Blog

Stay informed with free updates

Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.

US groups such as OpenAI that are racing to develop artificial intelligence are at risk of being undercut by cheaper rivals such as China’s DeepSeek, according to star investor and analyst Mary Meeker whose presentations on tech trends are followed across Silicon Valley.

Meeker, an early backer of companies including Meta, Spotify and Airbnb, told the Financial Times that new AI advances will mint “multiple companies worth $10tn and they probably will not all be based in North America”.

She added “the wealth creation will be extraordinary. We have never had a five billion-user market that one could get to so easily.”

Her latest report on industry developments highlights the growing challenges for US groups that have taken an early lead in developing large language models (LLMs).

Meeker argues that the costs to train models such as OpenAI’s GPT series or Google’s Gemini is rising, while increasing competition from cheaper domestic models and overseas groups such as DeepSeek is putting pressure on pricing power.

“The business model is in flux. And there are new questions about the one-size-fits-all LLM approach, with smaller, cheaper models trained for custom-use cases now emerging,” according to the report. “In the short term, it’s hard to ignore that the economics of general-purpose LLMs look like commodity businesses with venture-scale burn.”

Meeker was dubbed “the queen of the internet” as a result of her analysis while working at Morgan Stanley, which anticipated the success of Google, Apple and others. She joined venture capital firm Kleiner Perkins in 2010, before co-founding her own investment firm, Bond, in 2019.

Her firm’s latest presentation illustrates how explosively the market for generative AI tools has taken off since the launch of OpenAI’s ChatGPT chatbot in late 2022.

The tool has raced to more than 500 million monthly users in two years, spurring venture capital bets on a panoply of rivals.

Investment into data centres and other AI infrastructure has also accelerated. Capital expenditure from the six largest technology companies in the US jumped more than 60 per cent year over year to more than $200bn in 2024.

That has given the three leading US start-ups in the LLM race — OpenAI, xAI and Anthropic — a combined valuation of around $400bn and boosted the revenues and share prices of Big Tech companies such as Nvidia, Meta and Microsoft.

Meeker’s analysis shows that LLM companies also face growing threats: massive upgrades to chips and algorithms have slashed the costs of running models, opening the door for rivals including China’s DeepSeek to launch cheaper and more efficient models.

Training the most advanced models remains a prohibitively expensive task. Estimated costs to develop cutting-edge models have increased by 2,400-times in eight years, pricing out all but a handful of competitors which lack a clear path to profitability.

OpenAI, Anthropic and xAI have raced to a collective annualised revenue of around $12bn. But they raised a combined $95bn to do so. OpenAI’s valuation-to-revenue multiple “looks expensive”, according to Meeker.

Price wars and a proliferation of cheaper models are good news for consumers, but mean start-ups aiming to take full commercial advantage of their technology will need deep pockets and patient financiers.

Meeker compared their challenge to Uber, Amazon and Tesla, each of which burnt through cash in order to establish a large business they could ultimately monetise.

“The rules that hold well in these times of euphoria are only invest what you’re willing to lose, and take a portfolio approach,” said Meeker. “Putting all your eggs in one basket is a risk here, because everything is up and to the right — until it isn’t.”

This has been a banner month for X. Last week, the social network’s built-in chatbot, Grok, became strangely obsessed with false claims about “white genocide” in South Africa—allegedly because someone made an “unauthorized modification” to its code at 3:15 in the morning. The week prior, Ye (formerly Kanye West) released a single called “Heil Hitler” on the platform. The chorus includes the line “Heil Hitler, they don’t understand the things I say on Twitter.” West has frequently posted anti-Semitic rants on the platform and, at one point back in February, said he identified as a Nazi. (Yesterday on X, West said he was “done with antisemitism,” though he has made such apologies before; in any case, the single has already been viewed tens of millions of times on X.)

These incidents feel all too natural for Elon Musk’s social network. Even without knowing the precise technical reason Grok decided to do its best Alex Jones impression, the fact that it became monomaniacally obsessed with a white-supremacist talking point says something about what the platform has become since Musk took over in October 2022. Specifically, it validates that X has become a political weapon in his far-right activism. (To be clear, white farmers have been murdered in South Africa, which has one of the world’s highest murder rates, according to Reuters. But there is no indication of a genocide. In 2024, eight of the 26,232 murders nationwide were committed against farmers. Most murder victims there are Black.)

Read: The day Grok told everyone about ‘white genocide’

This has been obvious to anyone using the site or paying attention to Musk’s managerial decisions. He’s reinstated thousands of banned accounts (QAnon supporters and conspiracy theorists, and at least one bona fide neo-Nazi), and the platform is engorged with low-rent outrage porn, bigoted memes, MAGA AI slop, and, well, a lot of people proudly using racial slurs, frequently to attack other people. The platform’s defenders would likely argue that X is an experiment in free-speech maximalism and that it is one of the only truly neutral zones on social media. Musk and his sycophants have constantly cited his takeover as an attempt to “solve free speech”; Joe Rogan has suggested that Musk has done just that. (This isn’t quite accurate, as X has complied with government takedown requests, temporarily suspended journalist accounts, amplified accounts that promote Musk’s worldview, and tried to censor words its owner doesn’t like: Last year, it briefly warned users who attempted to use the word cisgender in posts, after Musk said he considers it a “slur.”)

But Grok’s white-genocide Wednesday is a major indication that the platform is not neutral. Either X has a natural bias, based on the site’s architecture and user base—that is, the chatbot, which is able to search tweets in real time, acts on an attitude that is endemic to the platform—or X is being directly manipulated to emphasize a certain viewpoint. In other words: Either way, X is racist. The only thing up for debate is whether this is a feature or a bug for those in charge.

Twitter always had an outsize cultural influence, and X—despite its marked decline under Musk—does as well. Yet mainstream culture is no longer dominant there: The media outlets and public figures are now punch lines for the site’s main characters, Musk and his MAGA acolytes. Platform events such as the Grok rampage and Ye’s “Heil Hitler” offer a window into the ways that X has become an accelerator for a broader, more durable culture of hate. It’s not only that some of this vile discourse seeps out into the physical world (memes about immigrants eating cats and dogs leading to harassment in Ohio, Trump bringing up conspiracy theories about white genocide during an Oval Office meeting with the South African president)—it’s that the worst of the internet is no longer relegated to the shadows. Instead, it is elevated, perhaps even at times normalized, by its proximity to everyone else’s content.

Last Wednesday, as I watched Grok bring up white genocide in response to an anodyne query about the Toronto Blue Jays pitcher Max Scherzer’s career earnings, I couldn’t shake the question: Why are people still using this website? The same thought had also occurred to me around the time that Ye released “Heil Hitler” and I toggled over to X’s algorithmic “For You” feed. It showed a smattering of the platform’s least savory commentators posting about how the anti-Semitic anthem was “the song of the year” and how it had become popular in Thailand. What happened next is pretty standard: By clicking on a few posts about the song, I’d expressed enough interest in it that the platform fed me a steady stream of “Heil Hitler” content: AI-generated remixes of the song, covers, dozens of memes about how the song was secretly popular. I saw a video of a white couple singing the song in their car, throwing up Nazi salutes. Not long after that, I saw a link to a crowdfunding campaign for that same couple, who were asking for money to “relocate” after their video went viral and they were doxxed and “threatened.” The couple set their funding goal at $88,000—a reference, almost assuredly, to “88,” a neo-Nazi code for “Heil Hitler.” This Russian nesting doll of irony-poisoned, loud-and-proud racism is a common experience in the algorithmic fever swamps of X.

It’s worth noting that Ye’s song was banned by other major streaming platforms and social networks. Writing about X, The New Yorker’s Kelefa Sanneh said, “West has given the platform a kind of exclusive hit single—a song that can be heard almost nowhere else.” Neo-Nazis and trolls expressed a palpable delight that all of this was happening on an ostensibly mainstream platform—wanton hatred not on 4chan or Stormfront, but on the same network where Barack Obama posted a condolence message about Joe Biden’s cancer diagnosis. “Heil Hitler” is almost assuredly not the global phenomenon that the fascists on the platform think it is, but its prevalence on X is not nothing either. As Sanneh wrote last week, “We now live in an era when a top musician can distribute a song called ‘Heil Hitler,’ and there’s no way to stop him. That is the true message of this song, which has spread and thrived beyond the reach of boycotts or shaming campaigns: no one is in charge.”

In July 2020, the Twitter user Michael B. Tager shared an anecdote that went viral. Tager was at “a shitty crustpunk bar” when the gruff bartender kicked out a patron in a “punk uniform”—not because the customer was making a scene, but because he was wearing Nazi paraphernalia. “You have to nip it in the bud immediately,” Tager recounted the bartender as saying. “These guys come in and it’s always a nice, polite one. And you serve them because you don’t want to cause a scene. And then they become a regular and after awhile they bring a friend.” Soon enough, you’re running a Nazi bar.

The Nazi bar is an apt analogy, yet it doesn’t fully capture the weirdness of a social network and of the strange, modern power of algorithms to sort and segregate experiences. Many people use X merely to post about sports, follow news, or look at dumb memes, and they’re probably having a mostly normal online experience; I don’t have any wish to judge them. To torture the metaphor, though, they’re sitting at a table outside the Nazi bar; their friends are there, they’re having a good time, maybe they hear a slur emanate from the window from time to time. Others fully recognize that they’re at a Nazi bar, but this was their bar first and they don’t want to cede the territory; they’re hanging around to debate, never mind that the bar’s owner is palling around with the new customers.

Of course, with a broadcast social network like X, everyone is both a patron and an owner of sorts. Followers can feel like a kind of currency, built up over years: Some people don’t leave the bar, because they’re invested and don’t want to dump their shares. Other people don’t leave, because the alternative hangouts aren’t enticing enough. Some simply don’t want to give the Nazis the satisfaction of successfully driving them out. There is plenty of commentary, even among users of other platforms, about how Threads is bloodless (and owned by Mark Zuckerberg), Mastodon is inscrutable, and Bluesky is humorless.

These quibbles make some sense in the brain-rot context of social media, where people have been conditioned to think it’s normal to have interactions with millions of strangers at the same time, but this is not really tenable or healthy. Nor is it something most people would tolerate in the physical world. If a billionaire bought one of your local haunts, renamed it, humiliated the employees, brought back many of the people who’d been banned for harassing other regulars, eliminated basic rules of decency, started having town halls with Republicans and a leader of the AfD, taking your business elsewhere would be perfectly rational. This is essentially what’s happened on X, only the reality is wildly, at times comically, more extreme. A critical mass of the nation’s politicians, news outlets, and major brands regularly post content for free to the exclusive streaming platform for the Ye song “Heil Hitler.” This platform is owned by the world’s richest man, a conspiracy theorizing GOP mega-donor who still holds a position in the Trump administration. Even if he winds down his official role, X will remain an instrument for Musk’s politics. Let’s pause to sit with the absurdity of these facts.

Acknowledging the role X plays in mainstreaming the worst constituencies makes for awkward conversations with those who continue to use it. These discussions grow exhausting, fast. There’s a definite purity-politics flavor to any suggestion that people should take a moral stand and leave a social network, but also a pretty airtight case to be made for boycotting it. There is no ethical consumption under tech oligarchy, etc. You’re not a Nazi simply because you use X—but also, what exactly are you doing there?

You may not have any interest in participating in a culture war. The problem is that on X, everything is a culture war. Culture war is the very point of the MAGA AI slop the platform traffics in and the viscerally cruel White House X account. Culture war is behind Tucker Carlson’s choice to debut his post-Fox show on X and why Alex Jones livestreams on the platform every day. West’s nihilistic neo-Nazi single is an act of culture war: Its message isn’t just that X has energized his ideas, but that the platform renders people like Ye unignorable. Only Musk could shut this machine down, but plenty of others lend it their credibility and happily turn the cranks, ensuring that the culture war grinds on and on.

In the consumer race, Anthropic’s Claude has seen less adoption than OpenAI [File]
| Photo Credit: REUTERS

Artificial intelligence developer Anthropic is making about $3 billion in annualised revenue, according to two sources familiar with the matter, in an early validation of generative AI use in the business world.

The milestone, which projects the company’s current sales over the course of a year, is a significant jump from December 2024 when the metric was nearly $1 billion, the sources said. The figure crossed $2 billion around the end of March, and at May’s end it hit $3 billion, one of the sources said.

While consumers have embraced rival OpenAI’s ChatGPT, a number of enterprises have limited their rollouts to experimentation, despite board-level interest in AI. Anthropic’s revenue surge, largely from selling AI models as a service to other companies, is a data point showing how business demand is growing, one of the sources said.

A key driver is code generation. The San Francisco-based startup, backed by Google parent Alphabet and Amazon.com, is famous for AI that excels at computer programming. Products in the so-called codegen space have experienced major growth and adoption in recent months, often drawing on Anthropic’s models.

This demand is setting Anthropic apart among software-as-a-service vendors. Its single-quarter revenue increases would count Anthropic as the fastest-growing SaaS company that at least one venture capitalist has ever seen.

“We’ve looked at the IPOs of over 200 public software companies, and this growth rate has never happened,” said Meritech General Partner Alex Clayton, who is not an Anthropic investor and has no inside knowledge of its sales.

He cautioned that these comparisons are not fully precise, since Anthropic also has consumer revenue via subscriptions to its Claude chatbot. Still, by contrast, publicly traded SaaS company Snowflake took six quarters to go from $1 billion to $2 billion in such run-rate revenue, Clayton said.

Anthropic competitor OpenAI has projected it will end 2025 with more than $12 billion in total revenue, up from $3.7 billion last year, three people familiar with the matter said. This total revenue is different from an estimated annualized figure like Anthropic’s. Reuters could not determine this metric for OpenAI.

The two rivals appear to be establishing their own swim lanes. While both offer enterprise and consumer products, OpenAI is shaping up to be a consumer-oriented company, and the majority of its revenue comes from subscriptions to its ChatGPT chatbot, OpenAI Chief Financial Officer Sarah Friar told Bloomberg late last year.

OpenAI has not reported enterprise-specific revenue but said in May that paying seats for its ChatGPT enterprise product have grown to 3 million, from 2 million in February, and that T-Mobile and Morgan Stanley are among its enterprise customers.

In the consumer race, Anthropic’s Claude has seen less adoption than OpenAI. Claude’s traffic, a proxy for consumer interest, was about 2% of ChatGPT’s in April, according to Web analytics firm Similarweb. Anthropic, founded in 2021 by a team that departed OpenAI over differences in vision, closed a $3.5 billion fundraise earlier this year. That valued the company at $61.4 billion. OpenAI is currently valued at $300 billion.

The story so far: Foreign students hoping to study in the U.S. may soon face increased scrutiny as the Trump administration is considering taking measures to collect more information from the social media accounts of foreign student visa applicants.

Ahead of these changes, reports emerged that the scheduling of student visa interviews have been put on hold as officials worldwide await further guidance. This triggered panic among students worried that their education in the U.S. could be delayed or blocked.

What is the new social media vetting rule for U.S. visa applicants?

The U.S. Department of State defines a social media identifier/handle as “any name used by the individual on social media platforms including, but not limited to, Facebook, Twitter, and Instagram.”

A cable in late May that was signed by U.S. Secretary of State Marco Rubio and shared by the Politico outlet, ordered a temporary halt to scheduling visa appointments.

“Effective immediately, in preparation for an expansion of required social media screening and vetting, consular sections should not add any additional student or exchange visitor (F, M, and J) visa appointment capacity until further guidance is issued septel, which we anticipate in the coming days,” Mr. Rubio’s cable quoted by Politico stated, referring to telegram guidance that would be issued separately.

The collection of visa applicants’ social media handles is not a new measure. The U.S. Department of State noted that it updated its non-immigrant visa online application form (DS-160), the paper back-up version of the non-immigrant visa application (DS-156), and the online immigrant visa application form (DS-260), to collect applicants’ social media identifiers. The changes were implemented from May 2019, citing U.S. national security as a top priority.

However, Mr. Rubio’s latest cable suggests closer scrutiny of visa applicants’ social media posts in the future, and especially if they are foreign students. Foreign visa applicants heading to Harvard University may also face this treatment.

Why does the Trump administration want to vet social media handles?

The new updates come as the U.S. government seeks to clamp down on American universities where students are participating in anti-Israel or pro-Palestinian protests.

Because Israel is a U.S. ally, the administration has deemed many anti-Israel demonstrations and forms of protest to be a form of antisemitism (anti-Jewish bigotry) aimed at Jewish Americans, rather than a form of anti-Zionism (opposition to Israel’s statehood and violence against Palestinians). Many pro-Palestinian protesters on university campuses have been termed as Hamas supporters by conservative Americans.

A White House factsheet published this January saw Trump stating, “To all the resident aliens who joined in the pro-jihadist protests, we put you on notice: come 2025, we will find you, and we will deport you. I will also quickly cancel the student visas of all Hamas sympathizers on college campuses, which have been infested with radicalism like never before.”

The U.S. Department of State’s ‘Online Nonimmigrant Visa Application’ (DS-160) form has a sub-section dedicated to the applicant’s various social media handles that they can choose from a menu.

A screenshot shows how certain visa applicants are asked to list their social media handles
| Photo Credit:
U.S. Department of State website

There is a separate section for other platforms/websites that the applicants wish to flag or identify independently.

Visa applicants are also allowed to volunteer other social media handles and websites
| Photo Credit:
U.S. Department of State website

The social media platforms already mentioned on the DS-160 form include names such as Facebook, Twitter, Instagram, and YouTube, as well as Chinese options such as Douban, Qzone, Weibo, and others. VKontakte is a Russian social media platform. The Trump-endorsed Truth Social, however, is not on this list. Vine, meanwhile, has shut down.

Visa applicants are being asked to provide social media handles they have used in the past five years.

One point of focus appears to be Chinese immigration. In a recent press statement titled, ‘New Visa Policies Put America First, Not China,’ Mr. Rubio confirmed that government bodies would start “aggressively” revoking visas belonging to Chinese students, “including those with connections to the Chinese Communist Party or studying in critical fields.”

What can foreign students do to safeguard their visa application?

First and foremost, carry out a social media audit of your own. This means making a list of all your accounts to make sure no key platforms are left out. Delete — not deactivate — inactive accounts or nearly forgotten handles you have not accessed in more than five years, to avoid confusion. Carry out this step as early as you can, because some social media companies keep even deleted accounts visible for 30-90 days to prevent fraudulent activity.

Next, compile a list of core social media identifiers — your existing usernames/handles — that you will submit as part of your visa application. Never share your passwords with others.

You can then choose whether to lock your social media accounts or not. While it might be natural for minors, privacy-minded users, or non-professional visa applicants to have private accounts, it will arouse suspicion if a major influencer or public figure with thousands of followers suddenly locks their official account before their visa application. Make the choice that fits your privacy requirements.

Whether your accounts are public or private, be sure to vet your own social media content. Be on the lookout for content you posted in the past that is outdated, inaccurate, offensive, abusive, or glorifies criminal activity even as a joke. Also check your replies, ‘likes’, and uploaded media for any content that could hurt your visa application. Take note of accounts you follow or interact with to ensure that they are in compliance with international laws.

You may also want to check content that does not strictly count as social media activity, such as your YouTube comment history.

You may feel strongly about potentially taking down content related to social causes that are important to your heart. Make an informed choice that aligns with your personal position.

Foreign students hoping to obtain a visa to study in the U.S. should stay in touch with their universities and international admissions officers in order to learn the latest updates.

NEW DELHI: Foreign portfolio investments in Indian markets reached record levels in May 2025, as confirmed by National Securities Depository Ltd (NSDL) statistics, quoted by ANI. The month recorded net FPI inflows of Rs 19,860 crore, establishing May as the strongest month for foreign investments in 2025.The period from May 26 to May 30 saw foreign investors maintain their investment momentum with net inflows of Rs 6,024.77 crore. While positive inflows characterised most trading days that week, Friday registered a net outflow of Rs 1,758.23 crore.Although May demonstrated robust performance, the cumulative FPI investment for 2025 remains negative. The period from January through May shows net outflows of Rs 92,491 crore. Nevertheless, the substantial May inflows suggest a possible shift in foreign investor confidence.The uptick in FPI activity correlates with the declining US dollar value and positive developments in the Indian stock market.The robust economic foundations of India continue drawing international investors, although FPI flows remain responsive to international circumstances and external challenges.Whilst the year commenced cautiously, the positive May figures might indicate a directional change, provided global conditions maintain stability.Earlier data indicated FPI stock sales of Rs 3,973 crore in March. January and February witnessed equity sales of Rs 78,027 crore and Rs 34,574 crore, respectively.The final trading day of May saw the Indian stock market close marginally lower, influenced by varied global indicators. The Sensex decreased by 182 points (0.22 per cent) to 81,451.01, whilst the Nifty 50 settled at 24,750.70, down 83 points (0.33 per cent).

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

Sequoia Capital is hoping to cash in on rising global demand for art house cinema after leading a $100mn investment in Mubi that values the indie streaming service at $1bn, marking a rare foray into film and media for the Silicon Valley venture capital firm.

The cash from Sequoia, best known for its early bets on high-growth tech companies including Google, Apple and YouTube, will help Mubi build out its global distribution network and invest in new films.

The fundraising marks a vote of confidence for the 18-year-old company, which has recently undergone a big transformation from niche indie streamer to Hollywood player.

Andrew Reed, a partner at Sequoia who led the investment, said the firm was banking on untapped global demand for highbrow films, including those being bankrolled by Mubi.

“We’re a Silicon Valley investor, so the question is, how many people around the world will projects like this resonate with? Our opinion is that this is going to resonate with a lot more people than anybody thinks.”

The Sequoia deal gives London-based Mubi financial firepower when the broader movie industry — a risky business at the best of times — is reeling from the disruptive effects of the streaming revolution. It also follows a series of big swings by Mubi, which this month beat out Apple and Netflix to acquire the rights to Lynne Ramsay’s Die, My Love starring Jennifer Lawrence for $24mn, in the biggest deal to come out of the Cannes Film Festival.

Mubi’s streaming service features a wide range of international films from Latin America, Europe and Asia alongside American indies. It has about 20mn registered users worldwide, with US subscribers paying $14.99 a month, but also distributes independent films in cinemas in the UK, US, Canada, Latin America, Germany and, starting this year, Italy. It now has offices in 15 countries with more than 400 employees.

“Mubi is high-growth and profitable, and in the film business that’s an impossibly rare combination,” said Reed.

The company established itself as a serious player in Hollywood last year following founder Efe Cakarel’s big bet on The Substance, starring Demi Moore and Margaret Qualley, which he bought for $12mn and went on to earn $84mn at the box office.

The Substance won an Oscar, a Golden Globe and a Bafta. “We took this body-horror film from a director that no one ever heard of and made it a cultural global phenomenon because of word of mouth,” Cakarel told the Financial Times.

Cakarel said he was not aiming to challenge Netflix or Apple, despite recent successes. “We are not really competing with them — we have a completely different business,” he said, adding that Mubi’s goal was to “elevate great cinema, both classic and contemporary”.

The company is planning more cinematic releases this year, including its first original film — director Kelly Reichardt’s The Mastermind — as well as Paolo Sorrentino’s Grace and Jim Jarmusch’s Father, Mother, Sister, Brother.

Despite a sagging overall box office, Cakarel said he believed there was a healthy global audience for such films, noting that seven of the nominees for the Best Picture Oscar this year were indies, including the winner, Aoralong with The Brutalist and The Substance.

“If you look at the box office for speciality films, it is growing year over year, very substantially. You look at the biggest 200 films over the past three years in the box office, 69 of them [were] speciality films. So we are not talking about some small niche here.”

Sequoia joins other outside investors in Mubi, including Summit Partners and Chinese billionaire Zhang Xin, who invested in the company last year through her New York-based film company Closer Media. She also sits on the company’s board.