SpaceX, the aerospace company founded by Elon Musk 24 years ago, has finally made its IPO filing public. And once the company goes public, Musk will be at its center as CEO, CTO, and Chairman of the board.

The hefty filing, posted after markets closed Wednesday, shows a company that has developed far beyond its initial pursuit of reusable rockets — although its long-term mission to create a multi-planetary species remains intact. SpaceX is now a technology conglomerate working on satellites and AI, and has become one of the world’s most valuable private companies.

When it goes public later this year on the Nasdaq exchange, it will become one of the most valuable publicly-traded companies. (Nvidia currently holds the crown with a market cap of $5.4 trillion.) SpaceX has chosen the ticker “SPCX” for the listing.

The regulatory filing, known as an S-1, offers the most vivid and financially illuminating public dissection of SpaceX’s business to date. And it comes just weeks ahead of what’s expected to be the largest IPO ever, both in terms of potential money raised (expected to be around $75 billion) and overall valuation (reportedly $1.75 trillion). It contains 36 pages of risk factors to SpaceX’s business, and details legal fights it faces following the absorption of Musk’s artificial intelligence and social media companies — battles SpaceX says will likely cost it $530 million.

Many of the headline details have been reported in the weeks since SpaceX first submitted a confidential version of its S-1 filing to the Securities and Exchange Commission on April 1. The company lost about $4.9 billion in 2025 on revenue of more than $18 billion, as Reuters reported last month.

The filing details a business that is currently dominated by SpaceX’s Starlink satellite internet offering, which generated more than half of the company’s revenue last year — around $11 billion. It also shows how much SpaceX has burned to get to this point: more than $37 billion lost since inception, according to the S-1.

XAI, the artificial intelligence company Elon Musk created and recently merged into SpaceX, is not helping on that front. The filing shows SpaceX directed around 60% of its capital spending in 2025 to its AI division, or around $20 billion. And yet that division — which houses the chatbot Grok — lost billions last year, and only grew revenue by about 22%. That’s far below the reported revenue growth rates at frontier AI labs.

But the company is, of course, making a lot of astronomical promises in the filing. One of the biggest? That it has “identified the largest actionable total addressable market in human history” of $28.5 trillion. The company attributes an enormous portion of that — $22.7 trillion — to “enterprise applications” of AI.

It’s all about the rocket

Despite SpaceX’s complex business, much of its future is pegged to the success of Starship, the fully reusable heavy-lift rocket that has had a series of explosions and technical revamps over the past several years. The company is expected to conduct the 12th launch of Starship as early as this week, and much is riding on its success.

SpaceX said in the filing that it expects Starship to begin payload delivery to orbit in the second half of 2026, leaving little room for error. Assuming SpaceX can hit that milestone, the company plans to begin using Starship to send its Starlink broadband satellites into orbit in the second half of 2026 and its next-generation V2 mobile satellites in 2027.

SpaceX’s plans for Starship stretch far beyond satellite launches. The company wants to use the rapidly reusable spacecraft, which is designed to deliver 100 metric tons to Earth’s orbit, for Mars exploration and to launch orbital AI data centers into space.

Pushing toward that goal has been costly for SpaceX, the S-1 filing shows. The space segment of the company invested heavily in research and development for the Starship program, spending $3 billion in 2025 and $930 million in the first quarter of 2026.

The cost is worth it, in SpaceX’s view. The company said Starship is critical to reducing the cost of reaching orbit by 99% or more relative to the historical average launch cost.

Starry-eyed visions

The S-1 details SpaceX’s many extreme goals, like making life multi-planetary, reaching the moon and Mars, and building orbital networks of satellites that can do space-based computing.

But there are other flashy, futuristic ideas included in the filing, too.

SpaceX is apparently still interested in using its Starship rocket as a terrestrial transportation system — an idea Musk first proposed in 2017. The company says it plans to “develop ultra-fast long-haul point-to-point Earth transport using Starship, enabling passengers and cargo to travel between major cities in a fraction of current transit times, revolutionizing global logistics and passenger travel with unprecedented speed and efficiency.”

The company caveats this idea as a “future market,” so it’s not anywhere in the near-term picture. As a result, the merits and risks of the idea of point-to-point travel doesn’t get the same kind of scrutiny in the filing as SpaceX’s core business.

Another “future market” listed is “space tourism.” SpaceX has flirted with this in the past, allowing private citizens to fly to space on its Dragon spacecraft. It also once planned a mission around the moon with Japanese billionaire Yusaku Maezawa, but that got canceled long before it could happen. In the filing, SpaceX says it expects “increasing interest in human space travel as it becomes easier and more common to access space.”

SpaceX executives also believe that the company will one day enable manufacturing facilities in-orbit and on the moon and Mars.

“We aim to establish in-space manufacturing facilities that leverage the unique microgravity conditions of space to produce materials, pharmaceuticals, and advanced components that are difficult or impossible to manufacture on Earth, opening new high-value industrial markets,” the filing reads. Facilities on the moon and Mars would be focused on producing fuel, construction materials, and other “essential resources,” along with solar energy production.

Lastly, SpaceX believes it could one day get involved in asteroid mining operations. As it is listed as another “future market,” there is little detail about how SpaceX plans to tackle this idea.

Total control

Make no mistake, this is Elon Musk’s company. Per the filing, Musk will be the the CEO, CTO, and Chairman of the SpaceX board after the IPO.

The S-1 shows he owns 93.6% of SpaceX’s Class B stock, which comes with 10 votes per share. Musk is therefore currently sitting on 85.1% of the voting power at SpaceX. That number is expected to drop following the IPO but will stay above 50%, allowing SpaceX to dodge certain rules about having independent directors on its board.

He was also handed a new compensation package at the beginning of this year, which could net him as many as 1 billion shares of Class B stock if he hits increases the value of SpaceX to $7.5 trillion and the “establishment of a permanent human colony on Mars with at least one million inhabitants.” He stands to reap even more shares if the company is able to to create space-based data centers capable of delivering “100 terawatts of compute per year.”

Elon Musk’s xAI lost $6.4 billion from operations on just $3.2 billion in revenue in 2025, according to SpaceX’s IPO filings. And the losses are poised to grow. SpaceX’s filing reveals plans to scale Grok to “multiple trillions of parameters,” a dramatic boost that will likely require significant additional compute spend.

Elon Musk merged his AI company xAI — which had previously acquired his social media platform X (formerly Twitter) — with his rocket and satellite company SpaceX in February before announcing that he’d take the combined company public this year. While AI competitors OpenAI and Anthropic are also eyeing public debuts in 2026, SpaceX’s is expected to be one of the largest in history with a potential $1.75 trillion valuation.

The filing marks the first public glimpses into xAI, and therefore X’s, financials. In 2024, xAI recorded a loss of $1.56 billion on $2.62 billion in revenue. By 2025, losses had ballooned to $6.4 billion on $3.2 billion, meaning the gap between what xAI earns and spends is widening. Meanwhile, competitor (and customer) Anthropic reportedly expects a 130% revenue jump to $10.9 billion in the second quarter, leading to its first operating profit.

The jump in revenue from 2024 to 2025 came in large part from “AI solutions and infrastructure revenue” totalling $465 million, which includes $365 million in X and Grok subscription revenue, and $88 million in data licensing. An additional $116 million came from advertising.

AI segment capital expenditures climbed from $12.7 billion in 2025 to $7.7 billion in the first quarter of 2026 alone. That’s an annualized capex run rate of about $30.8 billion, more than doubling year-over-year.

So far, that investment has resulted in growing, but still limited, user numbers. Per the filing, SpaceX recorded 117 million monthly active users for Grok AI features as of March 2026, out of 550 million total MAUs across Grok and X combined. That implies only one-fifth of the combined ecosystem is actively using Grok AI features.

Still, SpaceX intends to soldier on with Grok; its next-generation AI is expected to scale to “multiple trillions of parameters,” which the filing describes as a “step change in reasoning in depth and overall intelligence.” It’s an ambitious target, and one that’s now recorded in the audited annals of SEC history.

It’s also a target that will undoubtedly require more investment. The SpaceX filing’s “use of proceeds” section mentions an “expansion of our AI compute infrastructure.” Per the filing, xAI’s Colossus and Colossus II data centers — both of which came online in 122 days and 91 days, respectively — collectively provide about 1 gigawatt of compute power. These are both used for Grok’s training and inference. SpaceX claims that owning the compute infrastructure and vertically integrating across the AI stack lets them “train and iterate frontier models at lower cost and higher velocity.”

Another way that SpaceX might assuage investor fears about spending is by performing training and inference on orbital data centers, which Musk has promised to be a much cheaper alternative to terrestrial data centers. That sci-fi vision isn’t likely to happen for several years, though. The filing says SpaceX intends to begin deploying its orbital AI compute satellites as early as 2028 — the first concrete timeline set for such a launch.

“The future of AI will be determined by control of the physical stack,” the filing reads.

Nvidia announced another record revenue figure after market close on Wednesday, reporting financial results for the quarter ending April 26. Over those three months, the company brought in $81.6 billion in revenue (up 20% from the previous quarter) and a record $75.2 billion in data center revenue. On the strength of that revenue, the company is authorizing $80 billion in share repurchases.

“Our Blackwell architecture is everywhere, adopted and deployed by every major hyperscaler, every cloud provider, and every major model maker,” said Nvidia CFO Colette Kress.

Notably, Nvidia did project a slowdown in revenue growth, forecasting $91 billion in revenue for the next quarter, which will be 12% growth.

Chinese exports did not make any significant impact on the company’s earnings. While H200s have been approved for US export, “we have yet to generate any revenue, and we are uncertain whether any imports will be allowed into [China],” Kress said.

One surprise was the sheer volume of Nvidia’s stakes in privately held companies (listed in the filing as as “non-marketable equity securities”), which nearly doubled between January and April. The company began the quarter with $22 billion in privately held stakes, but ended with $43 billion, driven primarily by $18.5 billion in purchases over the course of the quarter. The previous quarter had seen only $649 million of equivalent purchases.

Notably, that figure does not include Nvidia’s recent investment in publicly traded companies like Corning and IREN, nor does it reflect future commitments that have not yet closed. Notably, Nvidia committed to investing $30 billion in OpenAI in February, although the precise structure of the deal was not disclosed.

On a call discussing the results, Jensen Huang emphasized the broad scope of Nvidia’s impact, including a pending buildout with Anthropic. “The amount of capacity we’re going to bring online for Anthropic this year and next year is going to be quite significant,” Huang told investors on a call. “Our coverage for Anthropic had been largely zero until this.”

Elon Musk’s xAI has gotten itself in hot water over its use of polluting generators at its data center near Memphis, Tennessee. Now it wants to buy even more of them.

In SpaceX’s IPO filing, released Wednesday, the company said its xAI division will buy another $2.8 billion worth of turbines for its AI infrastructure over the next three years. One deal, worth $2 billion, is specifically for “mobile gas turbines,” the kind that it’s currently being sued over.

The NAACP filed a lawsuit against xAI last month for operating dozens of unregulated gas turbines that worsen the air quality in one of the most polluted parts of the country. The organization has sought an injunction against xAI’s use of the turbines. So far, xAI has been granted permits for 15 turbines. As of a few weeks ago, it was using 46.

Each of the types of turbines xAI is operating have the potential to emit more than 2,000 tons of NO x pollution annually, a group of chemicals that contributes to asthma-inducing smog.

The company claims that it can operate the turbines for up to a year without permits because they are “mobile” — that is, they’re still on the trailer they were shipped on. The company appears to be exploiting a discrepancy between state and federal interpretations. Mississippi claims it doesn’t need to permit mobile generators.

But federal regulations say that turbines of that size, even if they’re on a trailer, are still subject to air-pollution regulations. The EPA ruled earlier this year that xAI was operating the turbines in violation of federal law.

SpaceX acknowledges the risks in its IPO filing. “We currently rely significantly on natural gas and gas turbine technology to power our data center operations,” it wrote. Injunctions or rescinded permits “would adversely affect our AI business.”

In Brief

Earlier this month, Anthropic surprised the AI world with a deal to buy 300 megawatts’ worth of compute — securing the entire output of the Colossus 1 data center near Memphis, Tennessee.

Turns out, compute at that scale isn’t cheap. Anthropic will be paying xAI $1.25 billion per month through May 2029, with a discounted rate for the first two months as xAI completes its ramp-up. All told, the deal could bring xAI over $40 billion in revenue.

Details of the transaction emerged from SpaceX’s S-1 filing with the SEC. The deal, the company said, “allows us to monetize unused compute capacity in our infrastructure.” The terms of the deal allow either side to terminate the contract with 90 days’ notice.

“We expect to enter into additional similar services contracts,” the filing stated.

The move has given xAI a hybrid stance in the AI market. Most players either build data centers for themselves or build data centers for others to use — rarely both simultaneously. This emerging model, sometimes called a “neocloud,” lets AI companies offset infrastructure costs by acting as a cloud provider when their own usage falls short of capacity.

SpaceX argues the arrangement is a savvy use of resources. “We believe our dual monetization strategy provides multiple pathways to generate returns on invested capital,” it wrote. But the subtext is hard to miss: xAI appears to have overbuilt its compute capacity and needed to find a way to monetize it ahead of a public offering. Usage of Grok — xAI’s flagship AI assistant — has dropped significantly in recent months, freeing up servers that the company is now selling to one of its closest competitors.

During a Y Combinator event on Tuesday night, Sam Altman had what YC partner Tyler Bosmeny called a “mic drop moment.” Altman offered $2 million worth of OpenAI tokens to every startup in the current class in exchange for equity in the startup.

In other words, he promised that OpenAI would invest in the whole class, not with cash but with an allotment of AI tokens that startups can use to build their products.

i am excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build.



openai offered to invest $2M in tokens into every startup in the current yc batch.



happy building! https://t.co/YSHYJoutuf — Sam Altman (@sama) May 20, 2026

Y Combinator has about 169 startups in this cohort, according to its directory.

As for how much equity each startup can expect to give up, that can’t be determined at the time it signs the deal. It will depend on how much the startup is worth when it raises its first priced round — a funding round in which investors assign the company a formal valuation.

Y Combinator managing director Jared Friedman tells TechCrunch that the deal will be offered as an “uncapped SAFE,” meaning, “it will convert in the next priced round, which is typically the Series A,” he said.

A SAFE is YC’s standard agreement structure for its early-stage companies that raise money before their first “priced” rounds with valuations involved. An uncapped SAFE doesn’t set a ceiling on that valuation, which can benefit founders because the higher the valuation at conversion, the smaller the slice of the company the investor receives.

We’ve seen some discussion on X that this deal could amount to OpenAI holding about 2% equity should a startup hit a $100 million valuation, though without seeing the actual terms, we can’t verify that.

For OpenAI, the deal works on two levels. Obviously, it gains equity in this crop of early-stage companies, meaning it profits if they succeed. But it also encourages them to build their business on and with OpenAI. Whether this locks them in for the long term or not, it does mean that they won’t default to OpenAI’s competitors, like Anthropic’s Claude Code.

The tokens themselves may sweeten the deal further: As inference costs continue to fall, what OpenAI is giving away today could cost it very little to produce tomorrow — making the equity it receives in return look increasingly cheap.

Unsurprisingly, there’s already plenty of commentary on X on why this is, and isn’t a good deal for startups.

The pro-deal folks believe the deal helps startups eliminate one of their biggest costs — AI infrastructure bills, which can spiral fast and consume a disproportionate share of an early-stage startup’s budget at a time when money, typically, is already scarce.

The buyer-beware folks have other warnings. Seed investor Jason Calacanis — who has his own competing accelerator and fund — went for the be-afraid-of-Big-Tech warning.

“If you take these tokens, there’s a non-zero chance that OpenAI will study exactly what your startup is doing, copy your idea and put your app into their free offering. This is the classic platform playbook — be careful, founders!” he posted.

The fear that OpenAI and Anthropic could swallow every good AI startup idea is real.

The truth is, should OpenAI want to do that, it can, even when startups simply pay OpenAI for the tokens. By taking an equity stake, OpenAI may have more incentive for the startup’s success, not less.

Plus, as the former head of Y Combinator and a recurring guest speaker, Altman has as much access to every cohort and its ideas as he wants, deal or not.

The bigger question for this YC batch is whether a budget of tokens from a single AI player is worth giving up additional equity. Y Combinator already takes a 7% stake for a $500,000 cash investment in its standard deal. In exchange, startups get access to YC’s powerful Silicon Valley network of VCs, potential customers, and other founders.

But equity is also precious for startups. Seed investors frequently take 20% or so, too. And startups need equity as compensation for their early employees.

The bigger danger is that a startup will blow through its OpenAI token budget without enough to show for it, having surrendered equity in the process. Still, that may be better than paying for the tokens with cash, an even scarcer resource at that stage.

SpaceX, the aerospace company founded by Elon Musk 24 years ago, has finally made its IPO filing public.

The hefty filing, posted after markets closed Wednesday, shows a company that has developed far beyond its initial pursuit of reusable rockets — although its long-term mission to create a multi-planetary species remains intact. SpaceX is now a technology conglomerate working on satellites and AI, and has become one of the world’s most valuable private companies.

When it goes public later this year on the Nasdaq exchange, it will become one of the most valuable publicly-traded companies. (Nvidia currently holds the crown with a market cap of $5.4 trillion.) SpaceX has chosen the ticker “SPCX” for the listing.

The regulatory filing, known as an S-1, offers the most vivid and financially illuminating public dissection of SpaceX’s business to date. And it comes just weeks ahead of what’s expected to be the largest IPO ever, both in terms of potential money raised (expected to be around $75 billion) and overall valuation (reportedly $1.75 trillion). It contains 36 pages of risk factors to SpaceX’s business, and details legal fights it faces following the absorption of Musk’s artificial intelligence and social media companies — battles SpaceX says will likely cost it $530 million.

Many of the headline details have been reported in the weeks since SpaceX first submitted a confidential version of its S-1 filing to the Securities and Exchange Commission on April 1. The company lost about $4.9 billion in 2025 on revenue of more than $18 billion, as Reuters reported last month.

The filing details a business that is currently dominated by SpaceX’s Starlink satellite internet offering, which generated more than half of the company’s revenue last year — around $11 billion. It also shows how much SpaceX has burned to get to this point: more than $37 billion lost since inception, according to the S-1.

XAI, the artificial intelligence company Elon Musk created and recently merged into SpaceX, is not helping on that front. The filing shows SpaceX directed around 60% of its capital spending in 2025 to its AI division, or around $20 billion. And yet that division — which houses the chatbot Grok — lost billions last year, and only grew revenue by about 22%. That’s far below the reported revenue growth rates at frontier AI labs.

But the company is, of course, making a lot of astronomical promises in the filing. One of the biggest? That it has “identified the largest actionable total addressable market in human history” of $28.5 trillion. The company attributes an enormous portion of that — $22.7 trillion — to “enterprise applications” of AI.

It’s all about the rocket

Despite SpaceX’s complex business, much of its future is pegged to the success of Starship, the fully reusable heavy-lift rocket that has had a series of explosions and technical revamps over the past several years. The company is expected to conduct the 12th launch of Starship as early as this week, and much is riding on its success.

SpaceX said in the filing that it expects Starship to begin payload delivery to orbit in the second half of 2026, leaving little room for error. Assuming SpaceX can hit that milestone, the company plans to begin using Starship to send its Starlink broadband satellites into orbit in the second half of 2026 and its next-generation V2 mobile satellites in 2027.

SpaceX’s plans for Starship stretch far beyond satellite launches. The company wants to use the rapidly reusable spacecraft, which is designed to deliver 100 metric tons to Earth’s orbit, for Mars exploration and to launch orbital AI data centers into space.

Pushing toward that goal has been costly for SpaceX, the S-1 filing shows. The space segment of the company invested heavily in research and development for the Starship program, spending $3 billion in 2025 and $930 million in the first quarter of 2026.

The cost is worth it, in SpaceX’s view. The company said Starship is critical to reducing the cost of reaching orbit by 99% or more relative to the historical average launch cost.

Starry-eyed visions

The S-1 details SpaceX’s many extreme goals, like making life multi-planetary, reaching the moon and Mars, and building orbital networks of satellites that can do space-based computing.

But there are other flashy, futuristic ideas included in the filing, too.

SpaceX is apparently still interested in using its Starship rocket as a terrestrial transportation system — an idea Musk first proposed in 2017. The company says it plans to “develop ultra-fast long-haul point-to-point Earth transport using Starship, enabling passengers and cargo to travel between major cities in a fraction of current transit times, revolutionizing global logistics and passenger travel with unprecedented speed and efficiency.”

The company caveats this idea as a “future market,” so it’s not anywhere in the near-term picture. As a result, the merits and risks of the idea of point-to-point travel doesn’t get the same kind of scrutiny in the filing as SpaceX’s core business.

Another “future market” listed is “space tourism.” SpaceX has flirted with this in the past, allowing private citizens to fly to space on its Dragon spacecraft. It also once planned a mission around the moon with Japanese billionaire Yusaku Maezawa, but that got canceled long before it could happen. In the filing, SpaceX says it expects “increasing interest in human space travel as it becomes easier and more common to access space.”

SpaceX executives also believe that the company will one day enable manufacturing facilities in-orbit and on the moon and Mars.

“We aim to establish in-space manufacturing facilities that leverage the unique microgravity conditions of space to produce materials, pharmaceuticals, and advanced components that are difficult or impossible to manufacture on Earth, opening new high-value industrial markets,” the filing reads. Facilities on the moon and Mars would be focused on producing fuel, construction materials, and other “essential resources,” along with solar energy production.

Lastly, SpaceX believes it could one day get involved in asteroid mining operations. As it is listed as another “future market,” there is little detail about how SpaceX plans to tackle this idea.

Total control

Make no mistake, this is Elon Musk’s company.

The S-1 shows he owns 93.6% of SpaceX’s Class B stock, which comes with 10 votes per share. Musk is therefore currently sitting on 85.1% of the voting power at SpaceX. That number is expected to drop following the IPO but will stay above 50%, allowing SpaceX to dodge certain rules about having independent directors on its board.

He was also handed a new compensation package at the beginning of this year, which could net him as many as 1 billion shares of Class B stock if he hits increases the value of SpaceX to $7.5 trillion and the “establishment of a permanent human colony on Mars with at least one million inhabitants.” He stands to reap even more shares if the company is able to to create space-based data centers capable of delivering “100 terawatts of compute per year.”

Microsoft is purchasing 650,000 metric tons of carbon-removal credits from startup BioCirc, the company said today.

As carbon-removal deals go, it’s not a big buy. But this one is notable because last month, two reports said the tech giant was pausing its carbon-removal deals. BioCirc confirmed for TechCrunch that the purchase agreement was signed in May, weeks after Microsoft reportedly paused new deals.

For the carbon-removal industry — and the startups that depend on it — there’s a big difference between a pause and a recalibration. Microsoft is reportedly responsible for more than 90% of the carbon-removal credit market, meaning its purchasing decisions alone can determine whether young companies in the space survive.

Microsoft repeatedly denied that it had paused its carbon-removal purchases. “Our carbon removal program has not ended,” Melanie Nakagawa, chief sustainability officer at Microsoft, told TechCrunch in a statement. “At times we may adjust the pace or volume of our carbon removal procurement as we continue to refine our approach toward sustainability goals.”

The new deal generates carbon-removal credits from five BioCirc biogas projects. The biogas plants take biomass waste — frequently from agriculture — and use industrial bioreactors to turn it into methane and carbon dioxide. BioCirc captures the carbon dioxide and stores it in an underground reservoir offshore. The methane is then burned in a power plant.

Microsoft’s sustainability goals have been strained by the company’s push into AI. To power its data centers in Texas, Microsoft last month said it was working with Chevron and Engine No. 1 to build a natural gas power plant in the state that could eventually generate 5 gigawatts of electricity. Emissions from that project alone promise to dwarf the deal with BioCirc.

Internally, Microsoft employees have also been debating whether to abandon the company’s goal of matching zero-emissions electricity with its energy use on an hourly basis. Today, the company matches on an annual basis. That approach gives Microsoft more flexibility to, say, use more natural gas to power its data centers at night, but it also makes the company’s clean energy claims harder to verify.

If Microsoft continues to pursue fossil fuel power plants, it’ll need to ramp up its carbon-removal purchases to meet its 2030 target of becoming a carbon-negative company (one that removes more greenhouse gases from the atmosphere than it generates).

Last year, Microsoft signed several deals worth millions of tons of carbon-removal credits. The program’s reported pause set off alarm bells throughout the carbon-removal industry, which is still in its infancy.

The new deal suggests that Microsoft is, in fact, recalibrating its carbon-removal program — not abandoning it. Whether that remains true as AI drives its energy consumption higher is something the industry will be watching.

OpenAI claims its new reasoning model has produced an original mathematical proof disproving a famous unsolved conjecture in geometry, which was first posed by Paul Erdős in 1946.

If this sounds familiar to you, it’s because this isn’t the first time OpenAI has made such a bold claim. Seven months ago, the AI giant’s former VP Kevin Weil posted on X: “GPT-5 found solutions to 10 (!) previously unsolved Erdős problems and made progress on 11 others.”

It turns out, GPT-5 didn’t actually solve those problems; it just found solutions that already existed in the literature.

Taunts from rivals like Yann LeCun and Google DeepMind CEO Demis Hassabis followed, and Weil promptly took down his premature post. Today, at least, it seems OpenAI didn’t make the same mistake twice. Alongside the announcement, the company published companion remarks in support of the disproof from mathematicians like Noga Alon, Melanie Wood, and Thomas Bloom, who maintains the Erdos Problems website, and previously called Weil’s post “a dramatic misrepresentation.”

“For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids,” OpenAI posted on X. “An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.”

The company said this marks “the first time AI has autonomously solved a prominent open problem central to a field of mathematics.” The proof, per OpenAI, came from a new general-purpose reasoning model, not a system specifically designed to solve math problems or even this problem in particular.

OpenAI says this is significant because it means AI systems are now more capable of holding together long, difficult chains of reasoning and connecting ideas across fields in ways researchers may not have previously explored. That has implications for biology, physics, engineering, and medicine.

“AI is helping us to more fully explore the cathedral of mathematics we have built over the centuries,” Bloom said in a statement. “What other unseen wonders are waiting in the wings?”

Industry insiders say the next big thing in AI is “proactive” systems: agents that can anticipate a user’s needs — and fulfill them — before the user even knows what those needs are.

One startup that’s looking to make headway in this area is IrisGo. The company, which closed a $2.8 million seed round led by Andrew Ng’s AI Fund earlier this year, is building a desktop companion for PCs that can learn about a user’s daily workflows and then automate them with limited to no human prompting.

IrisGo was co-founded by Jeffrey Lai, a former Apple engineer who helped to build the Chinese language version of Siri, the company’s automated assistant. (Somewhat slyly, Iris is Siri spelled backward.)

The core idea is simple: Show the program how to do something once, and it remembers that process for future automated use — no repeat instructions needed.

During a conversation with TechCrunch, Lai ran a demo, showing how his platform could learn to place a coffee order online. As I watched, IrisGo recorded the steps it took to select a latte from Philz Coffee (a popular Bay Area chain), fill out credit card information, and then hit purchase. Lai then asked IrisGo to repeat the order on its own; the agent dutifully complied.

Buying coffee, of course, is not really the point. Instead, the hope is that the system will automate a whole host of business-related tasks. Iris comes with a built-in “skills” library — things like email drafting, invoice processing, report building, document summarization, and many other ready-to-use automated workflows. At the same time, Iris learns from the user’s desktop behavior and automatically adds those tasks to its potential list of action items.

The application also includes a coding assistant — similar in concept to OpenAI’s Codex or Anthropic’s Claude Code — designed to assist developers as they go about their work.

“Our target audience is knowledge workers — white-collar companies. There’s a lot of repetitive tasks that those workers do every day,” Lai said, noting that, despite the high-octane power of today’s frontier models, AI-assisted office work can still feel incredibly manual and repetitive. The goal, he said, is to move away from that and toward a more fully autonomous workflow, where the human works on high-level conceptual work while agentic systems take care of all the clerical work in the background.

A particularly appealing feature of IrisGo is that it is designed to process a lot of data on-device, giving it stronger privacy protections than other applications that rely heavily on the cloud. Lai says that the system is still a hybrid architecture — meaning that larger, more complex tasks are ultimately processed through the cloud, although the company promises that cloud processing “only occurs when explicitly authorized by the user and uses end-to-end encryption.”

Part of the strategy for scaling Iris has been to garner credibility through association with prominent figures and organizations. Support from Ng — notably a co-founder of the formative deep learning research team Google Brain — has helped. Lai managed to set up a meeting with Ng through a shared connection: Both are alumni from Carnegie Mellon University. Lai and his co-founder demoed Iris during that meeting, and Ng’s AI Fund ultimately led the startup’s seed round. Nvidia and Google have also backed the company.

IrisGo recently launched the beta versions of its macOS and Windows apps, and the company is also currently pursuing deals with laptop companies to preinstall the app on new devices. It recently struck such a deal with Acer, and Lai said the hope is that the company can strike similar deals with other device makers soon.

It feels like short video clips from podcasts, songs, and movies are suddenly everywhere on social media right now, and that’s not an accident. Brands have realized that the format offers a highly cost-effective way to market products.

Brands and marketing agencies often outsource the process of finding the most compelling 30 to 90 seconds of a video, known as ‘clipping’, to independent creators. But managing these gig workers and determining exactly where to distribute those videos presents a massive operational challenge.

Clouted, a startup that went through the a16z’s Speedrun accelerator in 2024, is building the infrastructure to automatically handle both the distribution strategy and the logistics of the clipping process. The platform taps into a network of over 100,000 gig creators to edit clips, then uses AI to determine the best social media platform and target audience for promoting them.

Clouted co-founder and CEO Justin Banusing first applied the company’s technology to his personal passion: electronic music and festival production. As a longtime DJ, he used Clouted to promote and grow &Friends, a Manila-based electronic dance music and pop-culture festival that now draws over 20,000 people.

Clouted’s approach has caught investor interest. The startup just announced a $7 million seed round led by Slow Ventures, with participation from Gold House Ventures, Weekend Fund, Peak XV’s Surge, and others.

Unlike purely volume-driven marketing tools, Clouted doesn’t just chase high clip counts. Instead, its AI operates a continuous testing loop, experimenting with different formats and channel strategies to figure out what actually performs best. The practical effect is that each campaign makes the next one more targeted and efficient, as the system accumulates data on what works.

Clouted works somewhat like penetration testing for social media algorithms — a concept borrowed from cybersecurity, where researchers probe a system’s defenses by attempting to break them. Rather than looking for security flaws, Clouted’s AI and its network of creators test thousands of different clipping and distribution approaches to identify what triggers a piece of content to go viral.

“The result is that every campaign Clouted runs makes the next one faster, smarter, and more effective,” Banusing told TechCrunch. “The platform learns which formats win, which audiences convert, and which distribution channels compound over time.”

While Clouted competes directly with similar startups like Overlap AI in the automated clipping space, Banusing said he looks to larger marketing infrastructure players, specifically CreatorIQ and Hightouch, as the ultimate competition. Hightouch recently crossed $100 million in ARR, suggesting that the enterprise marketing infrastructure space is large and still expanding. That’s the market Clouted is ultimately building toward.

Airbnb’s core business was letting users book a room or a whole property owned by someone else. In most cases, the host would be a house or a property owner, and you would get a stay with basic services such as a bed, a bath, and a kitchen. The company is now introducing hotels to its platform to drive new revenue and offer more options for travelers to book stays.

The company has been testing out hotel listings for a few months and is officially adding a filter to search for them in its app. Initially, the travel tech company is partnering with boutique hotels in 20 cities, including New York, Paris, London, Madrid, Rome, and Singapore.

Image Credits:Airbnb

Users will see a pop-up to look at hotels if they are searching for a stay for one or two nights in a city. They can also choose to see only hotels by applying a particular filter in the search. Airbnb said that it is offering a price match guarantee and would refund the difference in app credits if you find the listing at a lower rate somewhere else.

“There are a few examples of the types of trips for which a hotel is probably better suited, such as last-minute bookings, one-night stays, and business trips. What we’ve done in the Airbnb app is that when we can see you’re searching for that kind of stay, we can recommend hotels without sending you to another part of the app or tapping on another tab,” Jud Coplan, VP of marketing at Airbnb, told TechCrunch on a call.

For Airbnb, this is also an opportunity to offer stays to travelers in locations like New York City and Singapore, where short-term stays are barred. That regulatory workaround is no small thing: it effectively expands Airbnb’s addressable market into cities that had legislated it out — some of the most valuable travel markets in the world.

Property booking is just one part of the equation now for Airbnb. The company also wants users to book experiences and services, which were launched last year, during their trip as well. In the last few months, the company has launched grocery delivery to properties and airport pickup services. With its summer launch, Airbnb is adding luggage storage in over 15,000 locations. Plus, it will launch car rental services this summer.

The moves reflects a broader convergence happening across travel tech: Uber, long a ride-hailing company, has been expanding into hotel and trip booking, while Airbnb is now effectively entering the transportation business. Both companies appear to be pursuing the same destination — a single app that owns the traveler’s entire journey.

Image Credits:Airbnb

Within experiences, the company is adding visits with local guides for 3,000 landmarks. It is also adding more than 2,500 food experiences in the mix. With these offerings, Airbnb will try to compete better with Viator or GetYourGuide.

Airbnb is redesigning the app to accommodate all the different parts, such as stays, experiences, and services. On the home page, users will see suggestions for all categories. But they can switch to individual tabs and look at a particular offering, too.

While the company is not launching any official loyalty program, it will start offering Airbnb credits for the first car rental booking or up to 15% credit on hotel bookings. Having these credits might nudge people to use the app more, potentially testing the waters for a long-term loyalty program.

But, where is AI?

Several travel companies have added AI-powered itinerary builders or trip-planning tools. Airbnb has stayed away from this trend. In its latest earnings call, Airbnb CEO Brian Chesky said that a chatbot as an interface is not suited for travel.

“I think that the way in which we think about AI is how it can be used to help people find what they’re looking for more effectively and efficiently. And you see with this release that we have AI spread across a number of features across the app on the guest side and the host side,” Coplan said.

The company is using AI in some other tools. With this release, it will let hosts add their address and let AI fill in details for easy listing creation. On the customer side, it is introducing category tags such as location, amenities, and family-friendliness that let you skim through reviews related to the tag.

Image Credits:Airbnb

The company is also adding a tool that compares properties in your wishlist and shows AI-generated summaries to give you information about which property you can pick.

The biggest area Airbnb is applying AI to is customer service. It rolled out the AI bot to the U.S. last year and is now eyeing global expansion with support for 11 languages. Chesky said during the Q1 2026 call that the chatbot handles 40% of its queries. The updated support will also have interactive cards, such as a way to update your trip or solve other issues.

Image Credits:Airbnb

Airbnb said that later this year, it will start using a voice-based AI assistant that users can talk to on a call. It said that it is working with multiple partners for this product, but didn’t specify any names.

All that doom and gloom about the state of the electric vehicle market? That’s just an American problem. The rest of the world can’t get enough EVs, according to a new report from the International Energy Agency.

EV sales surpassed 20 million units last year, capturing 25% of the global market. Growth was highest in China, and market share in other regions has also been picking up pace. In Latin America, for example, sales grew by 75%. Meanwhile, sales in the U.S. are stagnant, with EVs hovering around 10% market share.

The EV market has gone K-shaped, and automakers of all stripes — legacy and startup — had better pay attention.

Sales figures in the U.S. were held back last year by the One Big Beautiful Bill Act, which killed EV tax credits, along with policies that have prevented Chinese automakers from entering the market.

For startups like Rivian and Lucid, which are heavily invested in the U.S. market, it certainly makes for a more challenging road ahead. Legacy automakers are somewhat insulated since they can lean on more profitable fossil fuel vehicles — at least in the short term. But without a solid EV strategy, they stand to lose more global market share as consumer tastes and expectations shift.

Elsewhere, Chinese automakers have been driving the upper leg of the K higher. The growth has been most apparent in China, where nearly 55% of new vehicles were electric. Affordability helps: More than two-thirds of EVs sold in the country were cheaper than the average fossil fuel car.

Chinese automakers also helped drive EV sales higher in Southeast Asia, Latin America, and Europe. More than half of all EVs sold in Southeast Asia were made by a Chinese company, for example, while Europe imported over half a million Chinese EVs.

The stunning growth of EVs in Southeast Asia and Latin America punctures one prevailing theory that electric cars would be too expensive for developing economies. EV prices have been on par with internal combustion vehicles for the last two years in Thailand. “Imports of affordable electric cars from China have brought down prices and driven up EV sales in many emerging markets in recent years,” the IEA report said.

That may not last forever, though.

Chinese automakers exported more than 25% more vehicles than were bought in foreign markets. Dealers outside of China might resist accepting more EVs until they can sell what they have on hand. Plus, countries might begin to chafe at the flood of inexpensive Chinese cars and institute tariffs.

Even if that happens, it would be foolish to count Chinese brands out. The Communist Party has invested significant sums to turn its automotive industry into a powerhouse. As a result, the country has enough manufacturing capacity to fulfill 65% of global demand. Thanks to state support, Chinese automakers could produce an outsized number of vehicles far longer than other companies can remain solvent.

In the long run, though, EVs promise to undercut fossil fuel vehicles, even without subsidies. As early as next year, battery electric vehicles will be cheaper to make than internal combustion vehicles, according to Gartner.

The Trump administration is trying to steer the U.S. market back toward fossil fuels, perhaps convinced that the domestic market is different from others, but it’s pushing into stiff headwinds. The market for fossil fuel passenger vehicles and light trucks peaked in 2017, according to BloombergNEF, and while hybrid and plug-in hybrid sales are rising, they’re not growing as quickly as pure EVs.

Perhaps the most cautionary tale comes not from an American automaker, but a Japanese one.

Honda, which recently killed three EV projects, has imperiled its future as a global car manufacturer. By pulling back on EVs, it will forgo crucial lessons that have helped companies like Tesla and BYD slash the cost of their vehicles. And because EVs are ideal platforms on which to build software-defined vehicles, Honda stands to miss out on the other trend that’s sweeping the industry, one that has also helped companies trim expenses.

In all, it paints a grim picture for legacy automakers that have dialed back EV ambitions.

Companies that don’t get their respective EV houses in order could lose out to competitors in the global market, sacrificing revenue that could keep them competitive for years to come.

In Brief

A day after Elon Musk lost his lawsuit that threatened OpenAI’s structure, leadership, and finances, the AI giant is ready to move forward with its initial public offering, sources told the Wall Street Journal.

OpenAI chief executive Sam Altman reportedly hopes that his company will be ready to go public by September. The ChatGPT maker has been working with tech IPO powerhouse bankers Goldman Sachs and Morgan Stanley, and may file IPO paperwork confidentially with regulators within days or weeks, per the WSJ.

The news of OpenAI’s potential IPO, which by all accounts should be a blockbuster, comes as the world awaits the public disclosure of SpaceX’s IPO filings, which are expected to appear as soon as Wednesday, according to reports. Rocket-maker SpaceX is, of course, now one of OpenAI’s major competitors, after it consumed Elon Musk’s model maker, xAI.

Now that Musk failed to skewer OpenAI, the competitor he co-founded, through the heart with a lawsuit, it looks like the next Musk vs. Altman battle will take place in the world of finance. Which one will be the bigger IPO?

OpenAI did not immediately respond to a request for comment.

On Wednesday morning, Jeff Bezos said in an interview on CNBC that Americans earning in the bottom half of incomes should not pay taxes.

“Why is a nurse in Queens who makes $75,000 a year paying more than $1,000 a month in taxes?” Bezos said. “That’s $1,000 that could help with rent, or groceries, or anything … To me, it’s kind of absurd that we’re doing this. We shouldn’t be asking this nurse in Queens to send money to Washington. They should be sending her an apology.”

Jeff Bezos said the bottom half of Americans should pay zero federal income tax.



He cited a nurse in Queens making ~$75K and paying ~$12K in taxes saying “we shouldn’t be asking this nurse in Queens to send money to Washington.” pic.twitter.com/8KSgrO5TnE — Shay Boloor (@StockSavvyShay) May 20, 2026

Bezos argues that the lower half of earners contributes just 3% of total tax revenue — yet someone like that nurse is devoting roughly 16% of her salary to produce that negligible share. She’s paying a meaningful personal price to fund a rounding error in the federal budget.

This moment of empathy may be surprising coming from Bezos, one of the wealthiest people in the world. Billionaires like Bezos have taken advantage of holes in the tax system so that they pay income tax on just a small percentage of their annual gains. In 2007 and 2011, Bezos didn’t pay income tax at all. According to a ProPublica investigation, Bezos’ wealth increased $127 billion from 2006 to 2018, but he reported $6.5 billion in income. While this amounted to a huge $1.4 billion tax payment, this represented a tax rate of only about 1%.

This isn’t illegal. Americans are not taxed on unrealized capital gains, meaning that if Amazon’s stock balloons to make Bezos even more wealthy, he will only pay taxes once he sells that stock. The ultra-wealthy do whatever they can to hold on to their investments. Instead, they take out massive loans using their stock as collateral, live off of those loans, and avoid paying taxes on them, since those loans are technically debt.

Has Jeff Bezos finally realized how unfair this is? Does he now understand the frustration of the middle class, who can’t just take out loans on their heaping piles of company stock to avoid reporting capital gains?

“[Senator] Elizabeth Warren has made this point repeatedly … you and others are able to pay a lower tax rate — even though you’re paying an enormous sum in taxes — a lower tax rate than maybe I am, for example,” CNBC reporter Andrew Ross Sorkin told Bezos.

Bezos is hardly likely to invite Senator Warren over for dinner at one of his many mansions. In his words, the United States has a spending problem, not a revenue problem.

“We already have the most progressive tax system in the world,” Bezos said. “The top 1% of taxpayers pay 40% of all the tax revenue. The bottom half pay only 3%.”

Bezos on CNBC: "You could double the taxes I pay, and it's not gonna help that teacher in Queens. I promise you." pic.twitter.com/ocbf34XZhA — Aaron Rupar (@atrupar) May 20, 2026

Yet Bezos still pays taxes at a lower rate than most Americans. That fact remains true, even after he paid taxes on the Amazon stock he sold to fund Blue Origin, his space-exploration company.

“If people want me to pay more billions, then let’s have that debate. But don’t pretend that that’s going to solve the problem,” Bezos said. “You could double the taxes I pay, and it’s not going to help that [nurse] in Queens.”

It’s fundamentally difficult to imagine that we couldn’t find some productive use of the billions more dollars Bezos could be paying in taxes, even when juxtaposed with a gargantuan $7.4 trillion federal budget.

That nurse in Queens would probably find it helpful if she could reliably take public transit to work, or if she could send her children to public schools that actually have enough supplies to go around. She would probably appreciate it if she could go to the hospital in an emergency without having to worry about how she will pay thousands of dollars in medical bills.

Of course, these fantastical scenarios depend on our faith in the government to adequately distribute our tax dollars in ways that would help.

Bezos added, “If you really want to have a progressive tax system, you also want that money to actually be helping and not just dissolving in, you know, in like administrative bureaucracy.”

But money looks different when it’s a real, tangible thing — a finite number that appears in your bank account every other week, which must immediately cover rent, groceries, car payments, student loans, and other debts.

If it’s naive to dream of what we could accomplish if billionaires paid their fair share in taxes, then so be it. It’s also naive to think we can build data centers on the moon.

President Trump’s namesake cell phone provider and smartphone maker Trump Mobile is leaking customer data, including mailing addresses and email addresses, according to reports.

On Tuesday, YouTubers Coffeezilla and penguinz0, who both said they ordered Trump Mobile’s gold-colored T1 phone, said they had been alerted by a researcher who found the exposed data online.

“I know that because sadly I am one of those customers whose mailing address, email address, you know, everything short of credit card number is being leaked,” said Coffeezilla, who has risen to fame investigating crypto scams. “Do not order on trumpmobile.com unless you’re ready for your information to be leaked. It’s basically that bad.”

The two YouTubers said they ordered the T1 smartphone out of curiosity, not because they support the president and his business endeavors. Now, like other Trump Mobile customers, their information has been leaked.

They both said they were alerted of the leak by a source, who shared their personal information to prove they really had access to it. The researcher told them they were not able to get a hold of anyone at Trump Mobile, so the issue isn’t fixed yet. “All of us have been met with radio silence,” penguinz0 said.

Both YouTubers did not explain how hackers could access the data because they said it was very easy to do that, and the data is still available online.

Trump Mobile did not respond to TechCrunch’s request for comment.

Contact Us Do you have more information about the leak of Trump Mobile’s customers’ data? From a non-work device, you can contact Lorenzo Franceschi-Bicchierai securely on Signal at +1 917 257 1382, or via Telegram and Keybase @lorenzofb, or Do you have more information about the leak of Trump Mobile’s customers’ data? From a non-work device, you can contact Lorenzo Franceschi-Bicchierai securely on Signal at +1 917 257 1382, or via Telegram and Keybase @lorenzofb, or email .

According to the unique IDs in the leak, Coffeezilla said, it appears that only 30,000 people ordered the phone. That is a much lower number than expected. Last year, according to an estimate, there had been 590,000 preorders, with a cost of $100.

Trump Mobile was announced last year, with the promise of delivering a “Made in the USA” phone. As it turns out, according to NBC News, whose reporters got their hands on the device some nine months after it was supposed to be delivered, marketing materials now say the phone was “designed with American values in mind” and “shaped by American innovation.”

The Verge and others found that the phone sports an American flag with only 11 stripes instead of the dozen it should have, although it’s possible that the graphic designer intended the TRUMP MOBILE logo to be the 12th stripe. Some have also noted that the phone looks similar to a two-year-old HTC phone, suggesting it may just be a rebranded device.

The phone has been mired with issues since day one: 404 Media, for example, tried to order the T1 when it was announced and the order page failed and charged the wrong amount.

Enterprise software giant Intuit is letting 17% of its staff go, or about 3,000 people, as it seeks to divert resources toward baking AI into its products, Reuters reported, citing an internal memo sent to employees.

The memo by CEO Sasan Goodarzi said the layoffs are meant to reduce complexity by simplifying the company’s corporate structure and help it focus on AI efforts, according to Reuters.

The company, which makes accounting, tax, and personal finance software like TurboTax, QuickBooks, and Credit Karma, had 18,200 employees worldwide as of July 2025, according to its annual report.

Intuit did not immediately return a request for comment, or respond to questions about whether its management, directors, or its CEO would take a pay cut. Goodarzi’s salary was worth $36.8 million, including cash incentives and stock awards, during fiscal 2025.

The layoffs come during a bad year for the tech workforce. The tech industry has already cut more than 100,000 jobs this year, per Statista, and is on track to outpace both 2024 and 2025 if the layoff trend continues.

Companies such as Amazon, Block, Cisco, Cloudflare, Meta, Microsoft, and Oracle have let go of thousands of employees each, all of them citing a need to refocus expenditures around AI projects as a reason to cut jobs and restructure their organizations.

At the same time, all of these companies have recently reported strong revenue and profit, citing the apparent strong demand for AI products, services, or the infrastructure to power AI. Nearly all these companies’ share prices have risen, too, as investors bet that AI will serve as a new avenue of growth for software companies everywhere.

Intuit, however, hasn’t been perceived as a beneficiary of the AI boom, with its shares consistently underperforming in the broader S&P 500 over the past 12 months. The company has been caught up in the broader current of worries that traditional software-as-a-service firms will not be able to keep up or compete, as new and upcoming AI products and services threaten to change how software is developed and how it is used.

In its fiscal second quarter ended January, Intuit reported revenue of $4.65 billion, a 17% increase, and net profit of $693 million, a 48% improvement compared to a year earlier.

The company expects revenue to increase by about 10% in the third quarter, for which it will report results later today.

In Brief

Yesterday’s big news was Google’s plan to blow up its traditional Search in favor of an AI-powered experience — but Google isn’t the only company planning for the next generation of discoverability.AThis morning, Bloomberg has news of the Andreessen Horowitz-backed Exa Labs, which has raised $250 million against a $2.2 billion valuation to go after the same market. And it’s part of a wave of startups all chasing AI search, which has quietly become one of the most attractive targets in consumer AI.

From Bloomberg:

Exa is part of a wave of startups that are vying to transform the search industry, including Tavily, TinyFish and Parallel Web Systems. Led by former Twitter Chief Executive Officer Parag Agrawal, Parallel recently raised $100 million at a $2 billion valuation in a round led by venture firm Sequoia Capital, according to the Wall Street Journal.

At the same time, we’re also seeing conventional tech platforms like Amazon, LinkedIn, and Reddit looking to AI to revamp their search and discoverability features — so there will be plenty of potential acquirers if any of the startups start looking to sell.

The biggest competitor is ChatGPT, which still owns the interface layer and, prior to the Google launch, was handling the vast majority of the AI-powered searches taking place on a given day. But OpenAI can’t make Search a priority and Google has an ad business to protect, which could leave room for a smaller lab like Exa or Parallel to carve out a niche for itself.

Correction: A previous version of this article misstated Exa Labs’ valuation.

Stability AI, the company behind Stable Diffusion, is releasing a new family of audio models, called Stability Audio 3.0. The top model can generate professional-grade music of more than six minutes long, the company claimed.

The company is releasing four new models under the Stable Audio 3.0 name: small SFX (459M parameters), small (459M parameters), medium (1.4B parameters), and large (2.7B parameters). The duo of small models is suitable for on-device sound and music generation of up to two minutes.

Both medium and large models can create full compositions of 6 minutes, 20 seconds long that can maintain musical structure and melodic tone. This is more than double the length of what Stable Audio 2.0, released in 2024, was capable of generating.

Stability AI is making small SFX, small, and medium models available with open weights for anyone to use and modify. In 2024, the company released Stable Audio Open, which allowed for music generation of up to 47 seconds. The new family of models is a big step up from the previous open versions.

Image Credits:Stability AI

The large model is available only through the API and self-hosting paid services. Plus, companies with more than $1 million in revenue would need to get an enterprise license.

Many companies, including Google and ElevenLabs, are releasing models and tooling around music generation. However, as Suno’s and Udio’s ongoing court battles have proved, licensing of data and partnerships with music labels could become a key part of the long-term survival of these services.

Last year, Stability AI inked deals with Warner Music Group and Universal Music Group to develop models and music-creation tools. The company said that its latest set of audio models is built on fully licensed data.

The AI startup is developing a new suite of products for professional musicians but didn’t give more details on its features. Ethan Kaplan, former chief digital officer at Universal Audio and Fender, is joining the company to lead Stability’s professional music offering.

A number of AI companies are trying to bolster their credentials by hiring music execs. Earlier this year, Suno hired former Merlin CEO Jeremy Sirota as chief commercial officer. ElevenLabs has also hired Derek Cournoyer from indie music publisher Kobalt as a strategy lead for its music business.

Tesla’s Full Self-Driving (Supervised) driver-assistance software is now available in Lithuania, the second European country to approve its use, as the company pushes to position itself as an AI and robotics powerhouse and not just an automaker.

Making FSD available in Europe — which kicked off last month when the Dutch regulator RDW approved its use — is critical to Tesla’s and CEO Elon Musk’s ambitions. It’s also financially important for Musk, whose $1 trillion pay package is tied to hitting a number of product goals, including hitting “10 million active FSD subscriptions” by 2035.

Tesla has a long way to go before it hits that 10 million subscription figure. The company said during its first-quarter earnings call in April that it has nearly 1.3 million paying FSD customers globally.

Tesla’s FSD (Supervised) first launched in beta in late 2020. The advanced driver-assistance system, which still requires active driver supervision, has had regular updates since then to remove bugs and improve reliability and performance. Today, the system can handle driving maneuvers, such as steering, lane changes, and parking. Tesla used to allow owners to pay a one-time fee for the software; in January, Musk announced the only way to access the feature would be through a monthly subscription that is currently $99.

The rollout of FSD in Europe hasn’t been fast or furious — regulatory scrutiny has traditionally slowed deployment compared with the United States. Today, FSD is only available in the Netherlands, and now Lithuania. The rollout could accelerate if Dutch regulator RDW, which oversaw the first European approval, is successful in its bid for EU-wide acceptance.

Meanwhile, European countries can recognize Dutch certification and allow for FSD. And a number of them appear to be in the queue.

FSD Supervised now rolling out to Teslas in Lithuania 🇱🇹!



Making European roads safer, one by one pic.twitter.com/Uuj0bNG7pP — Tesla Europe, Middle East & Africa (@teslaeurope) May 20, 2026

The Greek transport ​ministry said Wednesday that an upcoming bill would grant approval of FSD, Reuters reported. Belgium is also expected to authorize the use of FSD, following the same authorization process used by RDW.

Outside of Europe, FSD is available in Australia, Canada, China, Mexico, New Zealand, Puerto Rico, South Korea, and the United States.

FSD (Supervised) isn’t the only product that fits within Musk’s vision of turning Tesla into a leader in AI and robotics, but it is the only one that is actually available to a wide swath of consumers.

FSD Unsupervised, a version of the software that handles all driving without any expectation of a human driver taking control, is not available to Tesla owners. It is used in a small fleet of about 50 Tesla robotaxis that operate in Austin, Dallas, and Houston.

Tesla’s Optimus humanoid robot, meanwhile, is not yet in mass production or available to consumers.

隨著企業AI應用逐漸從概念驗證進入正式環境，模型不再只是單一團隊測試使用，而是可能被多個部門、開發團隊或應用系統共同存取，連帶使模型存取控管、推論端點核准、用量控管與合規所需的用量報告，成為新的管理課題。

企業開源解決方案供應商紅帽（Red Hat）近日說明企業AI平臺OpenShift AI 3.4的模型即服務（Models-as-a-Service，MaaS）能力，能夠將AI模型作為內部可共用資源，透過API端點提供給不同團隊使用，減少各團隊自行架設與管理模型所形成的分散治理問題。

在OpenShift AI 3.4中，MaaS已列為可用於正式環境的企業功能，並內建AI推論閘道（AI inference gateway），企業不需要額外導入閘道工具，也不必另外管理一套生命週期，可減少模型服務上線後的部署與維運複雜度。這項能力由用於連接、控管與管理應用流量的Red Hat Connectivity Link核心技術堆疊支援，涵蓋政策管理、token配額、速率限制與API金鑰自助服務，並建立在代理伺服器Envoy、API政策管理專案Kuadrant，以及開源服務網格Istio等開放原始碼技術之上。

紅帽表示，管理者可透過Kubernetes原生的自定資源定義（Custom Resource Definition，CRD），設定各團隊的速率限制與token用量，降低預算超支的風險。開發者則可自行產生API金鑰，並限定在特定subscription（訂閱）資源範圍內使用。OpenShift AI會透過subscription綁定存取權限與用量規則，API金鑰可在建立時完成綁定，也能即時撤銷。

部分功能目前仍處於技術預覽階段，包括整合於OpenShift AI儀表板中的showback功能，可呈現內部成本與用量歸屬，並依模型與subscription群組追蹤token消耗量；企業身分驗證方面，OpenShift AI可透過開源授權服務Authorino控管使用者存取模型服務的權限，並支援採用OpenID Connect身分驗證標準的登入機制，可串接微軟企業身分服務Microsoft Azure AD、身分管理服務Okta，以及開源身分管理系統Keycloak。

此外，OpenShift AI 3.4也提供技術預覽階段的外部模型路由能力，可透過相容OpenAI Chat Completions API的/v1/chat/completions端點，將請求導向不同模型來源，包括透過大型語言模型推論框架vLLM執行的本機代管模型，或AWS代管式基礎模型服務Amazon Bedrock、Microsoft Azure OpenAI，以及AI模型業者Anthropic等外部模型服務。

對於已在企業內部使用API閘道或第三方代理工具管理AI流量的組織，紅帽表示，企業不必替換既有API閘道或代理工具，仍可將這些工具作為前端流量入口，把模型請求導向OpenShift AI代管的模型端點。該公司也發布第三方AI代理工具LiteLLM與AI閘道服務Portkey AI Gateway兩項參考整合案例，說明第三方代理工具如何連接OpenShift AI代管模型端點，並搭配LlamaStack等框架處理代理式工作流程，以及依團隊管理成本歸屬。

德國聯邦金融監理總署（BaFin）總裁Mark Branson於5月12日舉行年度記者會，將網路空間風險列為金融業重要風險之一。他指出，這類風險不只威脅金融機構本身的穩定，也可能影響消費者能否順利取得金融服務，因此已成為金融監理必須正視的議題。

Branson表示，網路威脅正在升高，部分原因在於AI與量子運算（Quantum Computing）等技術快速發展。他在演說中提到，數週前已有業者警告，新一代高效能AI模型可能對網路安全帶來影響。

路透社進一步指出，這項背景與AI業者Anthropic開發的Mythos有關；該模型已引發全球銀行業關注，部分美國銀行已取得存取權，監理機關也正檢視相關資安風險，以及金融機構是否已有足夠準備。路透社也引述資安專家看法指出，Mythos對銀行業及其老舊IT系統構成重大挑戰。

BaFin關切的是，新的AI模型可能大幅加快IT系統弱點的發掘與利用速度。Branson指出，這類AI系統能夠在新舊IT系統中快速找出大量弱點，並可能縮短攻擊者利用弱點的時間。對金融機構而言，這代表資安團隊必須重新檢視弱點管理、修補流程與老舊系統曝險情況。

BaFin認為，許多受監理企業目前已有良好防護，但現有資安能力未必足以因應接下來的威脅變化。Branson強調，企業必須準備面對更短的弱點修補時程：過去修補管理週期可以用月來計算，但未來可能必須在幾天內完成，甚至縮短至數小時內。這將提高中小企業的維運壓力，但在攻擊速度加快的情況下，縮短修補時程已成為必要的資安投入。

不過，修補速度加快也可能帶來系統問題與服務中斷。Branson以CrowdStrike事故為例指出，資安修補與更新若處理不當，本身也可能成為營運韌性風險。因此，BaFin強調，金融機構在加快弱點修補的同時，也必須符合金融業數位營運韌性法規（Digital Operational Resilience Act，DORA）要求，並維持良好的資安衛生。

為了強化監理量能，BaFin已擴大金融業資安與科技部門，後續將針對金融業者進行更聚焦的IT查核。Branson指出，BaFin在監理過程中已發現修補管理等問題，將藉由這類查核協助金融業衡量風險並強化韌性。