Rockets, models and a $60 billion IDE: the six months that changed the AI map

A lot has happened since I wrote the December wrap-up. Some of it I expected. Most of it I didn't.

Anthropic: the quiet compounder keeps compounding

Back in December I noted that Anthropic had gone from a $61.5 billion valuation to $183 billion in the space of a few months, and described it as compound execution with no drama. That trend has not broken. By the end of 2025, Anthropic's annualised revenue had reached $9 billion. By May 2026, research firm Sacra estimates it had hit $47 billion annualised — a number that, if accurate, would represent one of the most compressed revenue ramp-ups in the history of enterprise software.

In February 2026, the company launched an employee tender offer sized at $5–6 billion, allowing current and former staff to sell shares at an implied valuation of approximately $350 billion. Then in the spring, Anthropic closed a $30 billion Series G at a $380 billion post-money valuation — a figure that now puts it in the same conversation as the world's most valuable public companies, let alone private ones. The round was led by Coatue. The company has also confidentially filed a draft S-1 with the SEC, taking a cautious first step toward a public listing. No date has been set. But the direction of travel is clear.

What's striking about Anthropic's trajectory isn't just the numbers — it's the consistency of approach. No splashy consumer moments. No product launches timed to news cycles. Claude Code, the company's agentic coding tool, has quietly become a significant revenue driver in its own right. Over 300,000 businesses are now customers, with over 100,000 running Claude on Amazon Bedrock. The customers spending more than $100,000 annually have grown sevenfold in the past year. These are not the metrics of a company burning cash to stay relevant. They are the metrics of a company that has found repeatable enterprise motion.

SpaceX: the largest IPO in history, and the most complicated company on Earth

SpaceX listed on Nasdaq on 12 June 2026 under the ticker SPCX, targeting a $1.75 trillion valuation and aiming to raise $75 billion — the largest IPO in stock market history, surpassing Saudi Aramco's $29 billion raise in 2019 by a factor of more than two. On its first day of trading, SpaceX's market cap climbed to approximately $2.4 trillion. Morningstar called it 'significantly overvalued'. Plenty of retail investors disagreed.

The financial picture underneath the headline is worth understanding. SpaceX generated $18.7 billion in revenue in 2025 but posted a loss of nearly $5 billion — a reversal attributed largely to the integration of xAI, Musk's AI venture, which SpaceX acquired in July 2025 in an all-stock deal valuing xAI at $250 billion. In Q1 2026, revenue was $4.69 billion against net losses of $4.27 billion. The AI segment alone lost $6.36 billion in 2025.

Starlink, however, is a different story: $11.4 billion in 2025 revenue, $4.4 billion in connectivity operating income, and a subscriber base that surpassed nine million worldwide. It is the only part of SpaceX's business that currently looks like a conventional, profitable infrastructure company. The rest is a bet — on space-based compute, on Grok as a competitive AI platform, and on the idea that owning the infrastructure layer across rockets, satellites, data centres and the Colossus supercomputer eventually pays off at scale.

Grok's credibility problem — and the Cursor fix

Inside xAI's integration into SpaceX, the picture has been messier than the headline numbers suggest. xAI's Grok launched Grok Code Fast 1 in August 2025, positioning it as an agentic coding tool built from scratch for programming-heavy workloads. By May 2026, SpaceX launched Grok Build in early beta — a coding agent and CLI aimed at professional software engineers. The ambition was clear. The execution, less so.

The problem Grok faces isn't model quality in isolation — Grok 4 series models have performed credibly on benchmarks — it's the product layer. Developers don't choose AI coding tools based on benchmark scores. They choose them based on workflow fit, reliability, and trust accumulated over months of daily use. By May 2026, Cursor's market share among enterprise developers, which had peaked at 41% in mid-2025, had declined to around 26% according to spending data from Ramp. That still made Cursor the most widely adopted enterprise coding tool on the market.

On 21 April 2026, SpaceX announced an agreement giving it the right to acquire Anysphere — the parent company of Cursor — for $60 billion, or to pay $10 billion for an ongoing compute-and-collaboration partnership using xAI's Colossus supercomputer. The acquisition is expected to close in Q3 2026 and is structured in SpaceX stock. Cursor is used by more than half the Fortune 500 and was generating over $4 billion in annualised revenue. Cursor CEO Michael Truell described it as 'a meaningful step on our path to build the best place to code with AI.' Jensen Huang had previously called Cursor his 'favourite enterprise AI service'. Patrick Collison said every one of Stripe's 40,000 engineers uses it.

The strategic logic isn't subtle. Colossus has the compute. Cursor has the distribution and the developer trust. SpaceX needs AI revenue, and it needs it to be credible enough to justify the AI segment's valuation. Enterprise buyers using Cursor now face a period of vendor-identity ambiguity — model-neutrality assumptions that made Cursor attractive to non-xAI shops are now subject to renegotiation. The more precise read is that SpaceX is buying the one thing Grok cannot quickly replicate: a product layer that developers already trust inside their daily workflow.

What this means for the map

Six months ago I wrote that the battle lines between Anthropic, OpenAI, Google and the Musk empire were drawn more clearly than at any point this decade. That's still true, but the terrain has shifted. The contest has moved from model quality — where all the frontier labs are now broadly competitive — to distribution, infrastructure and enterprise stickiness.

Anthropic has enterprise stickiness and is building toward public markets with extraordinary revenue velocity. OpenAI has consumer scale and is heading toward its own IPO at a valuation that will test market credulity. Google has infrastructure and search distribution but continues to struggle to turn frontier model quality into enterprise momentum. And SpaceX has everything and nothing: compute, rockets, satellites, Starlink cash flows, and an AI product that still needs to earn daily developer trust.

The Cursor deal is SpaceX trying to buy time. Whether it works depends entirely on what developers decide to do when they realise the IDE they use every day is now owned by Elon Musk. Some won't care. Some will. The answer will tell us a lot about where the centre of gravity in enterprise AI actually sits. I'll be watching. As always, so will everyone else.