The Compute Crunch

The AI industry is encountering a problem it didn't expect to face this soon: the physical limits of compute are arriving faster than the business models designed around abundant, cheap inference. When GPU rental prices jump 48% in two months while companies ration their AI offerings, you're watching demand outpace supply at a scale that pricing alone won't fix.

The responses tell the story. Kepler Communications is flying GPUs in orbit, betting that the vacuum of space solves thermal management better than any datacenter. The Biological Computing Company is building chips from living neurons, pursuing energy efficiency through biology rather than silicon. These aren't incremental optimizations. They're admissions that conventional approaches to scaling AI infrastructure aren't keeping pace with adoption.

That adoption curve matters here. Half of employed Americans now use AI at work, crossing a threshold that cements these tools as infrastructure rather than experiment. But infrastructure demands reliability and accessibility. Rationing breaks both. The industry sold AI as boundless capability. Now it's discovering that physics, power grids, and chip fabrication timelines impose very real bounds. The gap between what was promised and what can be delivered is widening. That's the signal worth tracking.

Deep Dive

Orbital compute is not about data centers. It's about processing at the edge of Earth.

Kepler Communications launched 40 GPUs into orbit and signed 18 customers before SpaceX or Blue Origin put a single rack in space. That tells you something important: the near-term business case for space-based compute has nothing to do with giant orbital data centers. It's about edge processing for data that originates in orbit.

The use cases clarify the model. Synthetic aperture radar satellites generate massive amounts of data. Sending it to Earth for processing introduces latency and requires bandwidth. Processing it in orbit, right where the sensor sits, solves both problems. The U.S. military needs this for missile defense systems that detect and track threats in real time. Private satellite operators need it to make sense of sensor data without overwhelming ground stations. Kepler positions itself as networking infrastructure, not a cloud provider. It wants to be the layer that connects sensors in space to processing power in space, handling data before it ever touches Earth.

This matters for founders and VCs because it reframes what "space computing" means in the near term. The companies raising capital for mega-constellations of orbital data centers are betting on a 2030s timeline. Kepler and its customers are building revenue today by solving edge computing problems that exist right now. The hardware challenges are different too. Sophia Space is testing passive cooling systems because space-based GPUs face thermal management problems that terrestrial data centers never encounter. Solving those problems at small scale, with distributed inference workloads, is more tractable than trying to cool a warehouse-sized facility in orbit.

The Wisconsin data center ban that Sophia's CEO cited reveals another angle: regulatory constraints on terrestrial computing could accelerate space-based alternatives faster than anyone expected. That's a tail risk worth monitoring. When physical limits on Earth create openings in orbit, infrastructure assumptions start to shift.

Fifty percent adoption is where AI moves from tool to organizational problem.

Half of U.S. workers now use AI, and the Gallup data reveals something more important than the headline: organizations adopting AI report both more hiring and more layoffs than non-adopters. They're simultaneously expanding (34% report growth) and contracting (23% report cuts). That's not a contradiction. It's restructuring in real time.

The pattern makes sense when you look at what AI actually changes. Workers report productivity gains on specific tasks like drafting content or summarizing information. But only 10% say AI has fundamentally transformed how work gets done across their organization. The gap between individual efficiency and organizational transformation explains the turbulent staffing patterns. Companies are still figuring out which roles become more valuable when augmented by AI and which become redundant. That uncertainty shows up as simultaneous hiring and firing while they sort it out.

The implications split by company size. Large employers with 10,000-plus workers show a different pattern: 33% report workforce reductions versus 30% reporting expansion. Smaller organizations lean toward net growth. That suggests scale matters in how AI reshapes headcount. Larger organizations have more room to eliminate middle layers or consolidate roles. Smaller ones may see AI as a way to do more with existing teams rather than a replacement strategy.

For founders, this is the moment when AI strategy becomes inseparable from org design. You can't bolt AI onto existing workflows and call it done. The 18% of workers who think their jobs will be eliminated by AI in the next five years are concentrated in AI-adopting organizations, jumping to 23%. That's not paranoia. It's people reading the signals their employers are sending through restructuring. Leaders report stronger productivity gains from AI than individual contributors, which suggests the tools favor certain types of work over others. Managing that transition without tanking morale or losing institutional knowledge is now a core competency. The technology is the easy part. The organizational redesign is where most will struggle.

Signal Shots

France Bets on Digital Sovereignty with Linux Migration: France's Interministerial Directorate for Digital Affairs announced plans to replace Windows with Linux across government systems, part of a broader push to reduce dependence on American technology. The move extends beyond operating systems to collaboration tools, databases, and network equipment, with all ministries required to submit transition plans. This matters because it's a template for how other nations might respond to perceived tech concentration risks. Watch whether other European governments follow France's lead and whether this triggers trade tensions with the U.S., particularly given the current administration's sensitivity to policies that disadvantage American tech companies.

Japan's Corporate Coalition Targets Physical AI: SoftBank, Sony, Honda, and six other Japanese companies launched a joint venture to develop a one-trillion-parameter foundation model for physical AI by 2030, according to Nikkei Asia. This matters because it represents a nationalist approach to AI development in robotics and manufacturing, areas where Japan still holds competitive advantages. The consortium structure spreads risk across multiple corporations while pooling resources for compute-intensive model training. Watch whether this model attracts additional Japanese industrial firms and whether it produces commercially viable outputs faster than Western competitors focused on general-purpose models.

AI Tools Reshape Linux Development Process: Linux 7.0 officially added Rust language support while kernel maintainer Linus Torvalds noted that AI bug-finding tools are discovering corner cases at unprecedented rates, potentially establishing a "new normal" for the release process. This matters because it demonstrates AI's practical value in code review at massive scale, with kernel maintainers reporting that AI-generated bug reports have shifted from junk to legitimate overnight. Watch how this changes contribution patterns and whether other large open-source projects adopt similar AI-assisted quality control, potentially lowering barriers to participation while raising code quality standards.

Law Firms Face AI-Generated Work Deluge: Major law firms report lawyers spending significantly more time reviewing AI-generated documents from clients, potentially forcing increases in fixed-fee contract pricing. This matters because it reveals an unintended consequence of widespread AI adoption: when everyone uses AI to generate more output, professionals downstream face higher processing costs. Watch whether this pattern appears in other industries where AI accelerates document creation faster than review capacity scales, and whether it creates backlash against over-reliance on AI drafting tools.

Computer Science Enrollment Drops After 15-Year Run: U.S. colleges are seeing sharp declines in computer science enrollment after a sustained boom, according to new data. This matters because it signals a potential talent pipeline contraction just as AI development demands more specialized technical skills, though the timing suggests students may be responding to recent tech layoffs and uncertainty about AI's impact on programming careers. Watch whether this reverses if AI companies continue aggressive hiring or whether it persists as students conclude that AI tools will commoditize traditional software engineering roles.

Amazon Expands Auto Sales Beyond Hyundai: Amazon quietly added five automakers to Amazon Autos, including Kia, Mazda, Subaru, Chevrolet, and Jeep, expanding beyond its initial Hyundai partnership. The service now operates in over 130 U.S. cities. This matters because it shows traditional manufacturers willing to cede customer relationships to Amazon despite knowing the long-term risks, driven by pressure to match Tesla's direct sales model. Watch whether Amazon leverages this beachhead to insert itself into financing, insurance, and aftermarket services, where margins exceed vehicle sales, and whether dealers push back through state franchise laws.

Scanning the Wire

Elon Musk Launches TikTok Presence Ahead of SpaceX IPO: A verified @elonmusk account posted its first TikTok video promoting SpaceX and Tesla, drawing over 2 million views as the billionaire expands his social media footprint before taking his rocket company public. (New York Times)

Trump Crypto Venture Faces Investor Backlash Over Control Mechanisms: World Liberty Financial is dealing with an investor revolt after billionaire backer Justin Sun accused the project of building a backdoor that could blacklist investors, escalating tensions around the Trump family crypto venture. (Bloomberg)

OpenAI Commits to 500-Person London Office: The AI company plans its first permanent London location with capacity for over 500 staff, reinforcing its February commitment to make the city its largest research hub outside the United States. (CNBC)

Huawei Beats Apple and Samsung to Wide-Format Foldable: The Chinese manufacturer revealed the Pura X Max, a passport-style foldable launching next week in China, getting to market ahead of Western competitors rumored to be working on similar designs. (The Verge)

Netherlands Approves Tesla FSD After 18-Month Review: Dutch regulators became the first in Europe to authorize Tesla's Full Self-Driving software under UN standards following testing across 1.6 million kilometers of European roads. (The Next Web)

Rockstar Games Confirms Data Breach Via Third-Party Provider: The Grand Theft Auto publisher said hackers accessed data through Anodot, a cloud-monitoring service, though the company stated the breach will have no impact on operations despite ShinyHunters demanding ransom. (The Verge)

Basic-Fit Breach Exposes 200,000 Dutch Members: Europe's largest budget fitness chain by club count disclosed a hack affecting members across multiple countries, exposing names, addresses, email addresses, phone numbers, dates of birth, and bank details. (The Next Web)

Outlier

When Your Gym Gets Hacked, Check Your Threat Model: Basic-Fit, Europe's largest budget fitness chain, disclosed a breach exposing 200,000 Dutch members' bank details, addresses, and personal data. The weird part isn't that it happened. It's that your local gym now holds the same data profile as your bank, processes payments like a fintech startup, and sits in the crosshairs of the same threat actors targeting tech companies. Fitness chains weren't supposed to become critical data infrastructure, but subscription models and app-based check-ins transformed them into payment processors with weak security postures. This signals a broader pattern: every industry digitizing its customer relationships inherits cybersecurity problems it lacks the expertise to solve, creating an expanding attack surface that security teams never anticipated defending.

The physicists were right. Everything scales until it doesn't. We just didn't expect to hit the ceiling while half the workforce was still figuring out the prompts.