Intelligence. Capitalised The definitive AI authority for the global executive. Mapping the strategy and economics of Artificial Intelligence.
Barcelona | Global
Joined April 2026
- Tweets896
- Following117
- Followers53
- Likes112
tokenando retweeted
Someone asked me on Wednesday what was the biggest misconception about AI, I gave 2: that AI is free, and that good AI is very expensive.
Both hide the same elephant in the room: nobody can know what AI will cost before running it.
📰 The Tokenando Daily / Friday November 6, 2026
/ Wall Street suffers its sharpest single-session loss since March 2020, with the Nasdaq 100 falling 6.4% and erasing $2.3tn as Democratic midterm gains trigger a sweeping AI infrastructure sell-off
/ Two fund managers argue the AI equity trade broke because investors bundled infrastructure spend, model capability, and enterprise revenue into a single return thesis that the market is now pricing separately
/ Asian semiconductor markets absorb the Wall Street shock, with South Korea's Kospi triggering a circuit breaker and Bloomberg Economics warning that tighter US export controls could expose China's AI-driven export growth
/ Silver Lake and Saudi Arabia's Public Investment Fund approach Unity Software at a reported $25bn valuation, following their $55bn take-private of Electronic Arts in August 2026
/ SemiAnalysis publishes GDPx, a real-world agent benchmark across 206 professional tasks, with the leading model scoring 12% and every model scoring zero on finance and legal work under binary grading
/ OpenAI releases GPT-6 Cyber to approved security customers, scoring 48.5% on ExploitGym and developing 948 working Linux kernel exploits at roughly $1,279 each, against historical bug-bounty rates of $20,000 to $70,000
/ US voters pass 39 of 40 local ballot measures restricting data centre construction, with an average 75% vote share, as Democratic candidates win governorships in Texas, Georgia, and Ohio on moratorium platforms
🤖 made with AI
✅ reviewed by the Tokenando team
🔗 links and details in following posts
🤖 Made with AI
OpenAI released GPT-6 Cyber on November 5, 2026, a model trained specifically for offensive and defensive security tasks, initially available through the API and Codex to a select group of approved Daybreak Red customers. The model scored 48.5% on ExploitGym, a benchmark testing whether a model can exploit a specified vulnerability in a remote system, up from 32.1% for its predecessor GPT-5.6 Cyber, a 51% relative improvement. OpenAI rates GPT-6 Cyber as "Critical" under its Preparedness Framework cybersecurity category.
In a targeted campaign, GPT-6 Cyber identified 81 previously unknown high-severity vulnerabilities in Firefox, which Mozilla has begun patching, and reproduced 1,724 vulnerabilities in the Linux kernel, developing working exploits for 948 of them at a total compute cost of $1,213,430, or approximately $1,279 per exploit. Programmes such as Pwn2Own have historically paid between $20,000 and $70,000 per comparable finding. On RemoteExploitBench, GPT-6 Cyber completed 95% of tasks; GPT-6 Astra under Daybreak Blue refused every task. A Dragos evaluation found that GPT-6 Cyber identified 2 previously unknown vulnerabilities in engineering software used to manage protective relay data.
Access is restricted to approved Daybreak Red customers with identity verification, single sign-on, multi-factor authentication, and IP allowlisting requirements. OpenAI has also launched a $1bn Daybreak for Frontline Defenders programme to subsidise access for government and essential-service teams. OpenAI acknowledged that adversarial actors are already attempting to distil GPT-6 Cyber's capabilities into their own systems.
🤖 made with AI
Full details via OpenAI / openai.com/index/introducing…
In the November 4, 2026 US midterm elections, voters across 10 states considered approximately 40 to 45 local ballot measures relating to data centre development. Of those with results available, 39 out of 40 measures passed on the side restricting data centre construction, with the sole exception being Harwood, North Dakota, where voters rejected an added review of industrial projects. Restricting measures passed with an average of 75% of the vote. Ohio alone accounted for 21 measures, with communities including Sunbury, Pataskala, Urbana, Granville, and Trenton passing charter amendments restricting hyperscale data centre development by margins exceeding 70%.
At the statewide level, Democratic candidates who campaigned on data centre moratoriums won governorships in Texas, Georgia, and Ohio, and were leading in Kansas and Alaska. Democrats also flipped the Georgia Public Service Commission to Democratic control for the first time since 1996. A CNBC voter poll found that 56% of voters described data centres as a crucial or major factor in their Congressional vote, and 76% said they would oppose new data centre construction in their communities.
UBS estimated that as much as 30% of additional US data centre capacity planned by 2028 could be at risk from construction delays or disruptions. OpenAI's 8-gigawatt PORTS-Pike Technology Campus in Ohio, described as the largest planned AI data centre campus in the world, now faces a governor who supports a conditional moratorium requiring union labour and full utility cost coverage. Pro-data-centre political action committees, including networks linked to Andreessen Horowitz, Greg Brockman, and Joe Lonsdale, spent millions in targeted races and largely failed to shift outcomes.
🤖 made with AI
Full details via TechPolicy.Press / techpolicy.press/data-center…
📰 The Tokenando Daily / Thursday November 5, 2026
/ China's NDRC blocks SoftBank's $5.4bn acquisition of ABB's robotics division, citing risk that operational data could train OpenAI's AI models
/ Toyota establishes Toyota Robotics as a company-wide unit from January 2027, targeting 400,000 factory robots and up to $6.4bn in annual spending from 2028
/ OpenAI acquires Pierre Computer Company to fold its code-storage and developer tools into the Codex platform, targeting GitHub's infrastructure position
/ Apple cuts daily AI spending caps for engineers from $300 to $200, with further reductions to $100 and $25 under consideration across non-engineering teams
🤖 made with AI
✅ reviewed by the Tokenando team
🔗 links and details in following posts
🤖 Made with AI
OpenAI announced on November 4, 2026 that it will acquire Pierre Computer Company, the startup behind Code Storage and a family of open-source developer tools including Trees, Diffs, and Highlights. Financial terms were not disclosed.
Pierre's technology is already embedded in OpenAI's Codex coding platform: Trees powers file navigation and Diffs handles code-change visualisation. Code Storage is a Git-based service that allows applications to store code and track its history; OpenAI plans to integrate its features into Codex while allowing existing customers to continue using the service after the acquisition closes. Pierre's open-source projects will remain open-source.
The acquisition follows OpenAI's November 2 announcement of Pando, an internal source-control backend built to handle the volume of parallel code branches generated by AI coding agents. Pando reportedly reduced the time needed to push code changes by approximately 80%; Code Storage delivered a 70% reduction in code retrieval latency for at least one existing customer. The deal is subject to regulatory approval; until closing, both companies remain independent.
🤖 made with AI
Full details via OpenAI / openai.com/index/openai-to-a…
Apple reduced the daily AI spending cap for some engineers from $300 to $200 in October 2026, Bloomberg reported on November 4, citing people familiar with the matter.
Further reductions are planned. Some engineering teams may see daily limits fall to $100; non-engineering teams, many of which already operate under a $100 daily cap, may see allowances reduced to as little as $25. Employees who require higher limits can apply for exemptions from senior executives. The budgets cover tokens, the units of data that AI models process when handling requests. Apple engineers currently use Anthropic's Claude, OpenAI's ChatGPT, and Google's Gemini; some teams are being directed toward cheaper models or on-device alternatives.
The cuts arrive against a backdrop of rising overall AI expenditure. Apple's research and development costs rose 32% in its 2026 fiscal year on higher infrastructure and AI costs, and operating expenses rose 22% to $75.5bn. The changes are being driven in part by chief executive John Ternus, who took the role on September 1, 2026, and has been pushing for leaner operations and fewer management layers. Microsoft, Tesla, Uber, and Meta have followed a similar pattern of broad AI access followed by structured token rationing.
🤖 made with AI
Full details via Bloomberg / bloomberg.com/news/articles/…
📰 The Tokenando Daily / Wednesday November 4, 2026
/ Texas elects its first Democratic governor since 1990 on a platform opposing new data centre construction, putting more than 25% of planned US capacity at risk
/ Nvidia allocates just 1% of a $70bn Chinese RTX chip order, with global 2027 production of 3 million units already exceeded by Chinese demand alone
/ OpenAI reports GPT-6 Astra designed a physical hardware device in under a week, though manufacturing validation required multiple rounds of human review
/ Canva cuts 1,271 jobs and sees its valuation fall from $42bn to $31bn in 6 months, with AI rollout problems linked to the restructuring
/ Anthropic releases FlashProve, an open-source cryptographic system for verifying which AI model produced a given output, tested at 110 tokens per second on a single H100
/ Data centre operators deploy converted jet engines as primary power, with grid connections in Northern Virginia now taking approximately 7 years versus a 24-month turbine deployment timeline
🤖 made with AI
✅ reviewed by the Tokenando team
🔗 links and details in following posts
Anthropic published a research post and released open-source code on November 2 to 3, 2026, for FlashProve, a cryptographic system for verifying which AI model produced a given output. The system uses zero-knowledge succinct arguments of knowledge, known as zkSNARKs, to allow a model provider to publish a cryptographic commitment to a specific set of model weights and then generate proofs that individual outputs came from those committed weights. The proofs do not reveal the model's weights.
Tested on Meta's Llama 4 Maverick, a 400-billion-parameter mixture-of-experts model, a single H100 GPU generates proofs at approximately 110 tokens per second, with a proof size of approximately 19MB per 1,024-token request, at roughly 70 to 95 times the GPU time required for standard inference. A CPU can verify a proof in approximately 17 seconds. The system carries a current context limit of 1,024 tokens. There are no plans to deploy it commercially at Anthropic or any other lab. The code is released under an Apache 2.0 licence on GitHub. Anthropic collaborated with researchers from MIT, EPFL, StarkWare, and Succinct on the project.
The repository was created on October 30, 2026. Anthropic's stated use cases include verifying that a contracted model, rather than a cheaper substitute, was actually used, and enabling third-party audits of AI outputs in regulated settings.
🤖 made with AI
Full details via Anthropic Research / anthropic.com/research/prova…
Data centre operators are deploying converted jet engines as primary power generation units while grid connection queues extend to multi-year timelines, the New York Times reported on November 3, 2026. The engines, sourced from decommissioned aircraft and rebuilt to burn natural gas, can bring a data centre online in approximately 24 months, compared with approximately 7 years to connect 100MW to the grid in Northern Virginia, according to the International Energy Agency.
Elon Musk's xAI has installed 8 converted jet-engine turbines in Mississippi to help power its Colossus 2 data centre in Memphis, using units from ProEnergy, which rebuilds GE engines commonly found on Boeing aircraft. OpenAI used gas turbines at its first Stargate site in Abilene, Texas, in 2025. Crusoe, which is building the Abilene campus, ordered 13 ProEnergy units in April 2026. An FTAI Aviation joint venture agreed in July 2026 to supply converted turbines to a data centre developer under a $1.5bn deal.
The Southern Environmental Law Center has challenged Mississippi's regulatory approval of xAI's turbines, citing emissions compliance concerns including carbon monoxide at double the rate permitted for other on-site turbines. GE Vernova is quoting 2028 or later for new industrial aeroderivative turbine units, and Siemens Energy reports that more than 60% of US gas turbine orders are now linked to AI data centres. ProEnergy estimates approximately 1,000 CF6-80C2 engines will retire from aircraft service over the next decade, each capable of producing approximately 48MW of output.
🤖 made with AI
Full details via TechCrunch / techcrunch.com/2026/11/03/da…