Monday June 1, 2026
Stop Counting Tokens, Start Counting What They Actually Do
A company spent half a billion dollars on AI in a single month and only noticed when the invoice landed. The instinct is to treat that as a cautionary tale about expense controls. The more interesting reading is that it exposes the actual business model underneath the entire frontier AI boom, and the question almost nobody is asking out loud yet is who genuinely needs to be running the most powerful model on the market for the work sitting in front of them.
I caught myself in this exact bind last week. I was happily working in Claude Opus 4.7, and today I am simply defaulted into 4.8. It may well be the better model, but for the prompting and research I actually do, is the upgrade necessary, or am I now burning more to get the same answer I was already perfectly happy with? Multiply this one quiet default across millions of seats and you start to see where the waste really lives.
The waste is staggering once you look closely. When consultants audit how these tokens get burned, the recurring finding is that most users are running a Ferrari to take the trash to the end of the driveway. One audit turned up an agent that had been quietly running for five months and billing tens of thousands of dollars while the employee believed she was paying twenty dollars a month. Basic tasks get routed through the most expensive intelligence available, and staff are nudged to burn more because somebody decided token usage was the proof point that they had finally become a real tech company.
Here is the uncomfortable part for the AI labs. The incentives point in exactly the wrong direction. Anthropic and OpenAI make money when you consume, the apps are designed to help you consume more, and a single agentic workflow can eat a thousand times the tokens of a simple query. None of that is tied to whether the output mattered. It worked beautifully right until the first wave of audits hits a regular cadence and CFOs start pulling back. This pullback will arrive precisely when the AI companies are valued as though revenue only moves up and to the right.
So the metric quietly shifting under everyone is token intelligence density, the work you get per token rather than the score on a benchmark. The real franchise is not the smartest model, because within a short time everyone will have an excellent one. It is whoever helps the enterprise prove the spend was worth it, and cut the waste while they are at it. That is where I would be looking while everyone else stares at the racetrack.
◾ Treasury’s Bessent reiterates CBDC “off the table” (The Block)
◾ CFTC approves crypto perpetual futures (CFTC)
◾ Sen Lummis (R-WY): Clarity Act failure risks Chinese control of financial markets (X)
Government & NGO Actions
◾ ECB’s Schnabel: digital euro best response to USD stablecoins (Bloomberg)
◾ Fed’s Waller: stablecoins could broaden US monetary-policy reach (Bloomberg)
◾ BoE’s Greene: stablecoin demand may soon fade (Reuters)
◾ Japan ruling-party panel pushes yen stablecoins for Asian settlement, crypto ETF legal framework (Reuters)
◾ Vietnam ministry proposes SME digital-asset bank collateral (Vietnam News)
◾ SEC charges Texas man over $12M AI-trading-bot crypto scheme (SEC)
◾ US moves to halt Nvidia $NVDA AI-chip shipments to Chinese firms outside China (Reuters)
◾ Connecticut Gov Lamont signs AI hiring-notice mandate (Bloomberg Law)
◾ New Mexico officials urge AI governance plan for schools (Albuquerque Journal)
◾ Texas shifts $10M bitcoin from ETFs to direct custody (Crypto Briefing)
◾ Danish pension fund blacklists SpaceX IPO over “catastrophic governance,” “grossly overvalued” pricing (Bloomberg)
Financial Notices & Public Company Releases
◾ Financial updates:
Cango $CANG Q1’26: Revenue $102.0M | Op loss $254.4M | Net loss $261.1M | aEBITDA loss $154.1M (PR Newswire)
SATO Technologies $CCPUF Q1’26: Revenue C$1.4M -54% y/y | Net loss C$1.3M | aEBITDA loss C$578.3K (Newsfile)
Restructuring, Hacks, Losses & Legal Updates
◾ Ex-Celsius CEO Mashinsky files to vacate 12-year fraud sentence (Decrypt)
Crypto Protocols, Applications & Business News
◾ OKX to buy $53M Coinone stake (CryptoNews)
◾ Coinbase $COIN launches India with direct rupee rails (Coinbase)
◾ Coinbase $COIN brings global crypto derivatives to US (Coinbase)
◾ Nomura-backed Laser Digital wins conditional US national trust bank charter (Reuters)
AI Models, Applications & Developments
◾ Nvidia $NVDA unveils AI chip for laptops, PCs (Reuters)
◾ DMG Blockchain $DMGGF signs LOI for 50 MW BC AI data center (GlobeNewswire)
◾ Unitree, Nvidia $NVDA unveil H2 Plus humanoid robot for academic research (PR Newswire)
◾ $IREN taps Nvidia DSX Air digital twin to validate AI-factory deployment (PR Newswire)
◾ Mitsubishi UFJ $MUFG rolls ChatGPT to 35,000 employees (OpenAI)
◾ Anthropic flags unauthorized secondary share-sale platforms (Anthropic)
◾ SoftBank commits €75B to French AI data centers (CNBC)
The Attention Economy, Gaming & Interactive Shifts
◾ $META plans hardware lineup including new AI pendant (The Information)
◾ Kalshi launches first US perpetual futures (Kalshi)
Space Systems, Satellites & Cosmic Activity
◾ US Space Force awards SpaceX $4.1B space-based sensor contract (Space Force)
◾ SpaceX launches 50th Starlink mission of 2026 (Spaceflight Now)
◾ US Space Force awards Blue Origin NRO launch task order (Space Force)
◾ FAA approves SpaceX reusable reentry-vehicle flights (SpaceNews)