For the past year, as AI models kept getting better and seeping into work and personal lives, I kept noticing the same thing: while they are very capable at many tasks, they utterly lack taste. That undefinable element is a frontier where people still hold an advantage, and hopefully we will hold it for the long term. But lately, human content has been buried under AI slop, the machine-written padding that fills our feeds and nobody wants to read.
Pangram, which builds AI detection software, found in a study this summer that more than 40 per cent of long-form LinkedIn posts were entirely machine-written, roughly double the average across all the platforms the study covered. LinkedIn retired its own “enhance your post” writing assistant on July 30, 2026, replaced it with a proofreader that leaves the writer’s voice alone, and more importantly, added a button allowing users to report content that “seems like AI slop”.
Nine days earlier, on July 21, 2026, Chris Best, Substack’s chief executive, coined the term “Claudefishing”: a reader unwittingly investing attention in writing with no human thought on the other end. He believed people should know what they are getting for their attention, and Substack was introducing, in collaboration with Pangram, an AI scanner which any reader can use on any post over 100 words, to see, for instance on this issue, how much was written by me, a very organic person, and how much was contributed by a machine. (We always disclose, here and on LinkedIn, that our posts are made with the Tokenando Content Engine and reviewed by our team.)
While attention can be monetised, money still cannot buy taste, but a good amount of it can buy a lot of compute.
We started Tokenando on the realisation that there is an economic layer beneath the technical one, which was not yet visible to all leaders and decision makers. That awareness gap has become a financial gap, and it is actually widening faster than we predicted: from the cost of tokens inside a single company to the financing of data centres and frontier labs across the world.
Collateral Thinking
On August 10, 2026, Nvidia, whose chief executive Jensen Huang has spent the past three years selling the picks and shovels of this new gold rush, signed memorandums of understanding with six large financial institutions (Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR) to establish independent financing platforms, aiming to mobilise more than $500bn of third-party capital for AI infrastructure over time.
The platforms are designed to provide lending against compute, taking the chips as security and the revenue those chips earn as the repayment. It is a mortgage with silicon in place of bricks and mortar. “In AI, compute is revenue,” Mr Huang said in the announcement.
Nvidia is not the lender here; its role is to connect its customers to institutional capital. The buyers are hyperscalers, frontier labs, enterprises and neoclouds, the specialist providers renting out chip capacity. None of them can fund capital spending on this scale from their own balance sheets. Insurance money and pension money can, and it is always looking for somewhere to sit on the long-run. Larry Fink of BlackRock compared the arrangement to the creation of mortgage-backed securities in the 1970s and called it the start of the next era of financial engineering.
Nvidia has also offered to back the value of the chips if they are sold at a loss, up to 25 per cent of a given deal, but the full terms have not been published. Everything remains subject to execution of definitive agreements.
Off the Books
None of this is new. In June 2026, Apollo and Blackstone put together $35bn for Anthropic in one of the largest private credit deals ever done, as we covered at the time in the Tokenando Daily briefs.
The deal, nicknamed Project Big Sky, runs through a shell company set up especially for this purpose. The shell borrowed the money, bought the chips (designed by Google, built by Broadcom) and now leases them to Anthropic, using the rent money to repay the lenders. Broadcom guarantees up to $30bn of the amount, covering the shortfall if Anthropic stops paying and the chips resell for less than the loan balance.
All of that so neither the chips nor the debt appear on Anthropic’s own books, leaving them clean ahead of a potential IPO.
The pattern repeats across the industry, and in each case the company using the chips is neither the company that owns them nor the one that borrowed for them. Blackstone alone did several of these deals this year, while Meta’s Hyperion data centre is 80 per cent owned by Blue Owl funds and the remaining 20 per cent by Meta, which keeps most of the project debt off the company’s accounts.
OpenAI is not absent from the picture either. On August 17, 2026, Nvidia announced it had secured the land, power and shell capacity at the Ohio campus of SB Energy, a SoftBank subsidiary, and would provide credit support on the buildout. SB Energy will own and operate the site, while OpenAI takes it on a 20-year lease, for eight gigawatts of capacity arriving in phases from 2028. Nvidia is also investing $1.5bn in SB Energy, alongside SoftBank and OpenAI.
Something similar is happening in Europe, but with money flowing from different sources. On August 11, 2026, five European companies agreed to buy AI capacity from Mistral in advance: ASML, Amadeus, Capgemini, Caisse des Depots and CMA CGM. Their commitments convert into European Compute Units over roughly five years, with no early exit and no prices mentioned. Mistral currently operates about 77 megawatts of capacity across three sites, roughly a hundredth of the Ohio campus, funded by an $830m loan from seven banks including BNP Paribas, Credit Agricole and HSBC. It is targeting one gigawatt by 2030, which its chief executive Arthur Mensch has estimated will cost around $50bn against roughly $4bn raised so far.
Increasing complexity and opacity in the debt structures used to finance AI could increase risks to financial stability, the Bank of England said in its Financial Stability Report of July 2026, pointing to off-balance-sheet arrangements, securitised data centre structures and special purpose vehicles. The Bank judged the immediate risk contained, because the total stock of AI debt is still small, but said that is rapidly changing. The Bank for International Settlements said the terms of these deals are typically poorly disclosed, with a risk of the same asset being pledged more than once.
Routing Fees
A further, and perhaps simpler, embodiment of “compute is revenue” is a young New York company called OpenRouter, led by former OpenSea co-founder Alex Atallah, whose platform allows companies using AI to access more than 400 models from over 80 providers. Clients ranging from Nvidia to Zoom and Lovable send their AI requests through a single connection, and OpenRouter decides which model should handle it, based on the task, the price, the speed and the reliability, then routes it there.
OpenRouter raised $113m in May this year, in a Series B led by CapitalG, Alphabet’s growth fund, with participation from Nvidia’s venture arm NVentures alongside existing investors Andreessen Horowitz and Menlo Ventures, at a reported valuation of $1.3bn. This week it is being acquired by Stripe, the online payments platform, at a price estimated to exceed $7bn. That is more than five times its valuation of three months ago, and, based on the roughly $50m of annualised revenue estimated by research firm Sacra earlier this year, a revenue multiple of close to 140. Mr Atallah had described his company as Stripe for AI long before any of this.
Patrick Collison, Stripe’s chief executive, put the rationale on the premise that tokens are the central currency for companies building with AI, and that the economic potential depends on making good use of scarce compute. Stripe now spans both sides of the profitability equation, online revenue and compute spend, having already moved in that direction with a product called Token Billing.
It also gives Stripe a unique vantage point, as every request leaves a record of which model won and at what price. The volumes are considerable: weekly traffic of 25 trillion tokens, up from 5 trillion six months ago, with OpenRouter expecting to process more than a quadrillion this year across 8 million developers. Menlo Ventures, one of its investors, puts that at 15 to 30 per cent of Google’s token volume and 20 to 40 per cent of OpenAI’s.
The Singularity will not be Listed
On the same day it announced the OpenRouter acquisition, Stripe made a couple of rather larger claims in a letter obtained by Axios and signed by Mr Collison, his brother and co-founder John Collison, and William Gaybrick, president of technology and business.
Stripe told its investors that the “singularity” began on January 1 this year and that it has been operating on that basis ever since, borrowing a word that usually describes machine intelligence overtaking our own, and progress running beyond human prediction, but using it to mean a change in the rate at which people start companies. The company reports that this has accelerated its core business: first-half revenue up 41 per cent, free cash flow up 43 per cent, and 88 per cent of the Forbes AI 50, including OpenAI and Anthropic, now running on Stripe. The letter did not end there. “Being private”, it says, is “a growing advantage as we venture into the vicissitudes of the singularity”.
Every financial arrangement in this issue rests on the same bet: that a chip, a building or a five-year block of capacity will be worth something in ten years. Chipmakers are guaranteeing part of it. American frontier labs and European blue chips are committing for years, in some cases for decades. While nobody has published a price for any of it, this is set to change in the coming months: when the first of Nvidia’s platforms turns an MoU into a signed agreement with a credit rating attached, and when Anthropic and OpenAI file their IPO prospectuses. That will show what these assets are assumed to be worth, and this is the kind of intelligence we built Tokenando for.
Wirelessly yours,
Ziad Matar
Co-founder, Tokenando
Editor-in-chief, The Compute
After Hours
We hopped over to Madrid for a quick trip last week, which everyone tells you to avoid in August, but the temperatures were more than compensated by what the city itself had to offer.
Our evening started with two sunsets within the hour, as Madrid was just outside the path of totality of the solar eclipse, and ended in Charrua, a Uruguayan restaurant, where the side of wood-fired charred vegetables totally eclipsed the actual steak they came with. Who said that money could not buy taste?
Back in Barcelona, The Odyssey was the main event of last weekend, while at home we managed to catch up on Chernobyl, a very well crafted drama about the 1986 disaster, which was very hard to watch, but equally important to listen to its companion podcast.
On a lighter, but equally human note, we finished the fifth and final season of The Bear, which was way better than the fourth. Currently on our rotation are the latest seasons of Silo and Ted Lasso, with The Agency and Unfamiliar as new arrivals.
My AirPods played lots of Mumford & Sons this week, from The Road to Red Rocks, their live concert, I took a tangent to Africa with “Wona“, their 2016 collaboration with Baaba Maal from Senegal, Beatenberg and The Very Best, recorded in Johannesburg. Rock and West African percussion do blend well as it happens.
References
“How can I detect AI on Substack?”, Substack support documentation, updated July 21, 2026. https://support.substack.com/hc/en-us/articles/50891130623508-How-can-I-detect-AI-on-Substack
“LinkedIn adds a button to report AI-generated ‘slop’”, TechCrunch, July 30, 2026. https://techcrunch.com/2026/07/30/linkedin-adds-a-button-to-report-ai-generated-slop/
“NVIDIA Partners With Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to Establish AI Compute Infrastructure Financing Platforms”, NVIDIA Newsroom, August 10, 2026. https://nvidianews.nvidia.com/news/nvidia-partners-with-apollo-blackrock-blackstone-brookfield-goldman-sachs-and-kkr-to-establish-ai-compute-infrastructure-financing-platforms-to-mobilize-over-500-billion-of-third-party-capital
“Nvidia lines up $500 billion in financing as CEO Jensen Huang tells CNBC his chips are ‘investable asset’”, CNBC, August 10, 2026. https://www.cnbc.com/2026/08/10/nvidia-wall-street-asset-managers-500-billion-ai-push.html
“NVIDIA Guarantees SB Energy’s PORTS-Pike Technology Campus in Ohio to Exclusively Host NVIDIA AI Compute”, NVIDIA Newsroom, August 17, 2026. https://nvidianews.nvidia.com/news/nvidia-guarantees-sb-energy-s-ports-pike-technology-campus-in-ohio-to-exclusively-host-nvidia-ai-compute
“Inside the $35bn deal: Apollo and Blackstone’s chip-backed SPV for Anthropic signals a new financing era”, Capacity, June 10, 2026. https://capacityglobal.com/news/anthropic-blackstone-apollo-35bn-ai-infrastructure-spv/
“Bond Investors Push Back As AI Debt Heads Toward $570 Billion”, Forbes, July 17, 2026. https://www.forbes.com/sites/robertszczerba/2026/07/17/bond-investors-push-back-as-ai-debt-heads-toward-570-billion/
“ASML, Amadeus and others commit to backing Mistral’s data centre buildout”, Sifted, August 14, 2026. https://sifted.eu/articles/mistral-enterprise-customers-data-centre-buildout/
Financial Stability Report, July 2026, Bank of England. https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026
“OpenRouter Raises $113M Series B”, OpenRouter, May 28, 2026. https://openrouter.ai/blog/announcements/series-b/
“OpenRouter Now Processes More Than a Quadrillion Tokens a Year”, Menlo Ventures, May 26, 2026. https://menlovc.com/perspective/openrouter-now-processes-more-than-a-quadrillion-tokens-a-year/
“Stripe agrees to acquire OpenRouter to help businesses optimize token routing and usage”, Stripe Newsroom, August 19, 2026. https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter
OpenRouter revenue, valuation and funding, Sacra. https://sacra.com/c/openrouter/
“Scoop: Stripe says ‘the singularity’ has begun”, Axios, August 19, 2026. https://www.axios.com/2026/08/19/stripe-payments-openrouter-singularity
“Against Claudefishing”, Chris Best, The Substack Post, July 21, 2026.
