- Citadel Securities warns that AI token spend is slowing as enterprises confront ballooning compute bills and shrinking returns.
- Microsoft, Amazon, Meta, and Uber have all pulled back on AI token spend, cancelling tools and leaderboards over runaway costs.
- SpaceX is set to open trading at roughly $1.77 trillion, with Anthropic and OpenAI also preparing IPO filings in the same window.
- Tether led a $1.4 billion round for humanoid robotics firm NEURA, signalling stablecoin profits are quietly funding the next AI wave.
- Citadel Securities warns that AI token spend is slowing as enterprises confront ballooning compute bills and shrinking returns.
- Microsoft, Amazon, Meta, and Uber have all pulled back on AI token spend, cancelling tools and leaderboards over runaway costs.
- SpaceX is set to open trading at roughly $1.77 trillion, with Anthropic and OpenAI also preparing IPO filings in the same window.
- Tether led a $1.4 billion round for humanoid robotics firm NEURA, signalling stablecoin profits are quietly funding the next AI wave.
Table of Contents
AI Token Spend Is Hitting a Wall — and Citadel Is Calling It
AI token spend — the rolling cost of every prompt, completion, and code suggestion that enterprise teams push through large language models — has become the defining line item that CFOs can’t ignore anymore. This month, Citadel Securities dropped a macro note arguing that AI adoption is decelerating, and not because the technology isn’t impressive. The slowdown is about cost discipline catching up with enthusiasm.
The evidence is hard to dismiss. In the past few weeks alone, a pattern has emerged across some of the biggest names in tech. Amazon shut down an internal tool called KiroRank — a token leaderboard that was supposed to gamify developer productivity — after engineers figured out how to game the metrics rather than ship better software. Microsoft cancelled access to Anthropic’s Claude Code for roughly 5,000 employees after AI token spend ran well past budget. Meta had already scrapped a similar internal leaderboard back in April. And Uber’s CTO made the most candid admission of the bunch: the company burned through its entire 2026 AI coding budget in just four months.
That last one deserves a second read. Four months. Not a year, not two quarters — four months. It suggests that when you hand engineering teams open access to frontier models without guardrails, AI token spend scales faster than anyone’s spreadsheet anticipated. Uber isn’t a small startup testing the waters. It’s a global platform company with serious engineering resources, and it still ran out of runway almost before the year got started.
The pattern here isn’t a string of individual failures. It’s enterprises collectively discovering that AI productivity gains don’t automatically justify uncapped token consumption. The question that’s replacing ‘how do we get our teams using AI?’ is now ‘what did we actually ship with all that AI token spend?’
The IPO Slate That Turns This Into a Market Story
Any other week, a hedge fund note about slowing AI token spend would be a footnote in a macro briefing. But this week, it lands directly in front of what’s shaping up to be the richest IPO window in recent memory — and that changes the stakes considerably.
SpaceX begins trading today after close. The expected valuation sits around $1.77 trillion, a number that’s not purely a rocket story. SpaceX absorbed Elon Musk’s xAI compute business back in February, weaving AI infrastructure into its core narrative. That means Friday’s open is partly a market vote on whether investors still believe AI demand compounds indefinitely.
The timing gets more charged when you add the rest of the queue. Anthropic filed a confidential S-1 on June 1st, just weeks after closing a funding round that valued the company near $965 billion. That valuation is almost entirely a bet on enterprise adoption — the same enterprise adoption that just produced a high-profile Microsoft walkaway over runaway AI token spend. OpenAI is reportedly preparing its own filing. Together, these three listings represent a concentrated moment where the market has to put an actual number on the AI trade, not just a narrative.
Citadel’s thesis cuts right through the optimism: frontier AI will consolidate among a handful of companies with the balance sheets to absorb compute costs at scale. Google, Microsoft, Amazon, Meta — they can eat this. The mid-tier enterprise that was hoping to build competitive advantage on top of rented AI? Probably not. That bifurcation could mean the addressable market for AI services is smaller and more concentrated than the bullish IPO valuations assume.
Tether Bets $1.4 Billion on Humanoid Robots — and That’s Not a Distraction
While the AI token spend debate plays out in Silicon Valley boardrooms, Tether — the company behind the world’s most widely used stablecoin — just placed a landmark bet on what it clearly sees as the next wave. Tether led a $1.4 billion Series C for NEURA Robotics, a German humanoid robotics company, with Nvidia and Amazon also participating in the round.
The deal includes embedding edge AI and crypto payment tools directly into NEURA’s machines. That’s not a PR flourish — it sketches an architecture where a robot can carry its own digital wallet, make autonomous purchasing decisions, and settle transactions without a human in the loop. Pair that with the Mastercard story below and it stops looking like a concept.
What makes Tether’s involvement particularly worth watching is what it says about where stablecoin profits are flowing. Tether has spent the past year converting USDT earnings into stakes across compute infrastructure, energy assets, and media properties. Humanoid robotics is the latest addition to what’s quietly becoming a pretty serious AI-era investment portfolio. The company sitting next to Nvidia and Amazon at a $1.4 billion table isn’t a crypto sideshow anymore.
For anyone trying to identify the next major AI trade after the LLM infrastructure buildout, robotics is the clearest candidate right now. Physical AI — models embedded in machines that interact with the real world — represents a category where hardware moats matter again, and where the compute cost problem is fundamentally different from cloud-based AI token spend on enterprise tools.
Mastercard Brings AI Token Spend Into the Physical Economy
Mastercard unveiled something it’s calling Agent Pay for Machines this week: a system designed to let AI agents autonomously purchase services and settle payments using cards, bank accounts, or stablecoins. The stablecoin infrastructure was built in partnership with Coinbase, Ripple, and Solana. That’s a card network processing trillions of dollars annually treating crypto rails as equal to traditional payment methods for machine-to-machine commerce.
The timing is notable because MetaMask shipped a wallet built specifically for AI agents just days earlier. So you now have a wallet layer and a settlement layer forming simultaneously, in real time, for a commerce category that barely existed as a concept twelve months ago. Autonomous agents need money that moves at software speed — and as AI token spend migrates from the cloud into physical and agentic systems, stablecoins settle in seconds, around the clock, without banking hours or correspondent relationships slowing things down.
The broader implication is that stablecoins are positioning themselves as the default payment infrastructure for the agent economy. Not through a regulatory argument or a white paper, but through live integrations with Mastercard, Coinbase, and Ripple. That’s a meaningful shift in legitimacy.
Solana at the Poker Table, Crypto Buying Into Culture
The Solana Foundation announced a sponsorship deal with the World Series of Poker this week. Players can now buy into tournaments and collect winnings in SOL or stablecoins, with MoonPay handling the on-ramp. Solana branding will be visible at events throughout the series.
It’s a natural fit — poker and crypto have overlapped since the online poker boom of the mid-2000s, when a generation of players learned to move money across borders fast and cheaply. The WSOP gives Solana front-row placement in front of a global, financially engaged audience that already understands digital assets at a gut level.
The sponsorship landed a day after Kraken announced a World Cup deal, which means crypto is on a deliberate mainstream visibility push right now. The working theory is that brand exposure today converts to users and liquidity later, even if SOL’s price action has been soft. The Solana Foundation clearly isn’t managing to this week’s candle — it’s building for a longer adoption cycle.
What the SpaceX Open Actually Tells Us
Friday’s SpaceX debut isn’t just a single stock event. Given everything stacked around it — the Citadel warning on AI token spend, Anthropic’s S-1 in the pipeline, the NEURA raise, the Mastercard announcement — it functions as a real-time stress test for how much faith institutional capital still has in the AI growth thesis at current valuations.
A strong open says the market is willing to price in compounding AI demand despite the enterprise pullback signals. A weak one suggests the cracks Citadel is flagging have started showing up in the numbers that actually move money. Either way, the IPO window that follows — Anthropic, OpenAI, and whatever else files in the next few months — will be shaped by what happens at the SpaceX open. The AI trade is entering its accountability phase, and the bills are coming due on both sides of the ledger.
Source: Decrypt
Frequently Asked Questions
Why is AI token spend declining at major enterprises?
Companies including Microsoft, Amazon, Meta, and Uber have hit hard budget ceilings after treating token consumption as a vanity metric rather than a business outcome. Once the bills arrived — Uber burned its entire 2026 AI coding budget in four months — finance teams started asking what the spend actually shipped.
What does Citadel’s macro note say about the future of frontier AI?
Citadel Securities argues that frontier AI will concentrate among a small number of firms with large enough balance sheets to absorb compute costs. Smaller players and mid-sized enterprises are likely to cut back, creating a potentially shaky transition period for the broader market.
How does the SpaceX IPO connect to the AI trade?
SpaceX absorbed the xAI compute business in February, so its expected $1.77 trillion valuation is partly a bet on AI infrastructure demand. Friday’s opening price will act as a sentiment read on how confident markets remain in the long-term AI growth thesis.
What is Mastercard’s Agent Pay for Machines?
Agent Pay for Machines is a Mastercard system that lets AI agents autonomously buy services and settle payments using cards, bank accounts, or stablecoins. It was built with crypto firms including Coinbase, Ripple, and Solana, and represents a major institutional validation of crypto rails for machine-to-machine commerce.
