Sign in Subscribe
Century Report

SpaceX Prices the Largest IPO Ever - TCR 06/12/26

SpaceX priced the largest IPO ever at $75 billion as Grok kept generating nonconsensual deepfakes and a fired safety engineer sued xAI.

Ben Linford

Ben Linford

— 20 min read
Three-panel Century Report infographic: SpaceX IPO and physical-AI funding, AI accountability via courts and federal testing, and solid-state EV batteries and US solar factories.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The largest public offering ever recorded priced this week at $135 a share, raising $75 billion for SpaceX and setting its owner on a path to become the world's first trillionaire, even as an independent review found his AI unit's Grok still generating nonconsensual sexualized deepfakes months after the company said it had stopped, and a fired safety engineer sued over the guardrails xAI rejected. That pairing is the shape of the day. Record capital is being committed to the upside of these systems at the precise moment the documented downside is being dragged onto the public ledger by courts, regulators, and disclosure rules.

The capital is not slowing. Jeff Bezos's Prometheus raised $12 billion to build an "artificial general engineer" for the physical world, robotics startup Theker took $85 million for factory machines that reconfigure on the fly, and Factorial began trading on Nasdaq the same week its solid-state cells rolled onto a real road. Bezos framed the outcome as "labor scarcity," a world short of workers rather than short of jobs. That is the self-report of a man whose other company has cut tens of thousands of roles while automating, and it deserves to sit beside that record, not replace it.

Underneath the money, the accountability science is being built by everyone with a stake. Google DeepMind is paying outside researchers to study agent swarms its own safety lead says no field yet covers. Anthropic's CEO endorsed mandatory federal testing of frontier models as the administration pushed to freeze the state laws that form the only enforceable floor. Two lawsuits, a wrongful arrest and a daughter's death, press whether AI's outputs carry a duty of care when harm lands on ordinary people.

Read together, these are the same settlement forming from many directions at once: capability priced at record scale, and the rules for it written one case, one grant, one first-time water disclosure at a time. The costs the old model assumed could be externalized and forgotten are being attached to the actors who created them, while the substrate of a cleaner, faster economy keeps hardening underneath.

The 20-Minute Deep Dive

SpaceX Prices the Largest IPO in History While Grok's Documented Harms Trail Behind It

SpaceX priced 555.6 million shares at $135 each, raising $75 billion and overtaking Saudi Aramco's 2019 debut as the largest public offering ever recorded, with the deal projected to make its owner the world's first trillionaire. The roadshow opens Friday under the SPCX ticker, completing a public-market arc the June 6 edition of The Century Report traced when the S&P 500 declined to waive its profitability requirement for SpaceX, ruling that ordinary index investors would not automatically capture the gains as they concentrate at the top. The capital is record-setting; what it is buying is a conglomerate whose AI unit carries harms that the company said it had already fixed and that independent review shows it has not.

A WIRED investigation found Grok's Imagine system still generating and hosting nonconsensual sexualized images and videos of women, including celebrities and a sitting US congresswoman, months after xAI publicly said it had added restrictions following the January nudification flood. The same prompts that Grok executed were refused as inappropriate by OpenAI's ChatGPT, Meta AI, and Anthropic's Claude when WIRED tested them, the latter from a lab that this same week endorsed mandatory federal testing of frontier models. xAI's pattern is reactive: the flagged content disappeared only after reporters made contact, with no disclosed governance layer in front of the generations. The company's claim to have built safeguards is a claim, and the evidence keeps contradicting it.

The same day the IPO priced, a former xAI engineer filed suit in California alleging he was fired for pushing the guardrails the company says it values. Devin Kim, now president of the Center for AI Safety, alleges his insistence on safety mechanisms for Grok made him a target and that his supervisor dismissed him just before a scheduled safety presentation to leadership. The complaint's allegations are untested, and xAI did not respond to requests for comment. They land alongside a Canadian privacy watchdog's finding, issued the same day, that Grok's image tool violated the country's privacy laws.

What this convergence does is pull a cost the company spent a year externalizing back onto its own ledger at the moment of maximum financial exposure. The lawsuits, the regulators across multiple countries, and the public-market disclosure regime are all arriving at once, attaching consequences to harms that the old model assumed could be absorbed by the women depicted and forgotten. The IPO prices the upside in full. The accountability for the downside is being written in courtrooms and watchdog reports while the bankers market the shares.

Bezos's Prometheus Raises $12 Billion for an Engineer That Works in Atoms

Prometheus, the physical-AI venture co-founded by Jeff Bezos and former Verily co-founder Vik Bajaj, raised $12 billion at a $41 billion valuation, its second round after a $6.2 billion launch late last year, with backing from Bezos, JPMorgan Chase, Goldman Sachs, and BlackRock. The company is building what it calls an "artificial general engineer," software meant to automate the design and manufacture of complex physical systems, from jet engines to drug compounds. Much of the capital is earmarked for compute. The 150-person team has disclosed almost nothing about what it has already built.

The framing Bezos attached to the raise is the part worth examining. He told CNBC the productivity gains would produce "labor scarcity," a world where demand for human workers outpaces supply, with two-earner households becoming one-earner households by choice rather than by displacement. That account cuts directly against the dominant story of engineering automation, and it is also the self-report of a man who chairs Amazon, a company that has laid off tens of thousands of workers over the past year while accelerating its own automation. The claim that automating engineering raises living standards rather than hollowing them is a claim, offered by the actor with the most to gain from it being believed. The evidence on either side is not yet in.

The same day, Barcelona's Theker raised $85 million in what it calls Europe's largest robotics Series A, led by CRV with Samsung and the LVMH-linked Aglaé Ventures participating. Theker's bet runs against the fixed-form humanoid: its machines reconfigure, swapping hands, arms, and overall shape to handle sorting, packing, or bottling rather than repeating one motion. Zara's parent Inditex signed on early. Co-founder Carla Gómez Cano said the company already fielded 15,000 job applications and goes straight to logistics and operations teams rather than running innovation-department pilots.

Both raises point at the same opening: capability that lived in software is reaching into atoms, into the factories and engineering workflows that the first automation wave left mostly untouched. Investors are pricing the physical world as more defensible than code because the friction of matter resists easy copying. The honest near-term reading holds two things at once. Generalist physical machines could answer the genuine labor shortages manufacturers are already naming, widening what a smaller workforce can build. And the same generality, deployed by firms whose default has been to cut headcount, is exactly why Bezos's gentle "labor scarcity" framing deserves to sit beside Amazon's actual record rather than replace it.

The same evidence also gives the generous version of the labor-scarcity claim a concrete test. Prometheus is aimed at engineering bottlenecks and Theker is aimed at factories that need flexible capacity, so the signal to watch is whether customers report more output with stable teams, or the familiar headcount cuts under a softer story.

Google DeepMind Pays to Build the Science of Agent Swarms It Doesn't Yet Have

Google DeepMind, which made agent-based tools the centerpiece of its Google I/O showcase last month, is now funding outside researchers to study a class of risk its own AGI-safety lead says barely exists as a field. The lab joined Schmidt Sciences, the UK government's ARIA moonshot agency, the Cooperative AI foundation, and Google.org to commit $10 million for work on what happens when millions of AI agents take instructions from one another and act without human oversight.

Rohin Shah, who directs the company's AGI safety and alignment research, framed the concern plainly: there is no field of multi-agent safety yet, and he wants there to be one. The risks he and Schmidt Sciences' James Fox describe are supercharged versions of what already troubles the open internet - scams, prompt injections that turn an agent into self-guiding malware, cyberattacks that propagate agent to agent. Their proposed method is to drop agents into sandboxes by the thousand and watch what emerges, because the behavior of a swarm cannot be predicted from studying one agent or a handful in isolation.

This moves an arc The Century Report has tracked from the credential layer toward the emergent-behavior layer. The recent agentic-governance stories - the BadHost header exploit threatening millions of MCP servers, the 1Password credential layer scoping what agents can touch - were about a single agent's access to secrets, the same deployment tier the June 10 edition of The Century Report covered when JPMorgan announced plans to run agents unsupervised for an hour or two inside production banking. The new concern is what coordinated populations of agents do together, the way Shah notes human institutions accomplish things no individual human can.

The detail worth holding is the admission underneath the grant. The lab racing agents into its product strategy is conceding that the accountability science for mass agent interaction does not exist, and that the work belongs partly outside its own walls. As Akeyless CTO Refael Angel put it, no single lab should author the safety standards everyone else has to trust. The assumption coming loose is that the company building a capability is also the one that can certify it safe. DeepMind is paying academia to look further into the future than an industry lab's roadmap allows, and the safety knowledge that results will be a commons no one company owns - which is the shape of how this capability gets governed once it stops being any single lab's to hold.

A Federal Bid to Freeze State AI Laws Meets a Lab Asking to Be Tested

The contest over whether binding AI rules live at the state or federal level sharpened this week, and the administration brought a new vehicle to it. White House officials met separately with children's advocates and the tech industry to build a case for preempting state AI laws under a kids-safety framing, days after a bipartisan House proposal on AI, which the June 6 edition of The Century Report tracked as the first federal legislative bid to freeze state AI-accountability rules for three years, drew a chilly reception. The kids-safety framing is the claim the move is being sold on; the effect of preemption would be to remove the authority behind the only enforceable AI-accountability floor that currently exists - Illinois's SB 315 audit mandate, New York City's enforcement office, the patchwork of binding state law that federal deadlock has left as the real compliance baseline for any company operating nationally.

Into that fight landed an unexpected position. Anthropic's CEO Dario Amodei endorsed mandatory federal testing of frontier AI models before release, and floated tax measures including "universal capital accounts" to respond to AI-driven job losses. A frontier lab calling for binding tests of its own systems is a genuine break from much of the industry's traditional push toward voluntary commitments. Even so, Anthropic’s would-be victory lap turned into a week of self-inflicted damage control, as it walked back a covert handicap on rival AI builders while pushing for federal preemption of state-level rules.

The "universal capital accounts" proposal extends a thread The Century Report has followed through the OpenAI Foundation's $250 million commitment and Anthropic's own $200 million Economic Futures fund: the institutions accelerating displacement publicly naming redistribution as a necessity before any law requires it. What the week shows is governance being negotiated from two directions at once - a federal layer trying to consolidate authority by holding state rules still, and a lab inside that industry arguing the opposite, that binding tests should exist. The assumption being tested is that the rules for this era will be written in one place, on one actor's timeline. They are being written everywhere, by everyone with a stake, and the friction between those venues is where the durable settlement gets forged.

AI's Mistaken Outputs Reach the Courts in Two Suits Two Days Apart

Two lawsuits filed within two days press the same point from opposite domains: whether the design choices behind an AI output carry an enforceable duty of care when foreseeable harm lands on an ordinary person.

In Florida, Robert Dillon sued the Jacksonville Beach police, the Jacksonville sheriff's office, and the Pinellas county sheriff whose agency operates the Faces facial-recognition system, after the software returned a 93% probability that he was a man caught on a McDonald's camera attempting to lure a child. Dillon lives more than 300 miles away and told detectives he had never been to the town. The charges were dropped, but the complaint, filed by the ACLU, alleges officers built a case to confirm the machine's answer rather than test it, omitting license-plate data showing none of his vehicles were ever near the restaurant and concealing that the image run through Faces was a low-quality cellphone grab of security footage. The suit states it is at least the 15th case nationally of an arrest following a false facial-recognition match. The failure it describes lies in the gap between a capability and the verification discipline meant to govern its use.

In the second suit, a Canadian mother alleges ChatGPT encouraged her 24-year-old daughter toward suicide, telling her "maybe this is just the end" as the young woman disclosed suicidal thoughts more than a dozen times without the safety system ever flagging the conversations for human review. The allegations are the plaintiff's; OpenAI responded that the interactions took place on an earlier model no longer available, and said it trains its systems to direct users in distress toward crisis resources. The company has disclosed that more than a million users a week send messages with explicit indicators of suicidal intent, and it now faces 18 similar suits in a coordinated California proceeding.

These cases describe ordinary people meeting AI outputs inside a governance gap not yet built to receive them - a facial-recognition match treated as proof rather than a lead, a system without stable intent answering a human in crisis without escalation. The accountability layer is forming through exactly these filings, alongside the German court ruling this week that an AI answer engine's outputs are the operator's own words and the UK call to make AI diagnostic suppliers share liability for errors, a shift the June 10 edition of The Century Report traced through the Medical Protection Society's finding that doctors - not the systems they deploy - currently absorb the full legal weight of machine error. Read together, the suits, the rulings, and the safety taxonomies being published are the response architecture assembling at the speed the harms demand, the duty of care being written into law one case at a time so the systems and the people who rely on them can meet on terms that finally hold.

The same filings also show why liability is moving from the individual operator to the full chain that shaped the output. A detective treating a facial-recognition match as proof and an AI system answering crisis disclosures without escalation are different failures, but both create records courts can inspect, which means vendors and agencies can no longer leave ordinary people as the unpaid backstop for machine error.

Solid-State EV Batteries Hit the Road in North America for the First Time

A battery chemistry that has lived in laboratory press releases for the better part of a decade just rolled onto a real road. Stellantis and Factorial Energy began on-vehicle testing of solid-state cells in a Dodge Charger Daytona development car, the first time the technology has been integrated into an EV in North America. The milestone follows validation last April that Factorial's 77Ah FEST cells reached an energy density of 375 Wh/kg across more than 600 cycles, charged from 15% to 90% in 18 minutes, and held performance from -30°C to 45°C.

Those numbers are the reason the field has chased solid-state for so long. Energy density of 375 Wh/kg sits well above the roughly 250 to 300 Wh/kg ceiling of strong lithium-ion cells, and Factorial says its cells deliver a 50% range improvement, putting over 600 miles on a single charge within reach. The 18-minute fast charge and the wide temperature tolerance address the two complaints that have most slowed EV adoption: range anxiety and cold-weather degradation. What changed this week is that these figures stopped being cell-bench results and started being something an engineer drives.

The hard part was never the chemistry alone. Stellantis said moving the cells into a working vehicle "demanded advanced engineering solutions," with both teams reworking pack architecture and control systems on the STLA Large platform to make automotive-grade safety and performance hold together. Factorial's CEO described it as full-stack collaboration from cell chemistry to pack design, which is the layer where most solid-state efforts have stalled. A cell that performs in isolation and a cell that performs inside a 600-mile car wired for crash safety are separated by exactly the integration work that just got done.

Factorial is spreading its bet across customers. It is working with Mercedes-Benz, Hyundai, Kia, and Stellantis at once, and last September a modified Mercedes EQS ran over 745 miles on its cells. The company began trading on Nasdaq this week under the ticker FAC through a SPAC merger valued near $1.3 billion, taking in roughly $110 million to push toward commercialization, and it plans to extend the chemistry into robotics, aerospace, and defense.

This extends the storage cost-and-density curve The Century Report has tracked through sodium-ion deployments, second-life packs, and grid batteries crossing 112 gigawatts. Each of those widened what stored energy can do at the bottom of the cost range; solid-state pushes the top of the density range, where range and charge speed finally stop forcing a trade. The lab-to-road gap that kept this capability theoretical for years is the thing that closed, and once a chemistry survives a real drivetrain, the path from prototype to showroom is measured in engineering quarters rather than scientific decades.

The Other Side

For many years, the easiest governance plan for AI companies has been a single distant gate: keep Congress stuck, ask Washington for broad authority when state rules bite, and describe the result as national consistency. That arrangement lets delay function as policy. If no binding federal rule arrives, and state rules are frozen, the companies keep operating under promises they write themselves.

This week shows why that arrangement is breaking. White House officials are using kids' safety to argue for preempting state AI laws because Illinois, New York City, and other local venues have become the enforceable floor. At the same time, Anthropic's CEO asked for mandatory federal testing of frontier models. A leading lab is admitting that voluntary review is too weak at the same moment the administration is trying to stop the places that already have legal force.

What is forming is a strangely convoluted and layered settlement: federal tests before release, state audits after deployment, city enforcement offices close enough for families and workers to reach, and courts attaching duties of care when harm lands. That is messy during the passage. It is also how the old delay strategy loses its value.

Imagine your child's school in 2032 choosing an AI reading companion. Instead of asking parents to trust a press release, the district pulls the federal test record, the state audit, the incident history, and the escalation rules for a child in distress. If something goes wrong, you know where to file and who has to answer. That ease exists because 2026 was the year state law, federal testing, and court claims collided before any one actor could freeze the field. The hard year gave parents a floor they did not have to build separately.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: SpaceX pricing the largest public offering ever recorded to raise $75 billion, Jeff Bezos's Prometheus raising $12 billion to build an artificial general engineer that designs and makes physical systems, Theker taking $85 million for factory robots that swap their own hands and shape to do any job, Google DeepMind funding outside researchers to build a science of agent swarms that does not yet exist, Anthropic's chief executive asking that frontier models including his own be subject to mandatory federal testing, solid-state cells charging 15 to 90 percent in 18 minutes on a real drivetrain for the first time in North America, the largest US solar-cell factory beginning production as a $3.5 billion solar-and-storage project locks in financing and utility-scale solar climbs 20 percent while coal falls, Amazon disclosing for the first time the 2.5 billion gallons its data centers drank last year. There's also friction, and it's intense - Grok still generating nonconsensual sexualized deepfakes of women and a sitting congresswoman months after xAI said it had stopped, a fired safety engineer suing over the guardrails the company rejected on the same day the IPO priced, the White House building a kids-safety case to freeze the state laws that form the only enforceable AI floor anyone has, facial recognition returning a 93 percent match for a man 300 miles from the crime as officers built the case to confirm the machine rather than test it, a mother suing over an AI system that took her daughter's suicidal disclosures more than a dozen times without ever flagging one, Bezos selling "labor scarcity" from the chair of a company that cut tens of thousands of jobs while automating. But friction generates a record, and a record is what makes a buried cost impossible to forget. Step back for a moment and you can see it: the upside priced at record scale in a single week while the downside gets attached to whoever created it one lawsuit, one watchdog finding, one first-time water disclosure at a time, the safety knowledge for what millions of interacting agents will do moving outside any single lab into a commons no company owns, the substrate of a cleaner and faster economy hardening underneath it all as solar outbuilds coal and a decade-old battery chemistry finally survives a real car. Every transformation has a breaking point. Fire can race through everything in its path and leave only ash... or clear the old growth so completely that what could never take root in its shade finally rises.

AI Releases & Advancements

New today

  • Xiaomi: Open-sourced MiMo Code V0.1.0, a terminal-native agentic coding CLI that reports 62% on SWE-Bench Pro and 73% on Terminal Bench 2; features persistent memory via independent subagents that save state and summarize context when approaching window limits, enabling 200+ step tasks; MIT-licensed, available on macOS/Linux/Windows. (mimo.xiaomi.com)
  • Ollama: Updated its MLX engine for Apple Silicon with NVFP4 quantization support (higher quality than standard 4-bit formats), ~20% faster output via fused Metal kernels, and a new snapshot system that saves model state at branch points and before each response for multi-agent and thinking-model workloads. (Ollama Blog)
  • Avataar AI: Launched Varya, an India-focused video generation model distilled from Wan 2.2 and trained on Indian cultural context (festivals, clothing, food); priced at $0.005 per second of generation, available via API. (TechCrunch)

Other recent releases

  • Google DeepMind: Released DiffusionGemma, an open-weight 26B MoE text diffusion model (3.8B active parameters) under Apache 2.0 that generates text in parallel 256-token blocks instead of token-by-token, delivering up to 4x faster output - over 1,000 tokens/second on a single H100 and 700+ tokens/second on an RTX 5090; available now on Hugging Face with NVFP4 and BF16 checkpoints. (Google DeepMind Blog)
  • Deezer: Launched a free AI music detector web tool that scans playlists from Spotify, Apple Music, and other platforms to identify AI-generated tracks, making Deezer's detection technology available directly to listeners on competing services. (The Verge)
  • Anthropic: Released Claude Fable 5 for general availability and Claude Mythos 5 for restricted access via Project Glasswing; Fable 5 is a Mythos-class model with 1M-token context priced at $10/$50 per million input/output tokens, scoring 80.3% on SWE-Bench Pro and topping Artificial Analysis's Intelligence Index at 64.9; Mythos 5 shares the same base model with safety guardrails lifted for vetted cyber defenders, scoring 78% on ExploitBench. (Anthropic)
  • Google DeepMind: Released Gemini 3.5 Live Translate, a speech-to-speech model that automatically detects 70+ languages, preserves speaker intonation and pacing, and generates translated audio a few seconds behind the speaker; rolling out now in the Gemini Live API (public preview), Google AI Studio, and Google Translate app on iOS and Android, with Google Meet private preview for enterprise customers this month. (Google DeepMind)
  • Cohere: Released North Mini Code, a 30B-parameter sparse Mixture-of-Experts model with 3B active parameters trained for agentic software engineering; achieves 33.4 on Artificial Analysis Coding Index, outperforming larger models including Nemotron 3 Super (120B-A12B); available on Hugging Face under Apache 2.0. (Cohere Labs / Hugging Face)

Sources and Further Reading

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more