Vasco Pedro, co-founder and CEO of Unbabel, on the first day of the 2023 Web Summit at the Altice Arena in Lisbon.
Miguel Reis | SOPA | Lightrocket | Getty Images
LISBON — Unbabel on Wednesday announced a translation service powered by artificial intelligence, adding another rival to a highly competitive space — with its CEO warning that humans may not be needed for translation at all in three years.
Widn.AI is Unbabel’s new product and is based on the company’s proprietary large language model (LLM) called Tower. An LLM is an AI model that underpins applications like OpenAI’s ChatGPT.
Unbabel’s LLM allows AI translation in 32 languages, Vasco Pedro, the company’s CEO, told CNBC in an interview at the Web Summit in Lisbon.
“When we started in Unbabel 10 years ago, AI was not at the stage that it is now, and so we were very much focused on creating hybrid solutions that would combine AI and human,” Pedro said.
“But I think for the first time, we believe that translation is now fully in the realm of AI capabilities, and that you can do a lot of things without needing humans at all in the case of translation.”
Unbabel’s traditional product was one that combined so-called machine learning, a type of AI, to translate words, but with human editors to check the final product.
Pedro said Widn.AI will not require humans.
“I think humans still have a slight advantage in very hard use cases. But that advantage right now is so razor thin that except for really the … most difficult use cases, we believe AI is getting really there, and it’s hard for me to see right now how three years from now, you will need humans to be translating anything,” Pedro said.
“There’s still going to be humans responsible for making sure that things get translated and are delivered in the right places,” he added.
Widn.AI is the latest product in an increasingly competitive market which includes Google Translate and products from German startup DeepL.
Those companies see translation as a key area in which LLMs can be used effectively and have trained models specifically to tackle various languages.
Pedro acknowledges that the revenue per translated word is going to “drastically reduce.” But he said there will be an increase in the amount of content translated which will sustain the company’s growth.
Unbabel is speaking to investors and is looking to raise between $20 million and $50 million in funding to fuel the growth and development of Widn.AI, according to Pedro.
Thomas Fuller | SOPA Images | Lightrocket | Getty Images
Video generation startup Luma AI said it raised $900 million in a new funding round led by Humain, an artificial intelligence company owned by Saudi Arabia’s Public Investment Fund.
The financing, which included participation from Advanced Micro Devices’ venture arm and existing investors Andreessen Horowitz, Amplify Partners and Matrix Partners, was announced at the U.S.-Saudi Investment Forum on Wednesday.
The company is now valued upwards of $4 billion, CNBC has confirmed.
Luma develops multimodal “world models” that are able to learn from not only text, but also video, audio and images in order to simulate reality. CEO Amit Jain told CNBC in an interview that these models expand beyond large language models, which are solely trained on text, to be more effective in “helping in the real, physical world.”
“With this funding, we plan to scale our and accelerate our efforts in training and then deploying these world models today,” Jain said.
Luma released Ray3 in September, the first reasoning video model that can interpret prompts to create videos, images and audio. Jain said Ray3 currently benchmarks higher than OpenAI’s Sora 2 and around the same level as Google’sVeo 3.
Humain, which was launched in May, is aiming to deliver full-stack AI capabilities to bolster Saudi Arabia’s position as a global AI hub. The company is led by industry veteran Tareq Amin, who previously ran Aramco Digital and before that was CEO of Rakuten Mobile.
Luma and Humain will also partner to build a 2-gigawatt AI supercluster, dubbed Project Halo, in Saudi Arabia. The buildout will be one of the one of the largest deployments of graphic processing units (GPUs) in the world, Jain said.
Major tech companies have been investing in supercomputers across the globe to train massive AI models. In July, Meta announced plans to build a 1-gigawatt supercluster called Prometheus, and Microsoft deployed the first supercomputing cluster using the Nvidia GB300 NVL72 platform in October.
“Our investment in Luma AI, combined with HUMAIN’s 2GW supercluster, positions us to train, deploy, and scale multimodal intelligence at a frontier level,” Amin said in a release. “This partnership sets a new benchmark for how capital, compute, and capability come together.”
The collaboration also includes Humain Create, an initiative to create sovereign AI models trained on Arabic and regional data. Along with focusing on building the world’s first Arabic video model, Jain said Luma models and capabilities will be deployed to Middle Eastern businesses.
He added that since most models are trained by scraping data from the internet, countries outside the U.S. and Asia are often less represented in AI-generated content.
“It’s really important that we bring these cultures, their identities, their representation — visual and behavioral and everything — to our model,” Jain said.
AI-generated content tools have received significant backlash over the past year from entertainment studios over copyright concerns. Luma’s flagship text-to-video platform Dream Machine garnered some accusations of copying IP earlier this year, but Jain the company has installed safeguards to prevent unwanted usage.
“Even if you really try to trick it, we are constantly improving it,” he said. “We have built very robust systems that are actually using models we trained to detect them.”
Perplexity on Wednesday announced it will roll out a free agentic shopping product for U.S. users next week, as consumers ramp up spending for the holiday season.
“The agentic part is the seamless purchase right from the answer,” Dmitry Shevelenko, Perplexity’s chief business officer, told CNBC in an interview. “Most people want to still do their own research. They want that streamlined and simplified, and so that’s the part that is agentic in this launch.”
The artificial intelligence startup has partnered with PayPal ahead of the launch, and users will eventually be able to directly purchase items from more than 5,000 merchants through Perplexity’s search engine.
Perplexity initially released a shopping offering called “Buy With Pro” for its paid subscribers late last year. The company said its new free product will be better at detecting shopping intent and will deliver more personalized results by drawing on memory from a user’s previous searches.
Perplexity declined to share whether it will earn revenue from transactions that are completed through its platform.
The startup’s competitor OpenAI announced a similar e-commerce feature called Instant Checkout in September, which allows ChatGPT users to buy items from merchants without leaving the chatbot’s interface. OpenAI has said it will take a fee from those purchases.
Read more CNBC tech news
Etsy and Shopify were named as OpenAI’s initial partners for Instant Checkout, but it also inked a deal with PayPal late last month.
Starting next year, PayPal users will be able to buy items, and PayPal merchants will be able to sell items through ChatGPT.
Michelle Gill, who leads PayPal’s agentic strategy, said the company has been building out infrastructure and protections as AI ushers in the “next era of commerce.”
Part of that means keeping consumers and merchants connected to PayPal as they engage on new platforms like Perplexity, she said.
Perplexity said PayPal merchants will serve as the merchants of record through its agentic shopping product, which will allow them to handle processes like purchases, customer service and returns directly.
Through its “Buy With Pro” offering, Perplexity had served as the intermediary that completed purchases.
Gill said PayPal’s buyer protection policies, which can help users get reimbursed if there are problems with their orders, will also apply to transactions on Perplexity.
“We’re really excited about this launch because we will see it come to life during a period that’s so organic for people to shop,” Gill said in an interview.
Nvidia founder and CEO Jensen Huang reacts during a press conference at the Asia-Pacific Economic Cooperation (APEC) CEO Summit in Gyeongju on October 31, 2025.
Jung Yeon-je | Afp | Getty Images
Artificial intelligence chipmaker Nvidia is scheduled to report fiscal third-quarter earnings on Wednesday after the market closes.
Here’s what Wall Street is expecting, per LSEG consensus estimates:
EPS: $1.25
Revenue: $54.92 billion
Wall Street is expecting the chipmaker to guide in the current quarter to $1.43 in earnings per share on $61.66 billion of revenue. Nvidia typically provides one quarter of revenue guidance.
Anything Nvidia or CEO Jensen Huang says about the company’s outlook and its sales backlog will be closely scrutinized.
Nvidia is at the center of the AI boom, and it counts counts every major cloud company and AI lab as a customer. All of the major AI labs use Nvidia chips to develop next-generation models, and a handful of companies called hyperscalers have committed hundreds of billions of dollars to construct new data centers around Nvidia technology in unprecedented build-outs.
Last month, Huang said Nvidia had $500 billion in chip orders in calendar 2025 and 2026, including the forthcoming Rubin chip, which will start shipping in volume next year. Analysts will want to know more about what Nvidia sees coming from the AI infrastructure world next year, because all five of the top AI model developers in the U.S. use the company’s chips.
As of Tuesday, analysts polled by LSEG expect Nvidia’s sales to rise 39% in the company’s fiscal 2027, which starts in early 2026.
Investors will want to hear about Nvidia’s equity deals with customers and suppliers, including an agreement to invest in OpenAI, a deal with Nokia and an investment into former rival Intel. Nvidia has kept its pace of deal-making up, agreeing to invest $10 billion into AI company Anthropic earlier this week.
Nvidia management will also be asked about China, and the possibility that the company could gain licenses from the U.S. government to export a version of its current-generation Blackwell AI chip to the country. Analysts say Nvidia’s sales could get a boost of as much as $50 billion per year if it is allowed to sell current-generation chips to Chinese companies.