News · Weekly briefing
The Compute
A weekly briefing on the AI economy: GPU markets, frontier model releases, capital flows and regulatory shifts. Built for infrastructure leaders, operators and investors.
Published issues
A Red Phone for the Tech Cold War
Washington and Beijing agreed an AI incident channel in September 2026. Ricardo Tavares on what it must handle and what the Shenzhen summit should deliver.
Read on SubstackThe weakest link: AI, hackers and the human in the loop
How AI found a decade-old flaw affecting most of the world’s websites before the bad guys did, and a human mistake leaked it a day too early
Read on SubstackThe buyer who never switches off: why AI will pay any price for electricity
Energy executive Yasin Kasirga on the Compute Heat Rate: why AI data centres can pay 142 times the industrial power price, and who pays for the emissions.
Read on SubstackA matter of taste: maths, love, judgment and the oldest trick in the book
OpenAI says ChatGPT still has work ahead on its design judgment. On taste, design systems, Xerox PARC and the Renaissance art of sprezzatura.
Read on SubstackSuper powers: from New York to Washington and Barcelona
Washington renamed AI "Super Intelligence" with the same legal definition. Europe offers free supercomputer time, but the machines are full.
Read on SubstackRunning on trust: fast robots, slow deployments
Humanoid robot order books assume a boom that shipments do not yet show. Why the cost of putting a robot to work, not its price, sets the pace.
Read on SubstackThe Price of Losing Memory
AI accelerators took the memory. Phones, laptops and consoles are paying for it. Ricardo Tavares on the shortage, the HBM tax and the Apple exception. THE COMPUTE / BY TOKENANDO Issue XX / September 17, 2026
Read on SubstackFrom the kilogram to the token: what happens when the provider defines the unit
Read on SubstackFast forward to 1997
Portals, mispriced losses and features that became companies. The 1997 internet playbook, and why OpenRouter and Hugging Face did not have to wait three years.
Read on SubstackThirty Trillion, All the Way Down
Anthropic's $30tn TAM, taken apart layer by layer: users and tokens, racks and Macs, fabs and machines, and the arithmetic the ceiling requires.
Read on SubstackIntelligence, Collateralised: Inside AI's New Asset Class
Read on SubstackThe Elephants in the Room in the September U.S.-China AI Talks
Read on SubstackShould I stay or should I go
Read on SubstackRace to the top: when free stopped being cheap
Read on SubstackStandard definition: 5 AI words we use every day, and what each one actually means
Read on SubstackThe Music Dispatch: Human after all
Read on SubstackThe great turnaround: the jobs everyone thought were safe
Read on SubstackClaude-y with a chance of open bots
Read on SubstackThe cost of control
Read on SubstackA Tale of Three Continents
Read on SubstackWhat we cover
GPU and compute markets
Spot pricing, hyperscaler capacity shifts, and emerging inference infrastructure.
Token pricing moves
Provider price changes, cost benchmarks, and what they signal about competitive dynamics.
Model releases and benchmarks
What is shipping, what it costs, and whether the benchmarks hold up under scrutiny.
Capital flows and M&A
Funding rounds, strategic acquisitions, and where the smart money is moving.
Regulation and policy
EU AI Act updates, US executive orders, and what operators need to act on.
API and platform shifts
Rate limit changes, new modalities, deprecation timelines, and migration signals.
Methodology
The Compute draws on Tokenando’s pricing database (743 models, 75 providers), publisher feeds from primary sources, and original analysis. We do not reproduce press releases. Every claim is sourced, and our analytical frameworks are disclosed in full on the About page.