AI Revenue, Safety Gaps, and the Grid Running Out of Room

SpaceX is now pulling in more money from AI compute than from rockets. That single fact captures the strange moment tech is in right now: the AI revenue boom is touching everything, from chip makers to energy grids, and the consequences are only starting to show up.

SpaceX's AI business outpaces its space business

When SpaceX published its first quarterly earnings after going public, the headline figure was a doubling of total revenue year-over-year. But the more striking detail was where the growth came from. The company's AI division, which sells cloud compute to other firms, brought in roughly $2.6 billion, more than the rocket and Starlink sides of the business, according to The Verge's breakdown of SpaceX's earnings. Deals with Anthropic and Google were central to that figure, as TechCrunch reports in its SpaceX earnings coverage.

For anyone learning cloud infrastructure or AI engineering, the takeaway is direct: the market for compute is so hot that a rocket company is winning large contracts against established cloud providers.

Also worth noting: SpaceX spent $329 million on Tesla Megapack battery systems this year alone, showing how the energy demands of running AI infrastructure are funneling money across Elon Musk's portfolio of companies.

AMD's data center revenue more than doubled

The AI compute surge is just as visible at AMD. Its data center business brought in $6.7 billion last quarter, more than double the figure from the same period a year earlier, according to The Verge's AMD earnings report. Gaming revenue, meanwhile, fell. The shift matters for anyone studying hardware or systems: the career gravity in chip design and GPU computing has moved firmly toward enterprise AI workloads, not consumer gaming.

Texas hits pause on new data center connections

All that compute has to run somewhere, and the power grid is feeling it. Texas has suspended new data center connections to its electricity grid after demand from AI infrastructure overwhelmed capacity, as Ars Technica reports on the Texas data center grid halt. The state had been actively courting AI investment, so the pause is a sign of how quickly physical infrastructure limits can catch up with software ambitions. Energy policy and sustainable infrastructure are becoming real engineering constraints, not background noise.

Anthropic's $10B cloud deal and the AI infrastructure spending spree

Anthropic signed a reported $10 billion agreement with AI cloud startup Volta, adding to a string of large partnership announcements in recent months, TechCrunch covers the Anthropic-Volta deal. The scale of these contracts reflects how much compute capacity the leading AI labs need and how quickly new infrastructure businesses are forming around that demand. For anyone interested in startups, the AI cloud niche is producing companies that reach significant scale fast.

Open-weight AI models are closing the capability gap, but not the safety gap

A report from SaferAI found that Z.ai's open-weight model GLM-5.2 is approaching the performance of closed frontier models, with one critical difference: it lacks many of the safety guardrails those closed models include. TechCrunch's report on open-weight AI safety gaps explains that as open models become more capable, the absence of safety mitigations becomes a harder problem to ignore. For developers who use or fine-tune open-weight models, this is worth understanding directly: capability and safety aren't the same thing, and deploying powerful models without considering alignment implications is a real professional responsibility.

Nvidia's AI security alliance shows early momentum

A week after Nvidia helped launch the Open Secure AI Alliance, the group had already grown to over 120 member companies and put out its first proposals for defending against threats from AI agents. TechCrunch's piece on the Open Secure AI Alliance gives a sense of the pace. For cybersecurity learners, AI agents introduce new attack surfaces, and industry-led standards work like this is usually where the vocabulary and defensive frameworks first take shape.

AI coding agents are eating dev budgets

Engineering teams at companies like Replit and Kilo Code report that human developers are reading or writing code themselves only about one percent of the time now, with AI agents handling the rest. That shift is creating real new pressures: which systems are safe to hand to an agent, who cleans up the mistakes, and how to manage token costs that can spiral quickly. VentureBeat's reporting on AI coding agent budget overruns is one of the more grounded accounts of what that transition actually looks like day-to-day. If you're learning to code now, understanding how to evaluate, prompt, and supervise AI agents is becoming as useful as knowing syntax.

Telegram's App Store removal and what it exposed

Telegram was briefly pulled from Apple's App Store after child sexual abuse material appeared in a public chat. Telegram CEO Pavel Durov attributed it to an extortionist who planted the content to trigger a takedown. The Verge's report on the Telegram extortion and App Store removal raises a genuine question about platform moderation: if bad actors can weaponize a platform's own safety enforcement against it, the systems designed to protect users can also be exploited as leverage. Apple's decision to remove the app without contacting Telegram first has also renewed scrutiny of how App Store enforcement actually works in practice, covered further in CNET's look at Apple's CSAM enforcement process.

Hackers drain $130M from hardware crypto wallets

A vulnerability in Coldcard hardware wallets allowed attackers to drain funds even from devices that were offline, with blockchain monitoring firms tracking losses above $130 million. TechCrunch's report on the Coldcard hardware wallet exploit is a sharp reminder that "offline" doesn't automatically mean "secure." For anyone studying cybersecurity or blockchain development, hardware wallet vulnerabilities represent one of the harder categories of attack to defend against because the threat model is often misunderstood by users.

Signal adds multi-phone linking

Signal updated its linked devices feature to support multiple phones on a single account, not just tablets and computers. The Verge's report on Signal's multi-phone linking feature is a small update with a practical implication for privacy-conscious users who carry more than one device. For developers building on privacy-first platforms, Signal's incremental improvements to usability without compromising its encryption model are worth watching as a design reference.

Spotify moves toward paid AI remixes with independent labels

Spotify expanded its upcoming AI remix and covers tool by bringing Merlin, which represents over 30,000 independent labels and distributors, into the project alongside Universal Music Group. The tool will let fans create AI-generated covers and remixes while requiring artists to opt in and ensuring they get credited and paid, according to TechCrunch's coverage of Spotify's AI remix expansion. This is one of the cleaner examples of an AI product being built around consent and compensation from the start, which is an approach that other creative industries are still struggling to figure out.

The pattern across this week's stories is consistent: AI is generating serious revenue, serious infrastructure strain, and serious governance questions at the same time. Anyone entering tech right now is joining an industry that is working all three of those problems simultaneously, and understanding the connections between them is increasingly part of the job.