noticias ia para empresas openai agentes de ia

OpenAI is poaching large clients from Anthropic with discounts that cut bills by up to 90%: the list price is no longer the real price. Meanwhile, Apollo's chief economist offers the flip side: outside of Big Tech, AI has yet to move either margins or revenue per employee for U.S. companies. Starting today, Sweden's government is blocking ChatGPT and Claude for its staff, and money keeps flowing into AI that watches machines: Tractian raises $130 million. This picks up from the Sunday, Oct 25 radar.

OpenAI cuts AI bills by up to 90% to poach Claude customers: the list price is no longer the price

OpenAI is snatching large customers away from Anthropic, the company behind Claude, with aggressive discounts, according to The Information. Rippling, the payroll and HR software firm, now sends 52% of its AI coding workload to OpenAI, up from 16%, while Anthropic's share has dropped from 57% to 18%, according to its CEO, Parker Conrad. Replit, the coding platform, switched after OpenAI offered a discount of over 30% on a contract worth more than $100 million a year. DataCamp CEO Jonathan Cornelissen says OpenAI's pricing has cut his costs by 80% to 90%, and a U.S. bank with over $7 billion in assets has doubled its ChatGPT licenses to 500 employees at no extra cost: it pays less than $7 per user, compared to the usual $20 or more. OpenAI is funding part of the discount out of its own margin, but part of it is real: its newer models are more efficient, and Chinese competitors like DeepSeek and Kimi are putting pressure on both. As we flagged on Saturday, Oct. 24: both need revenue before going public, and their weapon is price. What it means for you: if you pay for ChatGPT, Claude, or Copilot licenses, or consume their models via API, get a written quote from a rival vendor this week and take it to your current one. Lock in the price for at least twelve months, with a clause that lets you review the contract if conditions change. And before switching, test the new model on ten real tasks from your business: the savings mean nothing if the results get worse. If you can't switch models in a week, you're not negotiating: you're taking whatever price they give you. Source

Apollo: AI has yet to improve a single margin outside Big Tech, and companies are starting to see it as a cost

Torsten Slok, chief economist at Apollo, one of the world's largest private equity firms, published a note on Saturday with a blunt conclusion: despite massive investment, AI has delivered "zero financial payoff to corporate America outside of tech." His charts separate the Magnificent Seven tech giants from the rest of the S&P 500 index, the other 493 companies: for the seven, revenue per employee and profit margins are rising; for the 493, they remain flat, and there are no signs that AI is accelerating productivity. According to Slok, without an effect on the bottom line, mainstream companies are pulling back and starting to see AI spending not as a source of growth, but as a cost eating into margins. His explanation: AI might be a great calculator, but in business, intelligence is primarily judgment. The takeaway: the data doesn't say AI is useless; it says buying it isn't enough. Pick a specific process, record what it costs today—in hours, in cost per order, or in errors—and measure it again three months after automating it. Anything that doesn't move one of those figures should be cut, even if it's cheap: a discount makes tokens cheaper, it doesn't make them profitable. AI that doesn't move a margin or a measured hour doesn't move you up a level: it just moves your bill up. Source

Sweden blocks ChatGPT and Claude on government computers starting today: no contract, no tool

The Government Offices of Sweden, the body serving the Prime Minister and ministries, are blocking ChatGPT and Claude on staff work computers starting this Monday, according to Swedish public radio. They want employees to use only approved services, including Microsoft's Copilot, which has had its security upgraded following a review. Mikael Wettercrantz, its director of digitalisation, explained on the intranet that using AI services without a contract can have "consequences for our information security." They aren't alone: the Swedish Tax Agency has already introduced blocks, and an internal audit of the transport agency this year found widespread, risky use of around ten different AI services, with issues concerning both the sensitive information being entered and how the outputs were used. However, some staff will continue using ChatGPT and Claude through an internal platform built by the government itself. What it means for you: the problem isn't which brand you use, but using it without a contract. Ask your team which AI tools they actually use, because the Swedish agency found around ten. Then choose one or two with an enterprise contract that stipulates in writing what the provider does with what you write, and roll it out to everyone before you block the rest: a ban with no alternative just drives usage underground. If you don't know which AI tools your staff use, you're not at the ASSISTED level: you're flying blind. Source

Tractian raises $130 million at a $1.4 billion valuation: the money is going to AI that watches machines

Brazil's Tractian, which uses sensors and AI to monitor industrial equipment, detect faults, and recommend maintenance before breakdowns occur, has closed a $130 million Series D round valuing it at $1.4 billion. The round was led by Sapphire Ventures and Catalyst, the growth equity arm of M&G, with Bosch Ventures and General Catalyst among other investors. Founded in 2019, it is generating revenue at a rate of nearly $100 million a year, half of it in the United States, and its technology operates in more than 2,000 facilities, including those of Coca-Cola, Whirlpool, Embraer, and Eli Lilly. It will invest the money over five years in AI for the physical world, proprietary language models, and Nvidia processors. The takeaway: this is the practical answer to Apollo's complaint. Predictive maintenance is one of the few AI use cases that measure themselves: machine downtime hours, cost of each breakdown, and parts replaced prematurely. If you have a plant, record those three figures from the last year before requesting a demo, and ask the provider who owns the sensor data and if you can take it with you: on Sunday, Sundial showed us what happens when a startup gets acquired. A sensor that alerts and a person who decides is INTEGRATED; UNATTENDED is when the work order opens automatically and someone reviews the failures. Source

What to watch this week

Microsoft reports earnings on Wednesday, Oct. 28, and Amazon on Thursday, Oct. 29: we'll see whether their enterprise customers are paying more for AI or starting to cut back, as Apollo suggests. And watch whether Anthropic answers to OpenAI's discounts with its own.