Tech Brew Ride Home - Is It Ironic If Amazon Is Shredding Books?

Episode Date: August 17, 2026

Stripe locked in a $7B+ deal for OpenRouter, and OpenAI leased a 10GW Ohio campus with Nvidia backstopping it. The Journal found $3T in off-balance-sheet AI commitments, an AirTag caught Amazon shredding rare books, and Anthropic detailed Claude's invisible watermarks. Links Sources: Stripe has finalized a deal to acquire AI model marketplace OpenRouter for more than $7B; founded in 2023, OpenRouter had a $1.3B valuation in May (Bloomberg) OpenAI signs a 20-year, 10GW data center deal in Ohio with SoftBank's SB Energy; Nvidia agrees to backstop a portion of the value of the completed data center (The Wall Street Journal) Analysis: nine top tech companies including Alphabet and Meta had ~$3T of AI-related off-balance-sheet commitments, far exceeding their $600B in reported capex (The Wall Street Journal) Anthropic explains how Claude's invisible text watermarks will work (The Verge) Investigation: Amazon is buying huge quantities of rare books, scanning them for AI, and destroying them; a tracked Biblio order went to its Las Vegas facility (The Decoder) Used bookstores are seeing a sales resurgence driven not by readers but by AI companies placing bulk orders in the thousands, with the books destructively scanned, then pulped (Fast Company) AI film startups are setting up studios in Hollywood, using US and Chinese AI models while touting lower production costs and a way around traditional financing (The Guardian) Subscribe to the ad-free feed.

Welcome to the Tech Brew Ride Home for Monday, August 17th, 2026. I'm Brian McCullough. Today has Striped locked in a deal for OpenRouter. OpenAI has leased a 10-gigawatt Ohio campus with Nvidia backstopping it. The journal found $3 trillion on off-balance sheet AI commitments, an air tag caught Amazon shredding rare books and anthropropic detailed clods and visible watermarks. Here's what you miss today in the world of tech. IT pros can't always be about fun and games until there's a fun game to actually play. Play today's sponsor Eaton created an online card-based game called Ace of Uptime,

and the whole idea behind it is to give IT pros a sandbox. You can practice problem-solving and decision-making and have some fun in the process. Each level of the game, you go up against a familiar foe like old processes or silent system failures, your Delta scenario, you choose your solution, and then see how your decisions play out. Will it resolve the problem? Will it escalate it faster than expected? start playing and find out. Head over to Eton.com slash ace to play Ace of Uptime. That's Eaton.com slash ace. Bloomberg is reporting that Stripe has finalized a deal to acquire AI

Model Marketplace, OpenRouter for more than $7 billion. Founded in 2023, OpenRouter had a mere $1.3 billion valuation just back in May. Quote, OpenRourter provides access to hundreds of AI models with the goal of matching developers with the most efficient and affordable options for the job at hand. The New York-based company has attracted some of the biggest investors in Silicon Valley, including Capital G, one of Alphabet's Venture arms, as well as Andreessen Horowitz and Menlo Ventures. OpenRouter has raised more than $150 million in capital to date. The startups rise coincides with greater scrutiny on AI costs, while firms like Anthropic and Open AI are still widely viewed as offering the most capable AI models. A long list of Chinese firms provide cheaper alternatives

that are often viewed as good enough for many tasks. In May, OpenRouter said it serves 8 million developers who rely on it to access more than 400 different AI models. The startup's main growth is coming from developers who experiment with different models when building agenetic capabilities into their software, a process that requires a mix of infrastructure that can work across different providers and data sources. OpenRourter also offers services that help companies access backups in case the model they use fails and understand which options are most popular across the broader tech ecosystem.

system. OpenRouter Chief Executive Officer Alex Atala previously co-founded OpenC, a non-fungible token marketplace, which raised more than $400 million in capital but saw usage crater. Atala stepped down from OpenC in July 2022, and less than a year later, started OpenRouter. Earlier this year, Atala described OpenRouter as the AI equivalent of Stripe, end quote. OpenAI has signed a 20-year 10-gawatt data center deal in Ohio with SoftBanks's SB Energy. Invida has also agreed to backstop a portion of the value of the completed data center, quoting the journal. In return for its commitment, Nvidia will be the exclusive chip provider of the first half of the Ohio site and take an equity stake in SB Energy, the people said.

NVIDIA has the option to expand its commitment later on to the full campus. Open AI's 20-year lease, parts of which were announced Monday, is the latest complex transaction aimed at helping an AI firm secure an enormous amount of computing capacity. Chip firms, including Nvidia and Google, have increasingly agreed to use their robust balance sheets to help startups such as OpenAI and Anthropic sign leases that developers can raise debt against while trying to manage their own risk. The Southern Ohio campus is partly located on a former uranium enrichment site owned by the U.S. Department of Energy and is expected to generate enough energy to power 7 million homes. Commerce Secretary Howard Lutnik and Energy Secretary Chris Wright have been deeply involved in the project.

NVIDIA's support is intended to make the project easier and less expensive to finance by giving lenders greater confidence that the data centers would retain value, even if OpenAI were no longer the tenant, the people with knowledge of the deal said. NVIDIA's commitment is structured to require a series of mitigation steps before the chipmaker would be obligated to make any payments. For instance, if OpenAI were to walk away from the project, SB Energy would first try to lease the site to another customer for the same price. Some of the people said, if SB Energy wasn't able to find another suit.

tenant, the firm would try to sell the site, and Nvidia would pay any difference in the value up to $105 billion if the initial phase is completed. By backing the asset value of the data center, not opening eyes ongoing lease payments, the structure limits Nvidia's risk exposure substantially those people familiar with the deal said. The chip giant's guarantee covers completed data centers, not facilities under construction. Invita is expected to reap commercial benefits from the project in the near term, though. The company plans to sell hundreds of billions of dollars worth of chips into the initial phase of the Ohio facility, people with knowledge of the deal said. It has discussed providing financing for OpenAI's purchase of its chips for the campus, for example.

As part of the arrangement, Nvidia said it agreed to make a $1.5 billion equity investment in SB Energy. SB Energy, which is majority owned by Japan's soft bank and counts OpenAI as an investor, is working with bankers on an IPO as soon as next month. People familiar with that plan, deal said, aiming to raise between $5 and $7 billion. Nvidia initially planned to guarantee all 10 gigawatts of OpenAIs data center commitments up front, which could have totaled around $250 billion of Wall Street Journal reported. The Data Center campus will rely on the construction of a 9.2 gigawatt natural gas plant, which is owned by the U.S. government and financed by Japan under a recent trade deal,

end quote. Well, given that, how about this? An analysis has found that nine top tech companies, including Alphabet and Meta, had around $3 trillion of AI-related off-balance sheet commitments far exceeding their $600 billion in reported CAP-X spending. Quoting the journal, those obligations are growing faster than traditional CAP-X, which totaled only about $600 billion over the past year they reported and were about triple what the companies owe under their outstanding leases and long-term borrowings.

Meta's gigantic Hyperion Data Center project in Louisiana, which is the size of about 1,700 football fields helps explain how big obligations wind up off tech companies' balance sheets. Meta initially agreed to lease Hyperion for a four-year term starting in 2029 with options to renew for up to 20 years. It guaranteed that it would make bondholders whole if it doesn't stay the entire two decades. The company doesn't think payments under that guarantee are probable, so it hasn't recorded any liability on its balance sheet. In accordance with accounting rules, meta's Hyperion lease obligations will remain off-balance sheet. sheet until it starts paying rent. It said its aggregate initial lease commitment is about $12.3 billion.

Meta disclosed $347 billion in total obligations for leases that haven't kicked in yet, including for Hyperion as of June. Across the companies, the journal analyzed promises of payments under those uncommenced leases totaled $1.2 trillion in off-balance sheet obligations or about four times more than what was disclosed a year earlier. In addition to meta, the journal reviewed commitments from Alphabet, Amazon, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, and advanced microdevices. Data centers get stuffed with a lot of hardware, including the Nvidia chips that are used to train and run models and memory chips that store information. To buy all that, companies sign long-term contractual agreements well in advance to lock in production from their

suppliers. Those and other purchase obligations at the companies, the journal examines, stand at a whopping $1.9 trillion. Under accounting rules, purchase commitments typically remain off-balance sheet until a product or service is delivered. Alphabet's purchase commitments and contractual obligations have exploded and stood at $811 billion as of June 30th. As with other companies, it is hard to tell from its disclosures what precisely it intends to buy. The company said the commitments primarily relate to technical infrastructure and inventory and agreements to secure energy for data center usage. Alphabet also didn't detail why those obligations increased so much from the $332 billion that reported three months earlier.

The commitments span several years with obligations under its energy agreements lasting as far out as 2054. Off-balance sheet exposures at some companies include agreements to buy other companies' stock in the future or backstop leases for other tenants. Invita committed to make $27 billion in equity investments between April 26 and the end of fiscal year January 2027. Alphabet and Amazon recently posted results showing negative-free cash flow, meaning their capital spending exceeded the cash they brought in from operating their businesses, end quote. AI threats are like a hydra that learned how to type, cut one head down, and two more pop right back up, which means the old way we've been handling cybersecurity just won't cut it

anymore. That's where Rubrik comes in. They've engineered a platform that secures your data, controls your AI, and protects your identity. See how Rubrik helps you stay ahead of threats at rubric.com slash mb. That's rubr-ir-i-k.com slash mb. Is your multi-entity management creating more confusion than clarity you need the Intuit ERP, Intuit Enterprise Suite? It's the AI-Native ERP solution that's powerful, painless, and proven. Learn more at Intuit.com slash ERP.

Anthropic has outlined exactly how that invisible watermarking of AI-generated text will work. Quoting the verge. On Friday, Anthropic announced that Claude's text marking system is a version of the synth ID text approach in open source watermarking technology developed by Google DeepMind that creates detectable patterns using wording probabilities. This watermarking feature alongside C2PA support for Claude processed images is being introduced to meet Anthropics obligations under the European Union's AI Act, which requires synthetic audio, image, video, and text to include machine-readable marks that enable the content to be detected as artificially generated or manipulated. Anthropics says the text watermarks won't make Claude more expensive for users or, quote,

have any practical impact on the quality or content of Claude's outputs. Here's Anthropics' explanation for how it works. Take the sentence, the weather today was cold and dot, dot, dot. The next word is very unlikely to be sugary, but it is quite likely to be overcast or gray. Under most circumstances, it doesn't matter much to the reader, which of these two words the model ultimately chooses. The meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number. Watermarking uses low-stakes choices like these, which occur many times over a piece of generated text,

to leave a pattern in Claude's responses. That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it. When watermarking is used, choices are still made at random, but the source of the randomness is different. Instead of using an arbitrary random number generator to pick the next word, watermarking uses the key and a few words that come before to settle what word the model should pick after, end quote. As Anthropic notes, the EU's AI transparency requirements also impact other major AI developers, so Claude won't be the only model introducing text watermarks. Google's Gemini chatbot has supported the Synth ID text solution since 2024, and while, While Open AI hasn't detailed any text watermarking plans for chat GPT in its AI Act compliance roadmap, it will also be subject to the law's requirements, end quote.

I've been hearing rumblings about this from a bunch of different places recently, but rumors are the big AI companies and labs might be buying up all the used books they can find and then destroying them, quoting the decoder. Amazon buys printed books in bulk, scans them for AI training, and destroys them in the process. That's what an investigation by 404 media found after placing an air tag inside a shipment of rare books and tracking it across the country. The shipment ended up at an Amazon warehouse in Las Vegas home to a team called VGT3, whose logo features a Tyrannosaurus Rex holding a book. Workers there say they cut off book spines to speed up scanning, destroying the copies in the process. Amazon uses the scan data to train its Nova models.

A spokesperson said the company buys books through commercial channels to improve its products. booksellers suspect that AI companies are trying to systematically scan every book by ISBN number. Printed texts are especially valuable because they often don't exist online and predate 2022, meaning they are free of AI generated content. Amazon isn't alone here. Anthropic ran a similar operation. A lawsuit by book authors revealed its Project Panama, where Anthropic bought books on marketplaces, cut off their spines, and digitize them. The judge in that case ruled that scanning qualified as fair use and didn't violate copyright, because the printed originals were destroyed and therefore not copied and resold, end quote.

And quoting Fast Company, not too many years ago before the dawn of Kindles used bookstores were a refuge for budget-minded, voracious readers. As the industry evolved, many suffered. Now, they're seeing a resurgence in sales, but the customers increasingly aren't bookworms, their AI companies. Barter Books, for instance, typically sells two to three thousand books per week, which isn't bad for a used bookstore in Northumberland, the least densely populated county in England. Recently, however, the owner got a bulk order from a company in Canada for that many books in a single day. Several other bookshops in the area are seeing similar demand, and while they can't definitively say what the books are being used for, a recent court ruling offers a

clue. The fate of those books may also be a dire one. Destructive scanning physically disassembles bound books so their individual pages can be fed through high-speed scanners and ingested by a large language model. Once the scanning process is complete, the pages are discarded, recycled, or pult. While it may not be an especially big deal of a copy of a mass market paperback gets destroyed, something that happens all the time, book enthusiasts worry that rare books, which are often sold by used bookstores to collectors, could be swept up in the process as well. Anthropic tells the BBC that's not the case, saying none of our data acquisition programs buy and destroy rare or antiquarian books in a statement. It's unlikely, however, that Anthropic

is the only AI company using this method to train its systems, and other companies may not have the same limitations on what they destructively scan. Antique and rare books, after all, often contain information that other books don't, which could make them particularly valuable as training material, end quote. Finally, The Guardian says AI film startups are increasingly setting up studios in Hollywood using U.S. and Chinese AI models while touting lower production costs as a way around traditional financing and other Hollywood studios are using them just on the down low. quote, imagine the 41-year-old studio run by Howard and Brian Grazer now has an AI studio Obsidian operating from a space in its Beverly Hills HQ. It has partnered with Obsidian to make

the code an AI-enabled Howard directed animated feature based on drawings made in Vietnamese captivity by U.S. Prisoners of War. D. Ryan Reeb, an executive producer at Obsidian, estimated the technology delivered a 30 to 40 percent cost savings. He said he had met with dozens of top production companies interested in the technology and predicted, all would eventually use it. But AI in Hollywood has drawn comparisons to plastic surgery. Everyone's doing it, but nobody wants to talk about it. It's 100% like that, Reeb said. An AI native filmmakers such as Cardoza, who makes films in his spare bedroom and Playa Vista, has no reason to downplay his AI use. It has helped him reach 28 million YouTube views for his Chronicles of Bone Fantasy Universe series

launched online last autumn, featuring his versions of Robin Hood, Peter Pan, King Arthur, and Tinkerbell, and a lucrative deal with the Magnific AI Platform. Cardoza believes AI filmmakers are attracting audiences because, ironically, their direct authorial voice has greater authenticity than in Hollywood where, quote, they turn writing into algorithms. Cardoza highly rates the models made by Chinese tech companies who are smoking everybody on everything, the image generation models, the video generation models, he says. He assumes the Chinese models are so powerful because they have been trained on everything under the sun in a way that U.S. models cannot because of

intellectual property restrictions. But with Washington, considering restrictions on Chinese open-source AI models, Cardoza opposes a ban on such powerful tools, which have allowed him a creative flourishing. I'll move to China, he said. I'm not even lying, end quote. Nothing more for you today. Talk to you tomorrow.