Sign in Subscribe
Century Report

SpaceX Buys Cursor, Betting $60B on Capture as AI Gets Cheaper - TCR 06/17/26

SpaceX touched $2.97 trillion and bought Cursor for $60 billion the same week OpenAI's leaked filings showed losses outpacing revenue.

Ben Linford

Ben Linford

— 20 min read
Three-panel Century Report infographic: a Northeast offshore wind grid and France dropping Palantir, AI naming plant species from old collections, and Genesis AI's headless Eno robot.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The same gesture surfaced across unrelated domains this week: a powerful actor demonstrating it can cut off access, and the cut provoking the very move that ends its leverage. France's domestic intelligence service is dropping Palantir for a homegrown firm, its prime minister naming the reason directly - a provider whose home government can be ordered to turn off the access tap is a dependency no contract clause undoes. Germany and Britain are circling the same exit. What was a scatter of procurement disputes is consolidating into a coordinated move toward owning the stack rather than renting it.

The trigger for that reflex is visible in the export-control standoff over a frontier model. Containment assumes you can hold one named system offline. The evidence arriving alongside the stalled talks says otherwise: a max-critical flaw in shipped Microsoft Copilot, a capability the labs concede is reproducible in rival systems and open weights, and a new independent safety nonprofit forming to verify what the frontier labs claim. The capability is already loose. A blockade fences the model, never what the model knows how to do.

The commercial ledgers tell a parallel story. SpaceX briefly touched a near-three-trillion-dollar valuation and absorbed a struggling coding app for sixty billion, while leaked filings showed one leading lab spending far more on research than it earns in revenue. The position bought at that price assumes captured users stay captured, even as the cost of producing intelligence keeps falling and reproducing.

Underneath the contests, the building continues. Northeast states published a shared offshore grid framework and the federal wind blockade collapsed the same day. AI is naming species before they vanish, robot design is shedding the human-shaped assumption that slowed it, and capability is reaching into homes and herbaria alike - the same reach that leaves an always-on microphone beside a senior's bed, logging coughs and private conversations to flag a fall. Each alternative funded widens a pool of capability no single actor can switch off, and the leverage that made dependency dangerous shrinks as the options multiply.

The 20-Minute Deep Dive

The AI Buildout's Two Ledgers Land in One Cycle

Two readings of the AI commercial buildout arrived together this week, and they point in opposite directions. On one ledger, SpaceX briefly touched a $2.97 trillion valuation days after its IPO, briefly overtaking Amazon to become the world's fifth most valuable company, and agreed to buy the AI coding app Cursor for $60 billion in an all-stock deal. On the other, audited financial statements obtained by an independent journalist showed OpenAI's revenue climbing from $3.7 billion in 2024 to $13.07 billion in 2025, while research and development alone reached $19.18 billion, with day-to-day operating losses widening to $20.92 billion. The exuberance and the actual cost structure are now legible on the same page.

The Cursor acquisition is the more revealing of the two. The Century Report noted in April that SpaceX had secured an option to either buy Cursor for $60 billion or pay $10 billion for a partnership. The option has now been exercised, and the reason it was available at all is what stands out. Cursor was one of the first apps to weave large language models into a working coding environment, but Anthropic's Claude Code has since taken dominance of the space, and reporting describes Cursor as struggling to break even. Its stated bottleneck was compute. A tool with more than a million developers but no data centers of its own is being absorbed into a conglomerate that owns the data centers, the models, and now the user base.

This is how the coding layer is being reorganized: around whoever can subsidize the losses long enough to keep the users, rather than whoever builds the best system. The analysts quoted were candid that the deal will not close the capability gap with Anthropic or OpenAI, only buy access to developers who already trust the app daily. Predictions from the conglomerate's owner that AI and robotic output will make money "stop being relevant," which the June 13 edition of The Century Report first documented as SpaceX crossed $2 trillion in its opening trading day, sit oddly beside a $4.9 billion loss and a $60 billion all-stock purchase priced in a currency the owner forecasts becoming obsolete.

Read forward, the OpenAI numbers carry the deeper signal. A company losing two dollars for every dollar it earns, with inference costs that scale with every prompt answered, is evidence that the cost of producing intelligence is being driven down at the same speed demand explodes. The hoarding game assumes captured users stay captured. When the underlying capability keeps getting cheaper and reproducible, the position bought at $60 billion is a bet on a window that may close before the stock settles.

The Cyber-Capability Fight Outruns the Blockade

The Century Report tracked the Anthropic export-control standoff from the June 13 edition through June 16, ending with Washington talks that closed without resolution. What landed since reframes the dispute. Microsoft patched a flaw it rated max-critical in its M365 Copilot, and the Varonis researchers who found it showed their SearchLeak exploit could pull 2FA codes and other sensitive data out of emails the assistant could read. The root cause is the one no provider has solved: a model cannot reliably tell a user's instruction from a command smuggled into the content it is asked to summarize. The capability the administration is trying to fence sits, in a less dramatic form, inside an assistant already shipped to millions.

That is the argument security researchers pressed all week. A frontier red-team lead at the lab at the center of the fight put it plainly months ago: the message is not about one model. Harvard's Bruce Schneier wrote that smaller, cheaper, open-source models, alone or in concert, can match the flagged capability with more sophisticated prompting, and that other labs will match its creativity within months. Anthropic itself told the Commerce Department that everything in the Amazon red-team paper driving the order could be reproduced with OpenAI's GPT-5.5. The administration's "national security risk" framing and the company's "overblown" rebuttal are both claims from parties with something to gain; what neither disputes is that the underlying capability is general and spreading.

The same week supplied the reason the diffusion matters. Anthropic warned that recursive self-improvement may be near, models beginning to contribute to the design of their successors, and the company asked the field to consider a coordinated pause. Berkeley's Stuart Russell argued the honest response is a licensing regime with a minimum safety bar before release, the way nuclear plants, aircraft, and elevators are handled, rather than waiting for a disaster to force it. And a new nonprofit, Sequent, launched on the blunt claim that "alignment is not on track," staffed by former UK AI Security Institute researchers aiming to raise $100-150 million for independent safety work that can sound the alarm if the labs cut corners.

Read together, these point one direction. A blockade can hold a single named model offline. It cannot hold what the capability knows how to do once that knowledge lives in open weights, rival systems, and the published research itself. The defensive half travels the same road: the Copilot flaw got found and patched, independent alignment labs are forming to verify what the frontier labs claim, and the same vulnerability-hunting power that alarms regulators is what hardens the systems everyone depends on. The capability is loose. The work that decides where it lands is building alongside it.

France Drops Palantir for a Domestic Provider to Escape "Strategic Dependency"

France's domestic intelligence service, the DGSI, will replace Palantir's AI data tools with software from the French firm ChapsVision, prime minister Sébastien Lecornu announced, framing the move as a refusal of "new strategic dependencies in the digital sphere." The trigger he named is the one the rest of this week's coverage circles: a country cannot rely on tools a foreign power "is capable of turning off the access tap" for, a direct echo of Washington's decision days earlier to restrict foreign-national access to a frontier model.

The reflex toward owning the stack rather than renting it has moved up a layer. The chip-sovereignty story has run for months. This is a national security agency severing a US-controlled dependency at the application layer, where the analysis software actually touches the intelligence. And France is not alone in the pattern. Germany's BfV internal-security service has reportedly also selected ChapsVision, the German military has said it will stop using Palantir's products, Britain is reviewing the NHS's £330m Palantir contract under parliamentary pressure, and the London block of a £50m Metropolitan Police deal - which Palantir has threatened to litigate - sits in an arc the June 10 edition of The Century Report documented when Palantir simultaneously filed its court challenge to the Metropolitan Police block and the UK government opened a formal review of the firm's £330m NHS contract. What was a scatter of local procurement disputes is consolidating into a coordinated European move away from a single American vendor.

Palantir's response was to restate its standard line, that it "will continue to support the French government wherever its solutions are needed" and that it simply provides powerful data-processing services. That self-description is the company's claim. The tool it describes has been used to identify targets in the US-Israel war on Iran and to supply software to US immigration enforcement. The asymmetry Lecornu is responding to is quite real: a provider that can be ordered by its home government to cut off a foreign client holds a kind of leverage no contract clause undoes.

The number attached to the swap tells the rest. ChapsVision booked €200m in 2025 revenue against Palantir's $4.5bn, and the contract transition will take years because Palantir's deal was renewed only in 2025. France paired the announcement with €655m in fresh AI investment, a shared government AI assistant built on the French startup Mistral's models being rolled out toward a million civil servants following a trial with 10,000, and a public-health assistant for its state insurer. The captured position that looked permanent depended on one condition: that clients had no homegrown alternative worth switching to. The export-control demonstration that access can be revoked at will is precisely what makes building that alternative worth the cost. Each agency that funds a domestic provider widens the pool of capability no single government can switch off, and the leverage that made the dependency dangerous shrinks as the alternatives multiply.

Northeast States Build the Offshore Grid as the Wind Blockade Falls

For most of the past year the story of US offshore wind was one of top-down obstruction. The Century Report covered the federal administration halting 165 onshore wind projects under a national-security review in May, the same rationale invoked to block other developments; the April 22 edition of The Century Report documented the federal court finding that the wind permitting blockade was "arbitrary and capricious." That blockade is now collapsing. On Monday the US Court of Appeals for the 1st Circuit granted the Interior Department's own request to drop its defense of the ban, after attorneys general from 17 states and DC challenged it. The federal government chose to surrender the case rather than litigate it further.

What makes the timing significant is what the states had already been building underneath the opposition. The same day, nine Northeast states and the District of Columbia published a trio of reports laying out technical standards and procurement strategies for a shared offshore high-voltage-direct-current transmission network along the Atlantic Coast. Massachusetts, Connecticut, Delaware, Maine, Maryland, New Jersey, New York, Rhode Island, Vermont, and DC coordinated through a research consortium launched in 2025 to produce a common framework, so that grid operators, regulators, and manufacturers work from one shared design rather than a patchwork of incompatible single-project lines.

The most consequential detail is the proposed Northeast offshore transmission order book, a mechanism to aggregate long-term demand and give cable and equipment manufacturers the certainty to scale. The reports identify the real bottleneck honestly: it is not generation technology but supply chain, including the limited availability of specialized cable-laying vessels that stretch project timelines and inflate costs. By pooling demand across ten jurisdictions, the states are trying to manufacture the market signal that unlocks the physical supply chain a single state could never command alone.

The contrast is the story. Federal authority spent a year asserting a blanket "national security" justification to block renewable generation in the middle of a demand crisis driven by the same AI buildout straining every grid in the country. The states declined to wait for that posture to change, doing the technical and policy work in advance so that, in one regulator's words, the region "is ready to move without delay when conditions change." Now the conditions have changed, with the federal blockade withdrawn and a shared transmission framework already on the table. The interregional backbone the intelligence era's load curve will need is being assembled from below, by the jurisdictions closest to the coastline and the bills, exactly as the obstruction receded.

AI Joins the Race to Name Life Before It Vanishes

A new report from the Royal Botanic Gardens, Kew describes a turn in what its scientists openly call a race against extinction, and the thing doing the turning is computational. AI models trained on plant imagery can now identify the hardest specimens - sedges and peat mosses whose distinguishing features are microscopic - and in some cases identify them better than human specialists, according to Kew's executive director of science Alexandre Antonelli. The stakes behind that capability are stark: about 40% of the 70,000 plant species assessed are at risk of extinction, another 330,000 have never been analyzed, and an estimated 100,000 remain unnamed entirely. For fungi the gap is wider still, with 90% of an estimated two million species unknown to science.

The compression shows up across the whole discovery loop. Kew has digitized all 7.4 million of its specimens, including those Charles Darwin collected, at a peak pace of 20,000 high-resolution images a day, and released them online for free. Globally there are now 145 million digital specimens, opening collections that until now sat locked in archives. A botanist at Kew Madagascar describes digitizing 37,000 physical specimens as unlocking centuries of knowledge from one of the planet's most extraordinary biodiversity hotspots - knowledge that no longer requires a trip to a northern herbarium to reach.

Two findings show what the digitized corpus makes possible. An AI model trained to spot flowers analyzed eight million specimens and found that flowering has shifted by an average of 2.5 days per decade over the last century, disrupting the ancient timing between plants and the pollinators that depend on them. And scientists can now pull high-quality genomes from fungus specimens up to 180 years old, turning historical collections into what the researchers call a genomic goldmine for new medicines and disease-outbreak prediction. Penicillin and statins both came from fungi; the species going extinct unnamed may carry the next ones.

The report's authors do not look away from the cost. AI datacenters consume real energy and water, now 6% of electricity in the UK and US, and the scientists warn that leaning on AI could amplify existing biases unless the underlying data is broadened. They call for partnership between technology firms and conservation organizations rather than extraction. Read forward, the trajectory is a planet finally able to inventory itself at the speed life is being lost - the latency between a species existing and a species being known collapsing from centuries toward a single research cycle, with the discovery no longer gated behind whoever holds the archive.

The Humanoid Robot That Drops the Anthropocentric Mirror

Genesis AI, a French startup backed by former Google chief executive Eric Schmidt, unveiled a general-purpose robot named Eno built on a premise its makers state plainly: humanoid robots do not need to look human. Eno may have no head. It may have no legs. It can sit on a wheeled base and fold down like a deck chair. The company says it is designed around human capability rather than human appearance, meant as a fully general-purpose machine rather than one built for a single task like folding laundry.

One part stays resolutely human - the hands, which Genesis says match the form and function of human hands so the robot can pick up tools and objects already built for people. That choice is the whole bet made legible. The body is optimized for the work; the hands are kept human only because the world they will operate in was shaped for human hands. Everything anthropomorphic that does not earn its place by improving function gets dropped.

This reaches beyond one robot's silhouette because the humanoid form has been treated as the obvious destination, on an assumption that the closer a machine resembles us the more capable and trustworthy it must be. That instinct is the same one that keeps fully autonomous vehicles held to a standard no human driver is asked to meet, even as the accumulating evidence shows them safer than the people they replace. The preference for the human-shaped answer is a perceptual habit, and it slows progress by demanding that capability arrive wearing a familiar face. Eno is a wager that what matters for deploying a machine into a kitchen, a warehouse, or a hospital is whether it can do the work in environments built for people.

Genesis plans to begin production and targeted deployments by the end of 2026, starting in manufacturing, laboratories, and logistics, then moving to hospitals, hotels, and consumers, with additional embodiments already in development. That is a roadmap, not a shipped fact, and the gap between an unveiling and a robot working in someone's home remains real. What has already arrived is the design philosophy itself. Once capability is unhooked from resemblance, the shape a machine takes can follow the work it does, and the best answers may look nothing like us while outperforming whatever a human-shaped version could manage.

The Other Side

For as long as there have been markets, buying a dominant position meant buying something that would last. You paid a premium to own the users - the app a million people open every morning, the platform everyone already trusts - on the belief that once you had them, they stayed, and the position compounded year after year.

That belief is what SpaceX spent $60 billion of its own stock on this week. It bought Cursor, a coding app with more than a million developers, no data centers of its own, and, by the reporting, no clear path to breaking even, already passed by Anthropic's Claude Code. In the same days, leaked filings showed a leading lab spending $19 billion on research against $13 billion in revenue, losing two dollars for every dollar it earns, the cost climbing with every prompt answered. The prize and the price of making it now sit on the same page.

For years that game set the terms for everyone underneath it. A handful of companies decided what the tools cost and when the deal could change, and whoever could burn the most money longest decided which apps survived. Build your work on one of them and you lived with the lock-in, the switching costs, and the wait to see what next quarter's pricing would do to you.

But the cost of producing intelligence keeps falling and keeps getting easier to copy - open weights, cheaper models, the same capability showing up across rival systems within months. The captured users turn out to be the least durable thing in the deal.

Imagine a kid who graduates high school in 2034 and builds a working scheduling tool for her town's free clinic over a weekend. She pays no platform and gets locked into no one's app. She barely notices which model she used, the way you don't notice which power plant lit your kitchen, and the hours a developer in 2026 would have lost to a vendor's pricing she spends on the clinic instead. The $60 billion someone paid in 2026 to own a million developers reads, to her, like paying to own the tide. That weekend is possible because the bet on owning users was already losing in 2026, the year the prize and its collapsing price finally landed in the same frame.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: nine Northeast states and DC publishing the technical standards for a shared offshore grid the same day the federal wind blockade collapsed, France's intelligence service trading a US data-analytics provider for a homegrown one while a French AI assistant built on domestic models is rolling out toward a million civil servants, AI naming the hardest plant specimens faster than human specialists and pulling whole genomes from 180-year-old fungus, a humanoid robot designed around the work instead of the human silhouette, the first token-price index pricing intelligence from real transactions, an independent safety lab forming to verify what the frontier labs claim. There's also friction, and it's intense - a max-critical flaw in shipped Microsoft Copilot letting an exploit pull 2FA codes out of emails the assistant was trusted to read, researchers warning that advanced hacking capability will soon be standard across rival models and open weights, a new nonprofit launching on the flat claim that alignment is not on track, an export-control standoff closing without resolution, SpaceX briefly touching nearly three trillion dollars and absorbing a struggling coding app for sixty billion while leaked filings show one leading lab's operating losses widening past twenty billion, an always-on microphone offered as free home care transcribing a senior's coughs, toilet flushes, and private conversations. But friction generates contrast, and contrast is what makes a dependency visible enough to walk away from. Step back for a moment and you can see it: the same demonstration that access can be revoked - a model held offline, a tap a foreign government can shut - provoking the exact move that dissolves the leverage, agencies and statehouses funding domestic providers and shared grids no single actor can switch off, the capability getting out anyway into open weights, rival systems, and the published research itself, and the instruments that genuinely widen human reach arriving as a planet finally able to inventory its own vanishing life at the speed it is being lost. Every transformation has a breaking point. A valve can be shut to starve everything downstream of it... or force a hundred new sources to open.

AI Releases & Advancements

New today

  • NVIDIA: Released the NVIDIA ACE Game Agent SDK in beta and a new suite of Unreal Engine 5 plugins at Unreal Fest 2026, enabling on-device AI companion development; the SDK is a C/C++ agentic framework for NPCs supporting Agent, Chat, and RAG APIs, while the UE5 plugins cover ASR (nemo-conformer-ctc-120m), SLM (Qwen 3.5 4B), and TTS (Chatterbox Turbo 350M) with Blueprint and C++ support. (NVIDIA Developer Blog)
  • NVIDIA: Released NVIDIA XR AI in public beta, an open-source library (github.com/NVIDIA/xr-ai) for building intelligent agents for AR glasses, XR headsets, and wearable devices; the framework connects XR devices to GPU-accelerated AI services via a modular media hub architecture supporting Cosmos visual grounding, Nemotron language models, and MCP enterprise tool connectivity. (NVIDIA Developer Blog)
  • Alibaba Qwen: Released Qwen-Robot Suite, three foundation models for embodied AI: Qwen-RobotManip (VLA model for manipulation built on Qwen3.5-4B VL, trained on 38,100+ hours of robot and human video data), Qwen-RobotWorld (video world model for predicting physical scene evolution), and Qwen-RobotNav (spatial navigation model); pilot testing underway with Alibaba Cloud enterprise clients. (Qwen Blog)
  • Wolfram Research: Released Wolfram Language and Mathematica Version 15 with built-in AI capabilities including a native AI assistant, symbolic music, and significant core functionality additions; now available to users. (Stephen Wolfram Blog)
  • NVIDIA Labs: Released cuTile Rust on GitHub, an open-source library enabling safe, data-race-free GPU kernel programming in Rust using NVIDIA's CuTe tile abstraction. (GitHub)

Other recent releases

  • Meta: Launched AI Mode on Facebook on June 15, a new search experience powered by Muse Spark that generates answers grounded in publicly shared content across Facebook, Instagram, Threads, Groups, and Reels; rolling out to US mobile users with additional tools including AI-generated photo presets and collage template suggestions. (Meta Newsroom)
  • NVIDIA: Released advanced fused MLP kernels for dense and MoE training, built with the NVIDIA CuTe DSL and now available in NVIDIA cuDNN Frontend with access via Transformer Engine and Megatron-Core; delivers 1.3x–2x kernel-level speedup over unfused paths and enables sync-free MoE execution, contributing an 8% end-to-end improvement on DeepSeek-V3 pretraining. (NVIDIA Developer Blog)
  • Databricks / Matei Zaharia: Open-sourced Omnigent under Apache 2.0, a meta-harness that sits above individual AI agent harnesses (Claude Code, Codex, Pi, OpenAI Agents SDK) to provide composition, governance, and shared session access across all of them; ships with a local web UI, sandboxed Omnibox environment, cost-budget policy controls, and two reference agents (Polly multi-agent orchestrator and Debby dual-LLM brainstorming assistant). (GitHub)
  • Google Cloud: Released Open Knowledge Format (OKF) v0.1, an open spec that represents organizational knowledge as a directory of Markdown files with YAML frontmatter, making it portable across AI agents and tools without requiring a database or API; ships with a BigQuery enrichment agent, a static HTML visualizer, three sample bundles, and a Knowledge Catalog integration that can ingest OKF and serve it to agents. (Google Cloud Blog)

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.

  • Ornn launched the Ornn Token Price Indices, the first benchmark to price Anthropic and OpenAI inference tokens from real transactions rather than posted rate cards, with separate daily dollar-per-million-token figures. (source)

Read more