Anthropic, Cloudflare, Elon's Terafab, and the Agentic Web

The connective thread this week is not any single product. It is the build-out of infrastructure that assumes AI agents are permanent economic actors, not temporary features. A $10 billion compute deal with a months-old startup, a Cloudflare wallet that lets AI pay its own bills in stablecoins, and a factory the size of a small city are not three separate stories. They are three corners of the same foundation being poured under a machine-driven economy. The labs are not just building smarter models. They are building the financial, physical, and computational substrate those models need to operate without human handholding on every transaction.

My bet is that within 18 months, the ability to run autonomous agents at scale, with their own compute budgets and payment rails, becomes a harder competitive moat than model quality alone. The model gap between frontier players is narrowing; the infrastructure gap is widening. Also below: why Tesla's data pipeline may matter more than any benchmark, why Apple's fitness angle for smart glasses is smarter than it sounds, and what Cloudflare's 57-percent-bots statistic tells you about who the internet is actually being built for now.

@KobeissiLetter broke the news: Anthropic has signed a 6-year, $10 billion computing deal with Volta, a months-old infrastructure startup backed by Nvidia, with Bitcoin miner Bitdeer as a delivery partner, confirmed by Bloomberg. Separately, a16z announced co-leading Volta's Series A, framing the company as a fix for the compute inequality that leaves startups paying premium prices for inflexible GPU access while hyperscalers lock in sweetheart deals.

The Anthropic deal is the demand signal; the a16z raise is the supply-side response. If you are building an AI product and your compute costs feel punishing, Volta is worth watching closely.

Cloudflare has launched stablecoin wallets that let AI agents pay for APIs and web content autonomously, no human approval required per transaction, as @coinbureau reported citing Cloudflare's own announcement. The company's chief strategy officer noted that 57 percent of all web traffic is already bots. The financial rails of the agentic web are not a future design question; they are being laid now, by one of the companies that already handles a significant portion of the internet's infrastructure.

The Verge flagged the obvious implication: if your product monetizes through APIs or paywalled content, your next customer cohort may not be human.

Jensen Huang named Elon Musk as his pick to win the AI race, and @stark0xbt's thread nailed the reasoning: it is not about model benchmarks, it is about Tesla's fleet of millions of vehicles functioning as a distributed real-world sensor network running continuously, feeding a training pipeline that no lab can replicate by writing checks alone, per CNBC.

The Dojo supercomputer is the processing layer on top of that collection advantage. What your product does with real-world behavioral data at scale matters more than model quality on a leaderboard, and that gap is only going to widen.

@rand_longevity's blunt post put it plainly: the days of humans fact-checking AI outputs in real time are over. The problem is not theoretical: it is live in medicine, law, finance, and code today. If you are shipping AI-assisted outputs to customers in any regulated domain, the liability question is no longer hypothetical and you need an answer before your users find out you do not have one.

Eleven Labs (sponsored)

The best voice models, now with full orchestration. Build real-time voice and chat agents on one low-latency stack: any LLM, your tools and knowledge, testing, Guardrails, and omnichannel deployment.

Start building free

When AI search engines began citing Reddit as a trusted source of genuine human opinion, brands noticed and started planting fake recommendations across major subreddits, as The Verge reported. Moderators are now playing whack-a-mole with astroturf campaigns, documented in detail by Bloomberg and Wired.

The feedback loop is the problem: AI elevates human conversation as a signal, humans corrupt the signal, and AI trains on the corruption. If your product depends on community-sourced trust signals showing up accurately in AI-generated answers, that signal is already being gamed and you should assume the degradation is worse than what moderators are catching.

Jensen Huang announced Alpamayo 2 Super, an open reasoning model built specifically for autonomous vehicles, with an explicit design goal of reasoning through ambiguity before acting rather than just reacting to sensor input. The distinction matters: most current AV systems are reactive pipelines; a model that deliberates before committing to a decision is a different architecture entirely.

NVIDIA is positioning itself as the cognitive layer, not just the chip supplier, for the next generation of robotaxis, trucks, and delivery vehicles. Competing AV developers who assumed NVIDIA was infrastructure-only should rethink that assumption now.

Elon's planned Terafab facility, detailed by @XFreeze and confirmed by Reuters reporting on the Foxconn partnership, is planned at 100 million square feet, roughly 10 times the size of Tesla's Giga Texas and covering approximately 2,300 acres. Tom's Hardware notes this is less a factory expansion than a rethinking of what manufacturing infrastructure means at AI-scale demand.

When the compute arms race requires a city-sized building to keep pace, the supply chain assumptions underneath every AI product on the market are due for a rewrite.

a16z's post highlighted Base Power, founded in 2023 by Justin Lopas and Zach Dell, which installs home battery systems that collectively form a distributed energy network, turning neighborhoods into resilient mini-grids rather than depending on centralized infrastructure. The company raised $200 million to deploy across Texas, where grid fragility is not an abstract concern.

The bet is that decentralization solves what top-down infrastructure investment has consistently failed to fix. For anyone building hardware or software that depends on reliable residential power, the distributed grid model is the infrastructure trend to track.

Apple is reportedly positioning fitness tracking as the anchor use case for its upcoming smart glasses, according to @RtoVR citing Road to VR. The logic is sound: fitness gives people a daily, concrete reason to wear something on their face, unlike AR overlays or AI assistants that require a use case to present itself.

If the glasses can replace an Apple Watch and AirPods simultaneously while tracking a run, the value proposition becomes legible to a mass-market buyer in a single sentence. Watch whether Apple prices these to replace two existing devices, because that framing is what turns a niche product into a platform.

The owners of MySpace told Dexerto they plan to relaunch the platform, adding "if that one doesn't work, we'll do it again," which is either admirable persistence or a complete absence of strategy. MySpace peaked at 100 million users before Facebook dismantled it, and the brand has been a cultural footnote for over a decade. Nostalgia is not a product-market fit.

The one scenario where this is not a waste of time is if they find a specific underserved community, music discovery being the obvious candidate, and build for that rather than trying to resurrect a general social network.

The counterargument worth sitting with is that infrastructure moats have a history of being less durable than they look at the moment of construction. The companies pouring the most concrete and signing the largest compute contracts are betting that raw capacity translates into lasting advantage. It might not. The history of platform transitions suggests that the layer which captures value shifts as the technology matures: first it is hardware, then infrastructure, then software, then application. Anthropic locking in ten billion dollars of compute and Elon building a city-sized factory could both look prescient in three years, or they could look like the peak of a capital cycle that got ahead of actual demand. That is the version of this story where the bet is wrong.

But here is why I think the bet holds. Cloudflare's 57-percent-bots figure is the tell. The internet is already predominantly machine-to-machine traffic, and that is before autonomous agents have payment rails or persistent compute budgets. The companies building that substrate today, the Voltas, the Cloudflares working on agent wallets, the Base Powers creating resilient power infrastructure, are not building for the current market.

They are building for a machine-client economy that is arriving faster than most product roadmaps acknowledge. If your business model still assumes a human is initiating every meaningful transaction, that assumption is worth stress-testing now, not after the rails are fully laid.

Markets don't wait for Monday. News breaks on a Saturday morning, and most traders can do nothing but watch.

Not on Liquid. Trade domestic and international equities, commodities, forex, crypto, and prediction markets — all from one account, 24 hours a day, 365 days a year. Liquid gives you access to any market, from anywhere, anytime. To us, access is arbitrage.

Getting started takes under 10 minutes: log in with Google, deposit with Apple Pay or a bank transfer, and trade from your phone or desktop — wherever you are in the world.

While everyone else is refreshing headlines and waiting for the open, you're already positioned. That's the difference between reacting to markets and actually trading them.

Trade 24/7 on Liquid Today

Most SaaS finance teams know their billing process is slow.Most SaaS finance teams know their billing process is slow. Few know what it's costing them.

The Tabs Billing Lag Calculator puts a dollar figure on it in two minutes — benchmarked against top SaaS companies.

See your number.