US United States

United States

The world's largest AI market: capital, models, platforms.

Your role

Text size

A migration roadmap sets firm deadlines for US agencies and their suppliers to adopt quantum-safe encryption.

NIST publishes a timeline for the shift to quantum-safe cryptography by 2030 and 2031.

The US agency NIST has released report NISTIR 8547. Federal agencies must move key exchange to quantum-safe standards by 2030 and digital signatures by 2031. The report also serves companies as a template for their own migration.

Executive

InfrastructureOctober 27, 2026NIST AI Risk Management Framework

Microsoft offers a voluntary exit to about 60,000 staff while posting the fattest margins in its history.

Microsoft offers a voluntary exit to a quarter of its workforce despite record margins.

Microsoft has offered roughly 60,000 employees across the US, UK, Canada and India a voluntary exit, about 27 percent of its 223,000 staff. Capital spending has climbed from 23 billion dollars five years ago to about 170 billion dollars now, and revenue per employee stands at 1.5 million dollars.

ExecutiveHR

Work & CostOctober 29, 2026Azeem Azhar – Exponential View

OpenAI shows how fast agents can replace classic chat tools across a whole company.

Codex share of OpenAI staff AI output jumps from under 10 to 99.8 percent in ten months.

A working paper by David Holtz, who heads AGIO Economic Research at OpenAI, shows that Codex went from under 10 percent of employees AI-driven output in August 2025 to 99.8 percent by June 2026. Legal and recruiting teams now run nearly all their AI work through the agent as well.

ExecutiveHRSales

Work & CostOctober 27, 2026MIT Initiative on the Digital Economy

Open AI models get caught between new US and Chinese export controls.

DeepSeek releases its own inference stack while the US and China tighten rules on open models.

China proposed pre-approval for exporting frontier models, and the US barred DeepSeek from federal procurement. At the same time DeepSeek open-sourced its inference stack DeepInfer, estimated to earn up to 90 percent gross margin on token sales.

ExecutiveSales

Markets & CapitalOctober 27, 2026Nathan Lambert – Interconnects

AI use is nearly universal in companies, but strong returns remain rare and concentrated.

The 2026 Wharton AI adoption report finds that 91 percent of firms use AI weekly, yet only 28 percent report strongly positive returns.

The fourth edition of the Wharton-GBK study tracks AI adoption, return, agent use and workforce impact over four years. In 2026, 91 percent of firms use AI at least weekly and 62 percent daily. Positive return is near-universal at 80 percent, but strongly positive return reaches only 28 percent, concentrated among firms measuring growth rather than efficiency alone.

ExecutiveHR

Work & CostOctober 13, 2026Knowledge at Wharton – AI

Chip-scale precise timing could give drones, ships and vehicles a fallback when GPS signals fail or are jammed.

NIST demonstrates a chip-based optical frequency comb built for precise timing without GPS.

Researchers at NIST and the University of Maryland built an optical frequency comb on a microchip that works without bulky lab equipment. Each chip holds up to 300 ring-shaped resonators generating precise timing signals, a step toward navigation systems that do not depend on GPS.

Executive

Models & TechnologyOctober 16, 2026NIST

Nearly all 94 US reactors are testing AI assistants because skilled staff are scarce and tech firms need power.

US nuclear plants are adopting AI assistants in control rooms despite the industry's strict safety culture.

Nearly the entire US fleet of 94 reactors has been offered AI integration, and most operators accepted. Vendors such as Atomic Canyon and Nuclearn now supply assistants for maintenance, compliance and operations.

ExecutiveHR

InfrastructureOctober 8, 2026IEEE Spectrum – Artificial Intelligence

The cost of a fixed level of AI performance is falling faster than any other technology on record.

Epoch AI reports plunging AI costs, billions of possible agents, and thin Chinese AI revenue versus US labs.

Epoch AI's October 8 brief puts the quarterly price drop for fixed AI performance at 47 percent, roughly 13 fold per year. Chips shipped through 2027 could run up to 2 billion cheap agents at once. Six Chinese AI firms together earn only a tenth of what OpenAI and Anthropic make combined.

ExecutiveSales

Markets & CapitalOctober 8, 2026Epoch AI

Chatbot use for election information is growing faster than the rules that govern it before the midterms.

Princeton researchers map four different approaches major AI firms take to election questions ahead of the US midterms.

A 2026 Pew survey finds 42 percent of US adults now use AI chatbots to search for information, while a 2024 Bipartisan Policy Center survey counted 27 percent asking AI tools election questions monthly. Princeton researchers compare how OpenAI, Anthropic, Google, Meta, xAI and Microsoft handle procedural, results, candidate and campaign content queries.

ExecutiveMarketing

RegulationOctober 8, 2026Princeton CITP – Freedom to Tinker

Florida takes OpenAI to court instead of just threatening sanctions.

Florida's attorney general has asked a court for an injunction against OpenAI and ChatGPT.

Florida Attorney General James Uthmeier filed for a court injunction against OpenAI on September 28. The motion cites an Axios report that leading AI labs are investigating tens of thousands of security incidents. The same day, Nvidia unveiled a new safety platform for AI agents.

ExecutiveSales

RegulationOctober 6, 2026Gary Marcus

AI performance is getting cheaper faster than any past transformative technology.

A study by Epoch AI finds that the cost of AI performance falls thirteenfold every year.

Epoch AI compared the cost of reaching set benchmark scores with AI models over time. The price per unit of performance drops thirteenfold per year, faster than DNA sequencing, computing power, or lithium batteries.

ExecutiveSales

Markets & CapitalOctober 6, 2026Epoch AI

Compute is growing faster than the question of who actually needs the output of these agents.

Epoch AI calculates that planned data center buildouts could support more AI agents than the US has workers.

Epoch AI estimates that compute available by 2027 could support between 140 million and 1.9 billion concurrently running AI agents, depending on the model used. That range implies up to 8 billion full time equivalent work hours per week, compared with roughly 100 million knowledge workers in the United States.

ExecutiveHR

Work & CostOctober 6, 2026Epoch AI

AI-GENERATEDInference costs fall faster than any earlier transformative technology.

AI inference costs are falling by a factor of 13 per year, according to Epoch AI.

Epoch AI tracked the cheapest way to hit fixed benchmark scores over time and found a 47 percent quarterly cost decline. One example: a 75 percent score on GPQA Diamond cost 30 cents per question in January 2025 and just 0.04 cents in mid-2026.

ExecutiveSales

Markets & CapitalSeptember 28, 2026Epoch AI

AI-GENERATEDProposed safety rules in Washington would target closed models only.

The White House discussed voluntary safety testing in August that would apply only to proprietary AI models.

In August, White House officials met with leading AI companies to discuss voluntary safety testing limited to proprietary frontier models, leaving open models exempt. MIT research shows closed models capture 96 percent of industry revenue at six times the cost of open alternatives, even as open models close much of the performance gap.

Executive

RegulationSeptember 23, 2026MIT Initiative on the Digital Economy

AI-GENERATEDBeyond 250 kilowatts per rack, mixed air and liquid cooling stops being enough.

At high rack power, near total liquid cooling becomes a technical requirement for AI servers.

IEEE Spectrum outlines why hybrid air and liquid cooling reaches its limits above 250 kilowatts per rack. With a common 70/30 split, 75 kilowatts of heat still needs air cooling. Manufacturer CoolIT, an Ecolab company, presents systems that capture nearly all heat through liquid, enabling fanless server designs.

Executive

InfrastructureSeptember 29, 2026IEEE Spectrum

AI-GENERATEDToken security becomes mandatory homework for firms running cloud and AI agent systems.

NIST publishes final guidance on protecting identity and access tokens from misuse.

On September 15, 2026, NIST and CISA released NIST IR 8587, giving organizations concrete steps to prevent theft and forgery of access tokens. The publication cites an attack in which forged tokens derived from one stolen signing key let attackers steal more than 60,000 emails from a single agency.

Executive

Safety & SecuritySeptember 15, 2026NIST AI Risk Management Framework

AI-GENERATEDOnly eleven percent of S&P 500 firms have AI deeply embedded in their business.

MIT research finds real AI integration at S&P 500 companies moves slower than the headlines suggest.

An MIT analysis of mandatory filings with the US Securities and Exchange Commission finds that by the end of 2025 only 11 percent of S&P 500 companies had AI deeply embedded in their business, while 45 percent were running pilots. Two thirds of deep integration sits in technology companies alone.

ExecutiveSales

Markets & CapitalSeptember 16, 2026MIT Initiative on the Digital Economy

Wharton economist Peter Cappelli explains why AI reshapes tasks more than it eliminates whole jobs.

Wharton professor Peter Cappelli argues that AI mainly changes tasks rather than eliminating entire positions.

In a conversation published on September 18, 2026, Wharton management professor Peter Cappelli reviews new research on how AI actually affects workplaces. His central point is that the technology reshapes individual tasks and job design far more than it removes whole occupations.

ExecutiveHR

Work & CostSeptember 18, 2026Knowledge at Wharton – AI

Fresh federal money targets regional cybersecurity talent pipelines in eight states.

NIST awards more than 1.7 million dollars to build cybersecurity workforce programs in eight states.

On September 18, 2026, NIST approved nine cooperative agreements under its RAMPS program. More than 1.7 million dollars will go to organizations in eight states to build local cybersecurity training pipelines.

ExecutiveHR

Work & CostSeptember 18, 2026NIST AI Risk Management Framework

AI-GENERATEDOpenAI uses its own language models to design its first custom chip faster than usual.

OpenAI unveiled Jalapeno, its first custom AI accelerator, designed with help from its own language models.

On August 25, OpenAI unveiled Jalapeno, its first custom AI chip. It delivers up to 13.4 petaflops of 4-bit compute and 232 gigabytes of memory linked at 15.4 terabytes per second. OpenAI says the chip cuts latency versus Nvidia's GB300 by up to 3.6 times.

ExecutiveSales

Models & TechnologySeptember 14, 2026IEEE Spectrum – Artificial Intelligence

AI-GENERATEDMandatory labeling of AI text is becoming standard practice across the industry.

Anthropic will embed a watermark in text from all future Claude models.

On 11 August Anthropic announced that all future Claude models will produce watermarked text. Google already applies its own method for Gemini, and OpenAI plans a similar approach.

ExecutiveMarketing

Models & TechnologySeptember 15, 2026IEEE Spectrum – Artificial Intelligence

More code gets written, but the path to customers remains the bottleneck.

A Wharton study finds that review and deployment hold back AI productivity gains in software.

A study tracking over 100,000 GitHub developers finds autocomplete lifts coding activity by 40 percent and autonomous agents by 180 percent. Finished software releases, however, grew by only 30 percent.

ExecutiveHR

Work & CostSeptember 15, 2026Knowledge at Wharton – AI

AI-GENERATEDA Wharton study puts a number on the bet behind Big Tech's AI spending.

A Wharton study finds that Big Tech's roughly one trillion dollar AI investment only pays off if productivity nearly triples.

New Wharton research finds hyperscalers need to raise AI productivity by a factor of 2.7 to justify close to one trillion dollars in spending by 2027. The authors warn that a shortfall could become the largest capital misallocation in history.

Executive

Markets & CapitalSeptember 1, 2026Knowledge at Wharton

AI-GENERATEDUnderstandable explanations aim to build trust in self-driving cars.

MIT and Motional built a system that translates self-driving car decisions into concepts humans can follow.

The new system, called CW-Net, explains why an autonomous vehicle brakes or swerves using concepts such as approaching stopped vehicle. In track tests and a simulation study, safety drivers predicted vehicle behavior more accurately with these explanations.

ExecutiveSales

Models & TechnologySeptember 2, 2026MIT News

Skip the manual steps and you skip the skills - a lesson from nuclear control rooms for AI era engineering.

A systems engineer warns that AI efficiency is undermining the training of future experts.

Richard Mitchell, who once led the control design for a US nuclear plant, deliberately kept manual steps inside a fully automated system. A Harvard study of 65 million workers at 280,000 firms shows why that instinct matters again: entry level jobs fell about 9 percent after genAI adoption.

ExecutiveHR

Work & CostSeptember 2, 2026IEEE Spectrum – Artificial Intelligence

AI-GENERATEDThe combined annualized revenue of Anthropic and OpenAI has more than tripled so far in 2026.

Anthropic has reportedly reached a 65 billion dollar annualized revenue run rate, while OpenAI stands above 40 billion dollars.

Epoch AI has published new figures on revenue growth at the two leading AI labs. The combined annualized run rate of Anthropic and OpenAI rose from 30 billion dollars in January 2026 to 105 billion dollars by August 2026.

ExecutiveSales

Markets & CapitalAugust 27, 2026Epoch AI

AI-GENERATEDA new tool makes the reasoning of large language models visible.

Startup Goodfire opens its Silico platform for interpretability research and companies.

The startup Goodfire has made its Silico platform generally available. The tool aims to show how large language models such as Claude, ChatGPT and Gemini arrive at their answers. Goodfire also launched a grant program worth 1 million dollars for academic and nonprofit interpretability research.

ExecutiveSales

Models & TechnologyAugust 26, 2026IEEE Spectrum – Artificial Intelligence

AI-GENERATEDAdaptability becomes the decisive skill for engineers in the AI era.

Employers expect 39 percent of core skills to change by 2030, according to the World Economic Forum.

The AI boom is changing tools, expectations and hiring odds for engineers. A World Economic Forum report finds that employers expect about 39 percent of workers' core skills to change by 2030. A June 2026 report by PwC names technology, media and telecom as the sectors with the fastest skill turnover.

ExecutiveHR

Work & CostAugust 24, 2026IEEE Spectrum – Artificial Intelligence

AI-GENERATEDThe biggest vendor shift of the year is happening in pricing

Anthropic's annualised revenue run rate passes 65 billion dollars

At the end of July the annualised run rate stood above 65 billion dollars, according to people familiar with the figures – up from 47 billion in May and around 9 billion at the end of 2025. A listing is expected in the autumn.

Executive

Markets & CapitalAugust 17, 2026CNBC

AI-GENERATEDEurope's labelling duty becomes visible inside other people's documents

Anthropic's watermark makes text produced with Claude permanently detectable

Every Claude model released since 2 August embeds a statistical watermark in its word choice to satisfy the EU AI Act. Lacking a regional switch, Anthropic applies the method worldwide.

ExecutiveMarketing

RegulationAugust 17, 2026THE DECODER

AI-GENERATEDNot fines, but design rules for recommendation systems

Twenty-nine US states seek court-ordered changes to Meta's algorithms

The suit targets not only monetary penalties but conditions on the recommendation systems themselves. Meta points to existing age limits, parental controls and terms of use.

ExecutiveMarketing

RegulationAugust 17, 2026IT BOLTWISE

AI-GENERATEDModel selection becomes an in-house measurement task

Artificial Analysis launches Optima, a platform for custom AI benchmarks

Users can build test suites from their own data and use cases and compare models by quality, cost and speed per task. It targets the known weakness of public leaderboards.

Executive

Models & TechnologyAugust 16, 2026THE DECODER

AI-GENERATEDThey can combine; so far they do not invent

Leading mathematicians identify a clear limit in large language models

Timothy Gowers and Peter Sarnak credit large language models with substantial ability but see limits when genuinely new ideas are required. The systems are strong at combining known methods.

Executive

Models & TechnologyAugust 16, 2026THE DECODER

Data centres become a local issue

Artificial intelligence is a major US election issue for the first time

A Washington Post analysis of more than 1,200 candidate websites shows AI features in almost 40 percent of House, Senate and gubernatorial races. The dominant sub-topic is data centres.

Executive

RegulationAugust 14, 2026Washington Post

Junior staff assess AI differently from the executive floor

Almost half of young US workers expect AI to harm their careers

In a CNBC and Generation Lab survey of 1,088 people aged 18 to 34, 45 percent expect AI to have a negative effect on their careers and 10 percent a positive one. Trust in leading AI executives is uniformly low.

ExecutiveHR

Work & CostAugust 13, 2026CNBC

Free subscription

Get the whole edition by email. Free of charge.

Every Tuesday morning, seven documented stories with what each one means for your decisions. One click to unsubscribe.

Subscribe free of charge

No costs, no advertising, no forwarding of addresses.

Free subscription

The whole edition free of charge by email.

Seven documented stories from research, public authorities and standardisation – and what they mean for executives, marketing, HR and sales. No costs, one click to unsubscribe.

Subscribe free

Double opt-in: nothing is sent before you confirm the link in the email.

Editorial principles

Always the original source

Every story names its source and links to it directly. We do not pass on information we cannot trace.

Interpretation, not excitement

Every story states what it means for decisions in your company – concretely, not as a buzzword.

Organised by country

The United States sets the pace, Germany sets the frame. The other markets follow by actual AI activity.