the Now & the Next

A bi-weekly speculative fiction suggesting the shape of things to come.
(sourced from trustworthy trade pubs, think tanks + frontier science news)

1,436 Signals Tracked
7 Collisions Identified
10 Frontier Science Cards
234 Source Domains

This fortnight brought 1,436 signals across corporate, startup, thinktank, and research sources into focus, and the throughline is the AI story is leaving the screen and colliding with the physical, human world. A Gallup poll found 71% of Americans don't want a data center near them — more opposition than nuclear plants draw — while the electricity demand behind those centers now exceeds any surge since the 1950s air-conditioning boom. Autonomous excavators and dozers started working real jobsites, a Swiss retrofit startup raised $200M from SoftBank to make any machine driverless, and Waymo shipped custom silicon built for the road. The store quietly became a media company: Walmart's ad business grew 38%, Target's profit doubled on Roundel, and both named Chief AI Officers. AI learned to read the body — skin analysis in 4,000 beauty stores, GLP-1 nutrition, wedding-ecosystem matchmaking — while hospitals put inference into the exam room at Epic, Mayo, and Ping An. Underneath the optimism, two anxieties surfaced: enterprises admitting they can't stop a runaway agent's spending, and China quietly catching up across chips, robot dogs, cars, and music. Fifteen fortnights in, the pattern isn't AI in the cloud anymore. It's AI in the neighborhood, on the jobsite, in the mirror, and in the exam room — and the friction that comes when software finally touches the ground.

01
The Neighbors Wake Up

The AI buildout stopped being an abstraction and became the hum next door — and the public, the grid, and the insurers all pushed back at once.

+

The data center went from press release to neighborhood grievance. A Gallup poll found 71% of Americans oppose an AI data center being built near them — with nearly half strongly opposed, a resistance that crosses party lines and exceeds opposition to local nuclear plants — much of it traced to the constant low-frequency hum from cooling systems that neighbors blame for sleepless nights and headaches. The economics are getting scrutinized too: economist Severin Borenstein warned that AI-driven interconnection requests may be inflated, even as he acknowledged the underlying demand surge is larger than any since the 1950s air-conditioning rollout. The physical layer is straining — Metro Colo Advisory found that while most North American operators have space, almost none can deliver the 140–160 kW-per-rack liquid cooling that high-density AI now needs — so operators are improvising: Nexus Data Centers raised a $1.3B mezzanine loan to build a behind-the-meter power campus in Texas, and Voltus and Sunrun began aggregating home batteries and thermostats to lend firm capacity back to the grid. And the risk is repricing: Allianz Commercial projects data-center investment will top $1 trillion a year by 2027 while the insurance market for fire, climate, and cyber exposure more than doubles to $24B by 2030.

⚡ The Now

The AI infrastructure boom has run into the one constraint that can't be provisioned in the cloud: the people, power grids, and insurers of the physical world. Communities are organizing against the hum, utilities are questioning whether the demand is real, and the money is flowing into behind-the-meter power and distributed batteries to route around a grid that can't keep up.

→ What's Next

Expect siting and community consent to become as decisive as chip supply. The winners will be operators who bring their own power (behind-the-meter gas, nuclear, or aggregated home batteries), who can prove their demand forecasts to skeptical regulators, and who treat noise, water, and insurability as design constraints rather than externalities. 'Social license to build' becomes a line item next to megawatts.

Space Daily
A 2026 Gallup poll finds 71% of Americans oppose AI data centers near their homes — crossing political lines and exceeding opposition to nuclear plants — citing noise, resource use, and quality of life.
Broadband Breakfast
Severin Borenstein notes AI-driven electricity demand exceeds any surge since the 1950s air-conditioning rollout, while cautioning that interconnection requests may overstate real demand.
Data Center Knowledge
A search for a 5 MW buyer needing 140–160 kW per rack with direct liquid cooling found most operators have space but few can support the highest-density AI workloads.
PitchBook
Nexus secured a $1.3B mezzanine loan to build a multi-phase AI data center with behind-the-meter power generation in Texas to feed energy-intensive workloads.
Data Center Knowledge
Voltus's 'Bring Your Own Capacity' program aggregates home batteries and smart thermostats so large consumers like AI data centers can lend firm, flexible capacity back to the grid.
Insurance Asia
Allianz projects data-center investment topping $1T/year by 2027, with the US and China at 62% of new capacity and the insurance market doubling to $24B by 2030 amid fire and cyber risk.
02
Nobody's Holding the Leash

Enterprises are racing to deploy autonomous agents they admit they can't yet control — and the governance layer is being built after the agents are already spending money.

+

The gap between AI ambition and AI control opened wide this fortnight. A VentureBeat survey found that one in five enterprises can't stop a runaway AI agent's spending in real time, even as they push toward multi-step autonomous systems. Deloitte put a number on the readiness gap: only 20% of U.S. organizations are prepared to redesign their processes for autonomous AI, while 61% of senior leaders expect widespread AI autonomy within four years — a mismatch rooted in fragmented data, undocumented processes, and entrenched work habits. The vendors are rushing to fill the void: Sierra AI shipped 'release governance' with automated agent checks, peer-review merge approvals, and split-traffic rollouts with fast rollback; Google launched its Gemini Enterprise Agent Platform with agent identity, centralized governance, and policy enforcement; and startups like xpander are pitching enterprises on owning their own control and context layer as agents sprawl. Meanwhile the human layer is exposed too: Dayforce found 77% of front-line managers feel unsupported by systems that can't give them coherent guidance, and Computerworld warned that 'vibe coding' — non-technical staff shipping generative-AI-built apps — is quietly introducing security and quality risk at enterprise scale.

⚡ The Now

Autonomy is being deployed faster than the guardrails to contain it. Companies are handing agents budgets, tools, and API access while admitting they lack the observability, rollback, and governance to stop them when they misbehave — and the same democratization is putting production code in the hands of employees with no software-engineering training.

→ What's Next

Expect 'agent governance' to become a distinct procurement category the way identity and access management did — spend controls, kill switches, audit trails, and peer-review gates baked into every agent deployment. The organizations that win won't be the ones that deploy the most agents; they'll be the ones that can prove, in real time, that they can stop one.

VentureBeat
A survey finds most enterprises are early in deploying multi-step agents, with control planes still immature — leaving many unable to halt a runaway agent's spending in real time.
HR Dive
Deloitte finds just 20% of U.S. organizations are ready to redesign processes for autonomous AI even as 61% of leaders expect widespread autonomy within four years.
Sierra AI
Sierra adds automated agent checks and simulations, merge-approval workflows with peer review, and split-traffic releases with fast rollback to make agent deployment safer.
VentureBeat
Google's new Gemini Enterprise Agent Platform adds agent identity, centralized governance, memory, tracing, and policy enforcement as enterprises confront agent sprawl.
HR Dive
A Dayforce report finds 77% of front-line managers feel unsupported by fragmented systems that slow decisions and raise risk under do-more-with-less pressure.
Computerworld
Enterprises training non-technical staff to build apps with generative AI accelerate innovation but introduce elevated security and quality risk from a lack of developer expertise.
03
The Gap Closes From Behind

China stopped chasing the frontier and started matching it — in chips, robot dogs, cars, and now generative music and film — reframing the whole 'AI race' narrative.

+

The assumption of a durable Western AI lead took a hit this fortnight. The Decoder's Frontier Radar concluded China has effectively caught up, leaving open the question of what edge remains — even as OpenAI paused parts of its Astra cyber model over risk concerns. The hardware embargo is leaking by design: China is letting Nvidia's H200 chips trickle onto the mainland, with ByteDance and Tencent each receiving roughly 10,000 units under NDRC approval, enough to keep domestic AI firms pacing the U.S. In robotics, C4ISRNET reported that China used Pentagon-funded U.S. designs to build robot quadrupeds and now leads in fielding them. The script has flipped in autos too: Chinese firms are offering to help Western automakers build cars faster on home soil, while LG repurposed North American EV-battery lines to make energy-storage cells for AI data centers. And China is now exporting AI culture, not just AI compute: Alibaba launched HappyShrimp 1.0, a text-to-song generator with Taihe Music signed up for artist co-creation, while the Motion Picture Association signed a first-of-its-kind deal with ByteDance to protect actors' likenesses in generative video — an implicit acknowledgment that Chinese platforms now sit at the center of global creative-AI governance.

⚡ The Now

The 'AI race' framing of a leading West and a following East is dissolving. China has parity in models, is routing around chip controls, leads in fielded robotics, and is now shipping consumer generative-AI products in music and video — forcing Hollywood and the chip industry to negotiate rather than dictate.

→ What's Next

Expect the competition to shift from 'who has the best model' to 'who controls the deployment surface' — factories, robots, entertainment platforms, and the standards around them. Western firms will increasingly find themselves partnering with, licensing to, or setting guardrails alongside Chinese platforms rather than simply out-innovating them, and export controls will look more like speed bumps than walls.

The Decoder
OpenAI advances GPT-5.6 Sol and cyber models like Daybreak and Astra but paused parts of Astra over risk — as the broader analysis concludes China has closed the frontier gap.
The Decoder
China permits limited H200 imports — ByteDance and Tencent each getting ~10,000 units under NDRC approval — to sustain domestic AI despite U.S. export restrictions.
C4ISRNET
China leveraged Pentagon-funded U.S. designs to develop advanced robot quadrupeds, now surpassing U.S. fielding and challenging its lead in unmanned battlefield tech.
The Drive
Chinese firms offer to help Western automakers build faster at home, while LG repurposes North American EV-battery lines to make energy-storage cells for AI data centers.
Music Business Worldwide
Alibaba's HappyShrimp 1.0 creates complete songs from a text prompt with customizable lyrics, instrumentation, and vocals, with Taihe Music signed for artist co-creation.
World IP Review
The Motion Picture Association strikes a first-of-its-kind deal with ByteDance to protect actors' likenesses in generative AI video, setting cross-border IP precedent.
04
The Store Becomes the Media Company

Retail's fastest-growing profit line isn't merchandise — it's ads, AI shopping agents, and cultural collaborations that turn the storefront into an audience business.

+

The most profitable thing a retailer sells now may be attention. Walmart raised its full-year guidance on the back of 38% ad growth, with its Sparky AI shopping agent up 70% year-over-year in users who spend 40% more per order. Target's profits doubled as its Roundel retail-media arm and same-day delivery powered a growth streak — and Target named its first Chief AI Officer, partnering with OpenAI and Google Gemini to grow AI-platform-driven traffic 3.5x faster than the industry. The ad machinery underneath is standardizing: Nexxen launched a universal spec for programmatic home-screen ads across TV OEMs, cutting onboarding time in half with AI-assisted creative resizing. And retailers are borrowing the playbook of media companies — Spotify and Depop staged an artist-curated fashion-and-music collaboration with live LA events, blending commerce and culture into fan experience, while Maisonette launched 'Neon Rebels,' a tween vertical with its own curated browsing experience, explicitly foregoing social ads targeted at kids.

⚡ The Now

The retail P&L is inverting: high-margin advertising, AI shopping agents, and cultural IP are becoming the growth engine while merchandise margins stay thin. Walmart and Target are effectively running media and audience businesses inside their stores, and the AI agent is both the salesperson and the ad unit.

→ What's Next

Expect every large retailer to build (or buy) a retail-media network and an owned AI shopping agent, and to compete on cultural relevance the way media companies do. The strategic question shifts from 'how do we move product' to 'how do we own the audience and the moment of recommendation' — and the retailers without clean product data or a compelling brand story get disintermediated by whoever's agent the shopper trusts.

Ad Week
Walmart raised guidance on 38% ad growth; its Sparky AI shopping agent grew users 70% YoY, and those users spend 40% more per order than non-users.
Ad Week
Target's profits doubled on Roundel retail media and same-day delivery; its first Chief AI Officer and OpenAI/Gemini partnerships grew AI-platform traffic 3.5x the industry average.
Ad Exchanger
Nexxen launched a universal spec for programmatic native home-screen ads across TV OEMs, using AI-assisted resizing to cut new-partner onboarding time up to 50%.
Spotify
Spotify and Depop paired six artists to curate resale shops and playlists with live LA events — blending music, fashion, and e-commerce into immersive fan experiences.
Glossy
Maisonette launched 'Neon Rebels' for tweens 7–14 with 100+ brands and a dedicated browsing experience, deliberately avoiding social ads targeted at children.
05
The Jobsite Goes Driverless

Autonomous machines moved from the highway to the dirt — excavators, dozers, farm robots, and logistics AI all crossed from pilot to payroll this fortnight.

+

The self-driving story quietly relocated from passenger cars to heavy industry. Bedrock Robotics began deploying fully autonomous excavators on real construction jobsites and laid out plans to extend the same stack to dump trucks, dozers, and loaders working collaboratively. The capital agrees: Gravis Robotics, an ETH Zurich spinout, raised a $200M Series A led by SoftBank for a bolt-on 'Gravis Rack' that retrofits existing machinery into autonomous robots — automating the equipment already on site rather than replacing it. In agriculture, UC ANR Innovate's field-day program is becoming the new due-diligence standard, putting physical-AI startups in front of growers to prove they work on hard crops like leafy greens before the term sheet. The intelligence keeps advancing on the road too — Waymo unveiled custom compute delivering a 20x power increase over eight years with dual independent engines for driver-free safety — and the paperwork behind the freight is automating in parallel: Descartes launched logistics-trained AI agents that turn trade documents into operational data, stripping manual entry out of customs and shipment workflows.

⚡ The Now

Physical AI is crossing the pilot-to-production line in the least glamorous, highest-labor-shortage corners of the economy: construction, mining, agriculture, and freight. The pattern is retrofit and augment — bolt autonomy onto existing machines and automate the documents around them — not rip-and-replace.

→ What's Next

Expect the value to concentrate in the autonomy layer that works across equipment brands and the field-validation networks that prove it on real terrain. Labor-short industries will adopt fastest because the alternative is idle machines; the winners will be the retrofit platforms that turn a customer's existing fleet driverless and the logistics AI that removes the human bottleneck between the physical work and the paperwork.

Construction Dive
Bedrock Robotics is deploying autonomous excavators on jobsites and plans to extend the tech to dump trucks, dozers, and loaders working collaboratively to extend operating hours.
Construction Dive
ETH Zurich spinout Gravis Robotics raised a $200M Series A led by SoftBank for a bolt-on 'Gravis Rack' controller that retrofits heavy machinery into autonomous robots.
AgFunderNews
UC ANR Innovate's field days put physical-AI agtech startups in front of growers and researchers to validate real-world use on hard crops before investment.
Waymo
Waymo unveiled custom autonomous-driving compute with deterministic ultra-low latency, dual independent engines, and a 20x power increase over eight years for driver-free safety.
Container News
Descartes added logistics-trained AI agents that convert trade documents into operational data, cutting manual entry from customs entry and shipment-creation workflows.
06
The Mirror Learns to Read You

AI is turning the body and personal context into data brands can act on — skin, weddings, diets, and travel are all becoming personalized, machine-read experiences.

+

Consumer brands started reading the customer's body and life-context directly. Grupo Boticário scaled Haut.AI's skin-analysis technology from a 24-store pilot to nearly 4,000 stores nationwide, deepening a three-year co-creation deal and making machine-read skin diagnostics a routine part of the beauty-counter experience. David's Bridal is building an AI tool that connects couples to the wider wedding ecosystem — photographers, florists, nail salons, event planners — turning a dress retailer into an AI matchmaker for an entire life event and harvesting the data that comes with it. Nestlé built proprietary AI platforms including 'Food Genie' to design nutrition products for GLP-1 drug users, simulating consumer behavior and predicting recipe performance for a body chemistry that's suddenly changing at scale. And Delta Air Lines rolled its AI-powered Concierge out to all SkyMiles members, blending self-service with human support to handle disruptions, rebooking, and refunds in real time with full account context. Across beauty, retail, food, and travel, the interface is becoming a model that knows something specific about you — your skin, your event, your metabolism, your itinerary.

⚡ The Now

Personalization is moving from 'people who bought this also bought' to machine-read models of the individual body and moment — your skin's condition, your wedding's full vendor graph, your GLP-1-altered appetite, your disrupted flight. Brands that can capture and act on that first-party context are building relationships competitors can't easily copy.

→ What's Next

Expect the frontier of consumer AI to be intimate context: sensors and models that read the person, not the purchase history. The trust stakes rise with the intimacy — brands that handle body and life data transparently earn loyalty, while those that feel extractive trigger backlash. Winners turn a one-time transaction (a dress, a foundation, a flight) into an ongoing, data-informed relationship around a life stage or a body.

Cosmetics Business
Grupo Boticário scaled Haut.AI's skin-analysis tech from a 24-store pilot to ~4,000 stores under a three-year co-creation deal, making AI diagnostics a routine beauty-retail experience.
Retail Brew
David's Bridal is building an AI tool connecting couples to the wider wedding ecosystem — photographers, florists, planners — expanding beyond dresses and deepening customer data.
The Dairy Site
Nestlé's proprietary AI platforms, including Food Genie, simulate consumer behavior and predict recipe performance to design nutrition products for GLP-1 drug users.
Future Travel Experience
Delta launched its AI Concierge for all SkyMiles members, combining self-service and human support for disruptions, rebooking, baggage, and real-time refunds with account context.
07
The Clinic Runs on Inference

AI stopped being a research demo in healthcare and became operational infrastructure — in the exam room, the EHR, the eye clinic, and the home-care visit.

+

Healthcare quietly moved AI from the lab into the workflow. Epic launched Ergo Visit, an AI tool that analyzes a patient's record before an outpatient visit and drafts notes and issue lists afterward — already live at Ochsner Health — putting inference directly into the clinician's day. Mayo Clinic named a new CIO as it doubled down on a Microsoft partnership to co-develop a frontier medical model for earlier diagnosis and treatment planning. The economics are proving out abroad: Ping An Good Doctor lifted profit 63.5%, with its AI Doctor now reaching 100% of the group's retail customers and 9.7 million users, and AI contributing materially to gross profit. Regulators are clearing the path — Zeiss won FDA 510(k) clearance for Clinic 360, a browser-based platform unifying diagnostic data across devices and EHRs. And the labor-strained edges are adopting fastest: four large at-home care providers including The Pennant Group are positioning AI tools as a staffing advantage to stretch scarce clinicians, while Radial acquired Mindful Health Solutions to fuse its AI-native RadialOS platform with multispecialty mental-health care.

⚡ The Now

AI in medicine has crossed from pilot to plumbing: it's in the EHR (Epic), the hospital strategy (Mayo/Microsoft), the P&L (Ping An), the regulatory pipeline (Zeiss FDA clearance), and the staffing model for home and behavioral health. The value proposition has shifted from diagnostic accuracy demos to operational leverage on a workforce that's stretched thin.

→ What's Next

Expect AI to become a standard line in health-system operating budgets, sold on clinician time saved and patient throughput rather than clinical novelty. The fault lines will be trust, liability, and reimbursement — but the labor shortage is a forcing function, and the systems that integrate AI into documentation, triage, and home care will out-capacity the ones that treat it as an experiment.

Healthcare Dive
Epic's Ergo Visit analyzes patient records before outpatient visits and drafts notes and issue lists afterward, live at Ochsner Health, to cut administrative burden.
Healthcare Dive
Mayo Clinic partnered with Microsoft to co-develop a frontier medical AI model for earlier diagnosis and treatment planning, naming a new CIO to drive the strategy.
Insurance Asia
Ping An Good Doctor's AI Doctor reached 9.7M users and 100% of the group's retail customers, contributing materially to a 63.5% profit lift.
Ophthalmology Management
Zeiss won FDA 510(k) clearance for Clinic 360, a cloud/browser platform unifying diagnostic and medical data across third-party devices and EHRs for disease-progression evaluation.
Home Health Care News
The Pennant Group and peers are piloting AI to improve clinician efficiency and job experience, positioning technology as a staffing advantage amid workforce shortages.
Behavioral Health Business
Radial acquired Mindful Health Solutions to fuse its AI-native RadialOS platform with multispecialty mental-health care, positioning for growth in tech-driven brain medicine.

Frontier Science Seeding the System

The research signals underneath the enterprise stories — ten breakthroughs from labs and universities this fortnight that didn't fit a collision but are too interesting to skip.

Neurotech
Ability Neurotech built a battery-free brain-computer interface that records from 128 channels at 30,000 samples/second and beams neural data through the skin via an infrared laser at up to 50 Mbps — now in clinical trials for speech and movement decoding in paralyzed patients.
AI for Science
Anthropic showed Claude can design protein binders from scratch, succeeding on 14 of 15 protein targets with 22–35% hit rates — well above the typical 10–15% — and some binders outperformed the best previously published results, with minimal human input.
Digital Health
Google Research's PhotoScan estimates 3D body composition and cardiometabolic risk from ordinary 2D phone photos, beating smartwatch impedance sensors and approaching DXA-scan accuracy for predicting insulin resistance — no specialized equipment required.
Synthetic Biology
Oxford researchers used AI to design bacteriophages ~97% genetically similar to a natural phage, mostly rearranging existing DNA but producing novel 3D protein structures and infection behaviors — showing AI can explore genetic combinations beyond natural evolution, and raising fresh biosafety questions.
Collective AI
University of Konstanz researchers found LLM-powered agents can reach group consensus in populations of up to ~1,000 — far past the human Dunbar number of 150–200 — mirroring majority-following behavior at a scale no human society sustains, with implications for large-scale AI coordination.
Telemedicine AI
Google's AMIE evolved from text diagnosis to real-time video consultations, perceiving non-verbal cues and guiding virtual physical exams. In a randomized study of 300 video consultations across 100 scenarios, it was benchmarked against board-certified primary-care physicians.
Industrial AI
Empa, SUPSI, and Bystronic combined acoustic and optical sensors with machine learning so a laser cutter can judge its own cut quality and adjust parameters in real time — retrofittable to existing machines with microphones, cameras, and a control PC, eliminating manual test runs.
Autonomous Logistics
HKUST's Aerial Robotics Group demonstrated a drone that can deliver packages to vehicles in motion — and potentially through open windows — using precise real-time navigation to interact with dynamic targets, a step toward contactless last-mile delivery to recipients in transit.
Human-Machine Interface
Augmental's Vox, a wearable necklace microphone paired with its Mouthpad, enables low-volume speech as a hands-free input method — with a long-term goal of a fully silent speech interface that tracks tongue movements for private, noise-proof device control.
Computational Chemistry
Microsoft Research's Skala 1.1, trained on 2.5x more data, predicts molecular properties like thermochemistry and reaction kinetics with substantially higher accuracy while staying computationally cheap — accelerating chemistry and materials-science simulation.