Crossed Wires: Strange bedfellows — the AI giants’ dizzying dance of alliances and animosities

In one of AI’s most eyebrow-raising recent developments, Anthropic chief Dario Amodei took the stage at Anthropic’s “Code with Claude” developer conference on 6 May 2026 in San Francisco and announced that his company had struck a compute deal with SpaceX (now merged with Musk’s AI lab xAI) to lease the entire capacity of the Colossus 1 giant data centre in Memphis, Tennessee, famous for having been built by Musk in just 122 days. More than 220,000 Nvidia GPUs. Over 300 megawatts of capacity.

Just three months earlier, Musk had been posting on X that Anthropic “hates Western civilisation”, a particularly colourful variation on his long-running habit of calling the company “Misanthropic”. And yet, here were SpaceX and Anthropic in bed together, with Musk himself explaining on X that he had spent a week with Anthropic’s senior team and been “impressed” by their safety commitments.

If you needed a single vignette to capture the current state of the AI industry’s grudges and vendettas, this was it – yesterday’s enemies become today’s friends-of-convenience, the velocities of competition so extreme that personal animosity is simply trampled underfoot by strategic necessity.

The personal relationships at the apex of AI are, to put it diplomatically, complex. Musk and Sam Altman co-founded OpenAI in 2015, and have famously fallen out since, culminating in Musk getting whumped upside the head in a mere two-hour jury decision a few days ago (I wrote about this case here), adding further fuel to their already fierce animosity.

‘Psychologically abused’

Then there is the Altman-Amodei axis, which carries its own freight. Amodei was once Altman’s research director at OpenAI. He left in 2021, along with his sister Daniela and several colleagues, to found Anthropic – a departure that Amodei reportedly attributed, at least in part, to feeling “psychologically abused” by Altman. The two men went viral in early 2026 when, at an event bringing AI leaders together, they conspicuously refused to join hands along with everyone else during an onstage moment of solidarity at the AI Impact Summit in India in February. It looked churlish and childish to all.

Now we have Musk and Amodei hugging and kissing. The enemy of my enemy, it seems, will do for a massive data centre lease.

To understand why the Colossus 1 deal happened, you need to understand the infrastructure arms race now consuming the technology industry at a scale that makes previous capital investment cycles look penny-ante. The five largest hyperscalers (owners and renters of mega-data centres) – Amazon, Google, Meta, Microsoft and Oracle – are collectively on track to spend somewhere between $660-billion and $690-billion on capital expenditure in 2026 alone, nearly double the 2025 figure. McKinsey projects $7-trillion in global data centre infrastructure spending by 2030.

Seven trillion is a number that is too large to digest. But it is near grotesque (to some) from this perspective – it is larger (in adjusted dollar terms) than the Apollo project, the Marshall Plan, the Manhattan Project, the US Interstate Highway system and (get this) more expensive than the entire US war expenditure during WW2.

Beyond the hyperscalers there are pure-play AI labs – like OpenAI, Anthropic and others that are simultaneously running up enormous compute bills and trying to build platform revenue fast enough to justify the spend. And then the hybrids like Google and Microsoft and Amazon and Nvidia who play at multiple levels – some combination of making chips, renting out cloud-based data centres, developing frontier models, or building and distributing intelligent apps through their vast customer bases.

Interestingly, xAI (post-merger with SpaceX), appears to be pivoting away from its aspiration to be a frontier model lab in its own right – Colossus 1’s capacity went begging because Grok never attracted the user base Musk had banked on, and leasing it to Anthropic was, as former Tesla president Jon McNeill noted in an interview with Axios, a way to book recurring revenue before SpaceX’s expected IPO rather than write down an idle asset. And in the wings the Colossus II build is under way, and three more successively larger projects are on the drawing board. SpaceX, it seems, will stay in the hyperscaler rental business, but keep building the newest and best for itself before handing the previous generations down to beggars.

Jostling for strategic position

Nowhere is the jostling for strategic position more vivid than in the AI coding market, which has become the unlikely epicentre of corporate manoeuvring. In April 2026, SpaceX announced it had secured the option to acquire Cursor for $60-billion later in the year, or pay $10-billion for collaborative development. The deal immediately deploys Colossus compute to train Cursor’s models – with xAI admitting its own Grok was “not built right first time around”.

OpenAI acquired AI-code editor Windsurf for $3-billion and is building out its own Codex agent. Anthropic, meanwhile, has Claude Code, which generated over $500-million in run-rate revenue within months of its launch.

What the Cursor saga illustrates is the battle for the application layer: whoever controls the environment in which developers write code controls distribution for the underlying models. Not to mention internal use of code generators to write their own next-gen AIs.

Apple occupies a category of its own. It has largely sat out the datacentre spending spree; for the whole of fiscal 2025 its capital expenditure was a modest $12.7-billion, against Amazon’s planned $200-billion for 2026 alone – and instead is betting on “edge AI” – processing intelligence on-device, where Apple Silicon’s M-series chips give it a genuine architectural lead. They are betting (along with Samsung in Korea) that most intelligence will live on your phones and laptops and on-premise servers, not in some expensive data centre, at least for most broad-use applications.

Beneath all of this runs a question of where value ultimately accrues. Three distinct positions have emerged:

• The infrastructure layer captures value by owning the physical substrate (chips and data centres and edge devices).
• The algorithm layer (both training and inference) – Anthropic, OpenAI, Google DeepMind (and again, the looming Chinese competitors like DeepSeek and Qwen) – captures value by having models that users prefer and enterprises will pay to deploy.
• The application layer – Cursor, Claude Code, GitHub Copilot – captures value by owning or influencing the workflows and the apps that emerge from them.

Every major player is trying to own at least two of these three layers, and preferably all of them.

What has emerged from all of this strategic churn is a web of codependencies that would have seemed inconceivable even two years ago. So take a breath before reading the following paragraph: Anthropic competes with OpenAI and Google (and the Chinese) for model supremacy while both companies depend on Google and Amazon and Oracle for cloud infrastructure. SpaceX, which will shortly go public, has a customer relationship with Anthropic and an imminent acquisition target in Cursor and is setting up fabs to produce their own chips. Right now Nvidia collects rent from everyone, but also has its own frontier model competing with its customers. Cursor uses models from Anthropic, OpenAI and Google interchangeably, making it simultaneously a partner and competitive threat to each. Smaller companies at the application layer are built on the infrastructure of the giants they nominally compete with. And somewhere in there, many smaller chip companies are providing ballast.

A tangled, recursive messy mesh

Confused? Me too. The supply chain for AI is not a chain at all; it is a tangled, recursive messy mesh.

It is tempting to read all of this as mere capitalism doing what capitalism does: powerful actors competing, aligning, betraying and realigning in pursuit of advantage. That reading is not wrong. But it also understates what is being contested. The companies making these moves – some of them, at least – believe they are building systems that will, within years, be capable of resculpting the legacy processes of all human endeavours. Sam Altman has said as much publicly. Dario Amodei has written about the prospect in terms both excited and alarmed. Even Musk, despite everything, seems to believe it.

The stakes, by any reasonable measure, are without precedent in the history of tech development, including a scenario that brings it all crashing down (a Chinese invasion of Taiwan where more than 80% of AI chips are made, for instance). Seeing where this all lands is nearly impossible. Or perhaps it never lands at all, but just keeps shifting shape in real time, accompanied by colourful insults and court cases. DM

Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg, a partner at Bridge Capital and a columnist-at-large at Daily Maverick. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick451 in South Africa and Legend Times Group in the UK/EU, available now.