Money talks
AI geopolitics, compute futures, and that moment when the models left the building
THE COMPUTE / BY TOKENANDO Issue II | May 14, 2026
President Trump arrived in Beijing yesterday for the first US state visit to China since 2017, accompanied by a delegation of US technology and business leaders, including Apple chief executive Tim Cook, Tesla chief executive Elon Musk, and — added at the last minute after a personal call from the President during a refuelling stop in Anchorage, Alaska — Nvidia chief executive Jensen Huang. AI and semiconductor access are now explicitly on the formal agenda of the most consequential global meeting of the year happening now at head-of-state level with the world’s most important chip maker in the room.
That confirms that AI has evolved from a technology conversation to a capital, infrastructure, and now geopolitical one. This issue of The Compute sheds a light on these three threads, which are becoming impossible to separate.
A note to our readers
If you are already feeling lost, and practically have very little idea what we are talking about here, you are not alone. This is something we have been hearing a lot from our readers since we started publishing The Compute as well as posting on the @tokenandoAI social media channels is that while the topic of AI is super interesting, our content is highlighting that the foundational knowledge among our audience is not yet in place.
While everyone has become an AI user very quickly thanks to the rapid spread of the technology, and the ease of using ChatGPT, Claude or Gemini; most people have not had the reason or opportunity to become fluent with concepts such as AI labs, providers, compute infrastructure, tokens and model pricing, and the difference between training, inference and agentic.
As the pace at which AI is becoming strategically relevant is outrunning the time available to learn these concepts, we at Tokenando are building something to close that gap: a focused programme for non-technical leaders who need to understand the business of AI, its value chain and the underlying economics.
Reply “teach me!” and we will make sure you hear about it first.
Vendor Moves
The models have left the building.
On May 4 and May 11, Anthropic and OpenAI each announced that they are entering the enterprise deployment business through private equity-backed entities. Anthropic launched first, with a $1.5bn joint venture seeded by Blackstone, Hellman & Friedman, Goldman Sachs, Apollo, General Atlantic, and others. OpenAI followed with the OpenAI Deployment Company (DeployCo), raising $4bn at a $10bn pre-money valuation, with TPG leading and 18 additional partners including Bain Capital, Brookfield, SoftBank, and Warburg Pincus.
Three management consultancies — Bain, Capgemini, and McKinsey — are among DeployCo’s backers, all three with core businesses built on enterprise transformation, all three now funding a company whose explicit purpose is to replace a significant portion of that work. The logic behind both announcements is the same: the gap between what AI models can do and the value enterprises are actually extracting from them is “enormous,” in the words of Anthropic’s head of product for financial services Nicholas Lin. Private equity firms own thousands of mid-market companies without the in-house capability to close that gap, turning them into a valuable distribution channel for the AI labs.. DeployCo has already acquired Tomoro to staff up immediately, bringing clients including Tesco, Red Bull, Mattel, and Virgin Atlantic.
Ben Bajarin of Creative Strategies, writing in his most recent report this week, explains the deeper logic: the model capability race is commoditising intelligence, and the next competitive front is enterprise software, the layer that sits between the model and the workflow and determines which vendor captures a chunk of the budget. The labs are not entering consulting because they want to be McKinsey; they need to own the software layer before Salesforce, ServiceNow, and Microsoft get there through existing enterprise relationships.
One announcement passed largely unnoticed this week: on the day jury selection opened in the Musk v. OpenAI trial, OpenAI and Microsoft revised their partnership agreement, removing the revenue share cap and giving OpenAI freedom to work with any cloud provider. This happening at the same time as both announcements is not a coincidence. All eyes are on the IPOs expected in autumn of the year, and every move is strategic in that direction.
Capital Watch
Compute is the newest asset class.
At the Milken Institute Global Conference in Beverly Hills on Tuesday, BlackRock chief executive Larry Fink said something that received less attention than it warranted. “A new asset class will be buying futures of compute. We just don’t have enough compute power right now.” He added:
“We are short power, we are short compute, we are short chips. I do not believe we are moving fast enough.”
BlackRock’s Global Infrastructure Partners is acquiring Aligned Data Centers for approximately $40bn, securing more than five gigawatts of capacity across 50 Americas campuses, and Mr Fink previewed a further one-gigawatt data centre partnership with an unnamed hyperscaler at the same conference. A single one-gigawatt facility now costs between $50bn and $75bn to build.
While Mr Fink did not give a specific timeframe, the direction is clear: compute is moving along the same trajectory that energy and agricultural commodities, from scarce physical resource to financialised instrument with a derivatives layer on top.
For more on these topics, check out the March 2026 arXiv paper by researcher Yicai Xing, who has already proposed a full compute futures contract design under CFTC jurisdiction, with an estimated optimal launch window of 2027 to 2028, and the May 2026 MTS exposé “The Economics of Data Centers”, with a seven-step walkthrough of the Meta / Blue Owl Hyperion data center deal.
On the Horizon
Beijing
The Trump–Xi summit concludes May 15 with an agenda spanning chip access, rare earth supply chains, AI governance, Taiwan, and the Iran conflict: the most concentrated set of technology and security disputes ever placed on a bilateral table at head-of-state level. One key negotiation item is whether China’s commercial AI sector gains access to Nvidia H200-class chips beyond the current case-by-case approvals with a 25 per cent surcharge, but the stakes are broader than one chip generation.
The Gulf precedent is relevant here: the HUMAIN and G42 chip access agreements concluded in 2025 established that sovereign AI programmes can negotiate direct access to advanced US hardware. Whatever framework emerges from the summit will signal to Riyadh, Brussels and London what Washington is actually willing to trade, and on what terms.
Oakland
The Musk v. OpenAI trial reaches closing arguments today after three weeks of testimony that put Elon Musk, Satya Nadella, and Sam Altman on the stand. The jury will decide whether OpenAI’s conversion from non-profit to for-profit constituted a breach of charitable trust; a verdict that carries implications for how every AI lab in the country thinks about governance, fiduciary obligation, and the relationship between stated mission and commercial structure.
Issue III of The Compute will cover both outcomes next Thursday, however stay tuned to our LinkedIn and X channels for updates in between
Wirelessly yours,
Ziad Matar
Co-founder, Tokenando
Editor-in-chief, The Compute
After Hours
This issue of The Compute was bookended by Negronis (a classic, and several sbagliatos), and anchored by a good ribeye accompanied by a good conversation on a sunny Barcelona terrace. Words were written over a background dominated by lyric-less Olafur Arnalds, Hania Rani and Nils Frahm music. The ongoing binge of Margo’s Got Money Troubles, raises the question of whether OnlyFans would be a monetisation channel for Tokenando at some stage.