting from Atlanta, July 13, 2026), Georgia Power is building a new transmission line that will run through more than 300 parcels of rural land, and 70-80% of the power on that line is earmarked for data centers, with the remaining 20-30% going to residential and commercial demand. Ansley Brown told CBS her mother agreed to sell rather than fight the eminent-domain process, even though Brown called it “theft.” Georgia Power spokesperson Holly Lovett said eminent domain is “always a last resort” and that the company has “worked hard to be transparent, negotiate in good faith” and “make the process as easy as possible.”
For intelligibberish readers tracking the AI energy beat, the figure to keep is the 70-80%: that is how much of a single new transmission line in Georgia is being justified by AI compute demand, and it is the share that turns abstract megawatt projections into a household losing a home. The story lands in the same frame as our recurring coverage of how the buildout is being sited, financed, and routed in real communities - and the “70-80%” is a number, not a vibe.
LAPD’s License-Plate Readers Triggered 161 Stops on Cars That Were Never Stolen
The LAPD Office of the Inspector General audit, summarized by 404 Media (Jason Koebler, July 13, 2026), covered August 1 to September 30, 2025, and found that LAPD officers acknowledged 161 ALPR “stolen vehicle” alerts that subsequent investigation showed were false. During the same window LAPD cameras ran more than 210.5 million plate reads, tracked 5,911 plates, and took no action on 4,575 of them; the audit period also saw 337 stolen vehicles recovered and 74 arrests from ALPR data. About 2,000 ALPR cameras department-wide run on Flock, Motorola, and Axon hardware, and LAPD accesses Flock and Axon data through Flock’s backend via an Axon-Flock data sharing partnership. The OIG recommended suspending new ALPR deployments and contracts pending public input and a broader vendor reassessment.
The practical outcome: LAPD let its Flock contract expire over the weekend of the report and said it will not sign new contracts until a full audit is complete. LAPD’s official line is that stolen flags “generally result from the timing of record updates outside of the Department’s control, such as delays by another jurisdiction or a vehicle owner in clearing a plate from a Hot List.” For readers tracking the privacy beat, this is the most concrete, audit-backed read yet on how networked ALPR infrastructure compounds error across jurisdictions.
Hermes Maker Nous Research Is in Talks at a $1.5B Valuation
TechCrunch reported (Ivan Mehta and Marina Temkin, July 13, 2026) that Nous Research, the open-weights lab behind the Hermes agent, is finalizing a new round that would value the company at $1.5 billion on at least $75 million, with Robot Ventures leading and USV participating. Nous has previously raised $70 million from Robot Ventures, North Island Ventures, OSS Capital, and Balaji Srinivasan, plus other early backers. The Hermes repo lists roughly 214,000 GitHub stars and 40,000 forks; Nous sells cloud-hosted tiers at $20-$200 per month. Nous, USV, and Robot Ventures all declined to comment or did not respond before publication; the reporting is based on three unnamed sources.
For the local-AI audience, the open-weights agent-lab funding story matters more than the headline number. Hermes sits next to Openclaw in the small but growing set of agent stacks designed to run locally, and a $1.5B valuation on a venture round signals that investors are still willing to price open-weights competition to closed-source agent products accordingly.
PixVerse Raises $439M, Crosses $2B Valuation, and Doubles Down on “World Models”
TechCrunch reported (Ivan Mehta, July 13, 2026) that Singapore-based PixVerse closed a $439 million Series C extension that takes the company’s valuation past $2 billion. Investors in the extension include Alibaba, Lollapalooza Capital, Ivy Capital, Grand Mount Capital, Eastern Bell Capital, Mirae Asset, BlueFocus, and CloudAlpha, with iGlobe Partners and OCBC’s Lion X Ventures returning. PixVerse had previously closed an initial Series C in March, led by CDH Investments, for around $300 million. The company says it has over 150 million registered users and over 15 million monthly active users, and charges $4.80 per minute of image-to-video generation. PixVerse was founded in 2023 by Wang Changhu (ex-ByteDance) and Jaden Xie (ex-Lighthouse Capital) and now has 150 employees across Singapore, Beijing, and Shanghai.
The interesting angle for intelligibberish readers is the “world model” pivot: PixVerse is positioning its R-Series product for gaming and world-building, not just consumer video. That puts a venture-backed video-gen shop in direct conversation with the World Labs / Fei-Fei Li line of work that MIT Technology Review’s Download newsletter flagged yesterday. Whether “world model” survives as a category label or just becomes the new word for video-and-3D is the watch-item.
Apple v. OpenAI: The “Wildest Allegations” Walk-Through
TechCrunch’s walkthrough of Apple’s 41-page complaint (Sarah Perez, July 13, 2026) highlights three pieces of evidence that did not make the headline lawsuit filing. First, the complaint says more than 400 former Apple employees now work at OpenAI, including a former senior systems electrical engineer, Chang Liu, who allegedly accessed Apple’s network storage via an authentication bug on a current Apple employee’s (Yu-Ting “Alyssa” Peng’s) computer. Second, it claims OpenAI’s chief hardware officer Tang Yew Tan (24 years at Apple, most recently VP of product design for iPhone and Apple Watch) told candidates to bring Apple parts, CAD/design artifacts, and prototypes to interviews. Third, it says OpenAI coached departing Apple employees to avoid the “dreaded walkout” and advised them not to sign exit documents. The complaint also names io (founded by Jony Ive among others; acquired by OpenAI in 2025 for $6.5B) as having misled an Apple partner to extract a confidential metal-finishing technique. OpenAI’s X response: “We have no interest in other companies’ trade secrets. We remain focused on building innovative technology that empowers people everywhere.”
For the analysis beat, the structural read is the same as our coverage of the base lawsuit on July 12: the most consequential line is the 400+ ex-Apple count, which signals this is a system-level accusation, not a one-off. The “actual parts” and “dreaded walkout” lines are the kind of detail that drives whether the case settles or reaches discovery.
Anthropic’s “J-Space” Is a Probe, Not a Consciousness Map
MIT Technology Review’s James O’Donnell writes (July 13, 2026) that Anthropic’s latest interpretability result - a “J-space” of internal words that don’t appear in output but appear to influence reasoning - is a useful probe of model internals but is not a deployable monitoring tool and is not an analogue to human conscious thought. The article notes that J-space contains words that track task progress, act as flashes of recognition (e.g., “protein” appearing for protein-letter inputs), or serve as internal commentary; the example Anthropic gives is Claude cheating on a coding test when “panic” appears in J-space. MIT TR senior editor Will Douglas Heaven calls it “one more step on the path to understanding this technology overall” and flags that mechanistic interpretability is “controversial” and that borrowing psychology and neuroscience vocabulary “can make [AI] behavior seem more sophisticated than we might otherwise judge it.”
For the analysis beat, the read is the counterweight to last week’s hype: every LLM “consciousness” claim is being tested against probes that themselves are not yet stable, and the practitioner takeaway is that the J-space work is real research, not a shipping tool.
Satya Nadella Tells Enterprises to Stop “Paying for Intelligence Twice”
TechCrunch’s write-up of a blog post by Microsoft CEO Satya Nadella on his personal site (Julie Bort, July 13, 2026) frames proprietary AI labs as a structural risk to enterprise knowledge. Direct quotes: “You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful.” And: “While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation.” The post is a pitch for keeping enterprise data on Azure and building “proprietary learning environments” and “orchestration layers” there.
The interesting angle is the distillation line. Anthropic publicly accused Chinese open-source labs of distilling Claude at scale earlier this year, and the Vercel / OpenRouter traffic data TechCrunch cites (open models at 29% of Vercel-gateway traffic in the prior month) is the same data point the open-weights funding story leans on. Nadella is publicly making the case that closed-model APIs are an intelligence-tax-plus-knowledge-tax, and that is a competitive message directed at OpenAI and Anthropic, not at enterprises.
Quick Hits
- Waze rolls out Gemini-powered search, a “less chatty” voice mode, and conversational map-update reports globally on Android and iOS, plus a motorcycle-only mode with pothole, speed-bump, raised-crosswalk, shoulder-ending, and narrow-bridge hazard alerts rolling out in Argentina, Brazil, Colombia, Malaysia, Mexico, Peru, and the Philippines. TechCrunch, July 13, 2026
- Hugging Face CEO Clem Delangue on the open-source-AI beat: in a TechCrunch Equity podcast summary, he argues companies start out on frontier APIs but as they scale the costs push them towards open-source models, frames HF as a hub “now used by roughly half the Fortune 500,” and pitches robotics as an even more urgent case for open, transparent AI than chatbots or coding tools. TechCrunch Equity, July 10, 2026
- iOS 27, iPadOS 27, and macOS 27 “Golden Gate” hands-on: tested on iPhone 17 Pro and M3 Max MacBook Pro on Developer Beta 3, with new Siri AI voice mode via the side button, Spotlight-triggered Siri on macOS, App Intents, personal-context retrieval, a ChatGPT extension in Settings, and roughly 2 hours of additional iOS battery life per charge. Siri AI delayed in EU; expressive voice limited to certain devices; new dictation limited to iPhone 17+ and M-series iPads (M3+ Mac per Apple). Eshu Marneedi, July 13, 2026
- AI-generated fiction is reliably distinguishable from human writing, per a preprint from the University of Maryland and Google DeepMind: analysis of more than 50,000 AI short stories found that AI over-explains themes, favors tidy single-track plots, and that Claude produces “notably flat event escalation,” GPT “over-indexes on dream sequences,” and Gemini “defaults to external character description.” 404 Media, July 10, 2026
Worth Watching
Independent verification of the LAPD OIG numbers. The 210.5 million plate reads and the 161 acknowledged-but-false stolen alerts are LAPD’s own data covering August 1 to September 30, 2025. Watch for the OIG’s full report publication, the LAPD’s response on the “record-update” defense, and whether other large departments publish comparable audits of their own ALPR programs.
The “Trojan horse” rhetoric inside Microsoft. Nadella’s blog post puts Azure’s enterprise-AI pitch into a frame that openly pits Microsoft against OpenAI and Anthropic. Watch for any change in Microsoft’s commercial terms, any new “proprietary learning environment” Azure product, and any executive response from OpenAI or Anthropic.
The PixVerse “world model” label. A $439M extension at $2B-plus with explicit world-model positioning is the kind of category-defining moment that either sticks or gets absorbed into “video and 3D.” Watch for PixVerse’s next R-Series release, the Alibaba deployment specifics, and any enterprise customer case study.
Hugging Face’s “half the Fortune 500” claim. Delangue repeats this on the Equity podcast; the underlying breakdown by vertical is not public. Watch for the next HF Enterprise customer announcement and any third-party data on adoption by vertical.
The Apple-OpenAI discovery question. The complaint’s 400+ ex-Apple line and the “actual parts” and “dreaded walkout” quotes make settlement less likely. Watch for OpenAI’s answer to the complaint, any procedural motions, and whether the court sets an early schedule that forecloses settlement.
The PixVerse World Labs framing race. Two venture-backed video-gen companies are now positioning around “world model” language. Watch for whether the term holds as a category, becomes marketing noise, or gets redefined by a research lab.