A leaker with a reasonable reputation has dropped the news that variable aperture is only coming to the iPhone 18 Pro Max, not both Pro models as previously expected.

Variable aperture is a feature that could narrow the gap between iPhone and DSLR photography still further, and it’s been rumored to be coming since at least 2024. More recent reports have said the camera component is now in production, but a new leak says it will be confined to the top of the range Pro model.

According to Ice Universe writing on Chinese social media site Weibo, the variable aperture feature will be exclusive to the iPhone 18 Pro Max. There are no further details, but the leaker implies that it’s because the component needs the larger size model.

“Even Apple isn’t immune to the trade-off,” writes the leaker (in machine translation), “where experiencing top-tier imaging capabilities means having to put up with the weight of a ‘brick.’”

The majority of smartphones, and all iPhones to date, have had a fixed aperture camera system. A variable one increases and decreases the hole through which light gets to the sensor, but it requires more internal volume as compared to a camera without to do so.

It’s the extra room needed by the apparatus that has proven to be a problem. It has previously been speculated that Apple’s moving to a camera plateau instead of a bump could provide enough space.

Given how strong the repeated rumors of this feature coming were, it has increasingly seemed certain that Apple has managed to include a variable aperture component. Then given that the camera system is a key selling point of the Pro models, it was presumed that it would come to both the iPhone 18 Pro and the iPhone 18 Pro Max.

Apple has done this before

There is a precedent for this with the iPhone 15 Pro Max in 2023. That model was the first to get Apple’s tetraprism system which meant it had a 5x optical zoom, compared to the iPhone 15 Pro’s 3x optical zoom.

In that case, the tetraprism feature was added to the non-Max version a year later, with the iPhone 16 Pro.

It’s likely that this will happen again, but it’s believed that there will be more of a size difference between the Pro and Pro Max models for 2026. As well as the regular difference in screen size, the iPhone 18 Pro Max is said to be larger than its predecessor in order to house a bigger battery.

Conceivably, the increased size of the chassis for the larger battery could mean there being more room for the variable aperture mechanism.

Note that Ice Universe has a generally good track record for leaks. Their most recent ones have centered on the iPhone Ultra dimensions.

Amazon’s updated terms of service now contain a clause that may take many subscribers by surprise. The online giant is revoking your right to file a class-action lawsuit against it if a dispute arises, according to its Conditions of Use page.

On Friday, Amazon sent an email to subscribers explaining that it will now settle any disputes via arbitration. Customers who have continued to use the corporation’s services after Aug. 14 implicitly waived their rights to bring new class-action suits against the online retail giant.

The language used in the Conditions of Use page states this new legal limitation outright.

“You and we agree that any dispute or claim relating in any way to your use of any Amazon service, or to any products or services sold or distributed by Amazon or through amazon.com, will be resolved by binding arbitration rather than in court,” the document reads. “There is no judge or jury in arbitration, and court review of an arbitration award is limited. A neutral arbitrator will resolve the dispute or claim and must follow the terms of this agreement as a court would.”

A class-action waiver is also included, which prevents customers from filing requests for arbitration together. Although Amazon is allowed to resolve batches of arbitration proceedings under detailed mass arbitration rules, each dispute must be filed individually.

Amazon’s language frames the change as a positive for its customers, since the company pledges to “pay most of the cost” for customers who choose to arbitrate disputes.

A representative for Amazon did not immediately respond to a request for comment.

A company spokesperson told CNET’s sibling site PCMag over email that Amazon has and will continue to “continually update our Conditions of Use to better serve our customers.”

“We determined that reinstating the arbitration clause will offer customers a fast, cost-effective way to resolve disputes while still giving them the option of going to small claims court,” the spokesperson wrote.

But at the end of the day, that arbitration clause may serve to stifle customer disputes by making it more difficult to seek financial restitution.

The big takeaway: Receiving compensation from Amazon may be harder

There are very few exceptions to Amazon’s new arbitration clause, and none of them are particularly great for customers. You can still take the company to small claims court, but the compensation you can receive is extremely limited.

Before the terms of service update, it only took one motivated individual to perform research, consult legal professionals and get the ball rolling on a class-action lawsuit. From there, other affected parties were able to sign on and receive compensation with comparatively little effort.

That’s not the case with an arbitration policy, which has a higher bar for individual effort and personal costs that may dissuade many customers from hashing out their financial disputes.

Before any arbitration proceeding can be filed, customers must submit a dispute claim to Amazon itself, giving the company 60 days to resolve it. If Amazon doesn’t respond during that two-month window, you can move on to the actual arbitration process.

Filing an official complaint with JAMS, Amazon’s chosen third-party arbitrator, requires you to pay a $250 fee upfront, which is a financial hurdle that the average class-action member likely won’t or can’t be bothered with.

From there on, “a neutral arbitrator will resolve the dispute or claim and must follow the terms of [the Conditions of Use] as a court would,” according to Amazon. “The arbitrator shall issue a written award that states the disposition of each claim and provides a concise statement of the essential findings and conclusions on which it is based.”

If you’re keeping a running tally of every hurdle presented to customers by the arbitration clause, you’ll realize that this process demands more time and money than any class-action lawsuit settlement claim does.

Technically, a critical mass of maligned customers could initiate mass arbitration and be compensated with more money than would ever be awarded to them in a class-action lawsuit. In fact, Amazon previously revoked an arbitration clause in its terms of service in 2021, after 75,000 people simultaneously filed disputes alleging Alexa was recording them without their consent.

Now, the e-commerce giant is trying to force customers into individualized claims once again, likely betting that the legal red tape of the arbitration process will put off many affected customers from ever fighting for financial restitution in the first place.

Existing lawsuits aren’t governed by the new legal language

The terms of service update only limits US Amazon customers from proposing new class-action suits, which means any case filed prior to Aug. 14 will be allowed to proceed. 

That means a bevy of existing lawsuits, including separate cases that allege Amazon purposefully shortened the lifespan of its first- and second-generation Fire TV Sticks and violated the privacy of millions of Americans through Ring camera AI features, will still have their day in court.

The e-commerce giant recently settled a $2.5 billion class-action lawsuit that alleged it had used deceptive “dark patterns” to push customers into paying for Prime subscriptions, before making it intentionally difficult to cancel the recurring charge.

Amazon admitted no wrongdoing as the case concluded, but more than $1.5 billion was earmarked to refund customers after legal fees were paid out. A costly, high-profile case like this one may have been part of the reason why the company changed its terms of service.

Changes to Amazon’s subscriber agreement won’t insulate the company from litigation by government agencies, so it will still be subject to lawsuits filed by the US Federal Trade Commission.

Tyler is a writer for CNET covering laptops and video games. He’s previously covered mobile devices, home energy products and broadband. He came to CNET straight out of college, where he graduated from Seton Hall with a bachelor’s degree in journalism. When Tyler’s not asking questions or doing research for his next assignment, you can find him in his home state of New Jersey, kicking back with a bagel and watching an action flick or playing a new video game.

You can reach him at tgraham@cnet.com.

See full bio

Advertisement

BGMI is one of, if not the most popular games in India, with new events and modes arriving every few months. While this keeps things fresh for everyone, not every phone can run BGMI smoothly because it’s too heavy. And given that every phone today ships with less RAM, KRAFTON is bringing a lighter version of its popular battle royale game to the country. The company has officially announced that BGMI Lite is coming to India by the end of 2026, although details about its gameplay and hardware requirements are still under wraps.

What is BGMI Lite?

While KRAFTON hasn’t explicitly said that BGMI Lite is designed for budget smartphones, a lighter version could make the game more accessible to users with older or entry-level devices. The standard BGMI experience can be demanding on smartphones with limited processing power, RAM, or storage. A less resource-intensive version could potentially allow more players to run the game smoothly without upgrading their phones. However, KRAFTON has not confirmed the system requirements or hardware targets for the game, so it’s too early to say exactly which devices will support it.

KRAFTON’s previous experience with PUBG Mobile Lite offers some context for what a Lite version of BGMI could look like. PUBG Mobile Lite was launched in 2019 as a smaller version of PUBG Mobile for smartphones with limited hardware. It used a smaller installation size and reduced some aspects of the full game while retaining the core battle royale experience. The game featured a smaller map and supported up to 60 players, compared with 100 players in the standard PUBG Mobile. Matches were also considerably shorter, typically lasting around 10 minutes. That said, these specifications should not be treated as confirmed BGMI Lite features.

Reach Capital announced Tuesday the close of a $265 million Fund V. The thesis of the 11-year-old, San Francisco firm is to back founders building AI applications that can “expand human potential,” Tony Wan, head of platform at Reach Capital, told TechCrunch. In practice, he added, that means looking at founders building across three areas: learning, health, and work.

“We believe AI should serve human flourishing, not replace it,” Wan said. The firm’s previous investments include Replit, ClassDojo, and Coral Care. 

The new fund will write checks of $1 million to $10 million, spanning pre-seed through Series A, into roughly 50 companies over the next three years. So far, no companies have been backed through Fund V.

Limited partners include Capricorn Investment Group, the Los Angeles Fire and Police Pensions, the LEGO Foundation, and College Board. Speaking to TechCrunch, general partner Jomayra Herrera said fundraising went smoothly and that the team was able to raise the new fund in less than six months. 

“The vast majority of our LPs doubled down, and we brought on a few new marquee LPs,” Herrera said. “We attribute this to LP interest in sector-focused boutique funds that focus on conviction-based investments.” 

Reach Capital’s new fund is noteworthy given the barbell shape the broader fundraising market has taken in recent years, with capital flowing overwhelmingly to giant, brand-name funds on one end and to sharply focused specialists on the other, with generalist firms in the middle struggling to get LPs’ attention.

Analysis by PitchBook and the National Venture Capital Association found that established firms captured more than 90% of the roughly $62 billion raised across U.S. VC funds through May of this year, leaving a smaller pool of first-time and mid-sized managers to compete for whatever’s left. Reach’s thesis, with over a decade of edtech and impact-investing, fits the mold of the kind of specialist fund LPs have remained open to funding.

The outfit previously raised $215 million for Fund IV in 2023 and $165 million for Fund III in 2021.

One of its most recent exits came in June, when Superhuman — the productivity platform now owned by Grammarly — acquired GPTZero, the AI-detection startup co-founded by Princeton graduate Edward Tian. Terms weren’t disclosed, but GPTZero had grown to more than 19 million registered users and $30 million in annual recurring revenue on just $13.5 million raised, and Reach was one of several investors in the company, alongside Uncork Capital, Footwork, and Jack Altman’s Alt Capital.

Cursor began rolling out Origin, its own code hosting platform, to paid users on Monday morning. Roughly three and a half hours later, GitHub’s status page lit up with what became a six-hour-and-forty-two-minute global degradation — error rates near 20% across pull requests, issues and the API, and near 50% on archive and raw file downloads, according to GitHub’s incident log. Enterprise single sign-on went down with it: SAML, OIDC, SCIM provisioning and Team Sync all failed. So did Copilot.

The developer internet did what the developer internet does.

“You can now host your repos in Cursor Origin and deploy to Vercel via Cursor Origin which is itself hosted on Vercel,” Vercel chief executive Guillermo Rauch posted on X. “And unlike GitHub, it’s online 😁” Asked why he was smiling, Rauch replied: “trying to make light of the situation. We ourselves are stuck because of github rn!”

Matt Palmer, who works at Cursor, quote-tweeted his own company’s launch with the day’s best line: “We were going to ship this earlier, but GitHub was down.” A GitHub outage, in other words, delayed the launch of a GitHub competitor.

Product launches get locked weeks in advance, and no evidence suggests Cursor timed this one. But the coincidence did the company an enormous favor, because it dramatized the argument Origin exists to make. For eighteen years, choosing where to host your team’s source code has been the least interesting decision an engineering organization makes. Cursor is betting that AI agents have made it interesting again — and for technical decision makers, that is the real news here. Not a new product, but a new procurement question with a governance problem attached.

Inside Origin: what Cursor’s code hosting platform actually does

Origin lives in a new Codebase tab inside Cursor. Teams name a codebase, which becomes part of its URL, then push to it over the command line. From there they get the machinery you would expect from a forge — the service layer that wraps Git and handles storage, permissions, checks and merges. Every repository comes with pull requests: timelines, commits, checks and files changed. Reviewers read the diff, leave comments and merge, without ever opening a browser tab.

What Cursor built around that machinery is the part worth studying. Agents now operate in the same surface as the code and the pull requests they are modifying. “Your code, PRs, and agents are now in the same place,” the changelog reads. A developer can ask questions about the file on screen, hand an agent a review comment and have it revise the pull request in place, or tell it to push a branch — all inside the editor where the code was written.

Three integrations shipped on day one, and the choice of partners is telling. Vercel spins up a preview deployment for every pull request and ships to production on merge, available in public beta for Pro and Enterprise customers, its developer account said. Depot and Buildkite run continuous integration, and critically, both execute existing GitHub Actions workflows unchanged. Buildkite adds native pipelines on top.

That compatibility layer is the whole strategy in miniature. Cursor is not asking teams to rewrite their build system, retrain their engineers or rip out their deployment pipeline. It is asking them to try a second window onto code they already have — which is a far easier request to approve.

More partners are coming, the company said, and the ones it landed first are the ones that matter to a platform team evaluating whether Origin can carry real work. A forge without deployments and CI is a code viewer. A forge that runs your existing Actions workflows and ships previews to the CDN you already pay for is a candidate.

Why letting GitHub stay the source of truth is Origin’s smartest design choice

Here is the decision enterprise buyers should study most closely, because it determines whether Origin survives a security review at all.

Cursor does not ask you to leave GitHub. Connect a GitHub organization, pick repositories, and they appear alongside Origin-native ones. “Pushes keep going to GitHub, which stays the source of truth for anything started there,” the changelog says. Access permissions mirror GitHub’s existing read and write settings rather than establishing a parallel system. Pull request conversations sync in both directions — comment in Cursor and it posts to GitHub; reply or react on GitHub and it surfaces in Cursor “within seconds.”

This is a classic wedge, and a well-executed one. Rip-and-replace migration of source control ranks among the highest-risk projects an engineering organization can undertake. It touches continuous integration, compliance evidence, audit trails, branch protection rules, every integration in the toolchain and the muscle memory of every engineer on staff. Almost no chief technology officer approves that for a product in early beta.

A read-mostly mirror that leaves GitHub authoritative approves itself. It costs nothing to try, breaks nothing if abandoned, and quietly relocates the place developers spend their working hours. If Cursor’s review experience proves better — and Cursor spent real money to make sure it would — the source of truth eventually follows the attention.

That money went to Graphite, the code review startup Cursor bought in December 2025 for what Axios reported was well above its $290 million Series B valuation. Graphite built stacked pull requests, the workflow that lets developers keep shipping dependent changes without waiting on approvals. Announcing the deal, Cursor wrote that “the boundary between where you write code and where you collaborate on it feels increasingly arbitrary,” and promised “some more radical ideas we can’t share just yet.” Origin is the radical idea. Graphite co-founder Tomas Reimers unveiled it on stage at Cursor’s inaugural Compile conference in June and leads its development.

How AI agents turned code review into software’s new bottleneck

The case for an agent-native forge rests on a claim that is easy to state and, unusually for this market, well supported by evidence: writing code stopped being the constraint. Reviewing and integrating it became one.

Google’s 2025 DORA report, drawn from nearly 5,000 technology professionals, found that 90% of developers now use AI at work, spending a median of two hours a day with it, and more than 80% say it made them more productive. But AI adoption showed a positive relationship with software delivery throughput and a negative one with delivery stability. More output, more breakage. The report’s authors describe AI as “an amplifier” that “magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones.”

Trust has not kept pace with volume. Stack Overflow’s 2025 developer survey of 49,009 respondents across 177 countries found 84% using or planning to use AI tools, while trust in their accuracy fell to 33% from 43% a year earlier and distrust climbed to 46% from 31%. Two-thirds named “AI solutions that are almost right, but not quite” as their leading frustration. GitLab’s ninth annual DevSecOps survey, of 3,266 practitioners polled by Harris, put numbers on the operational drag: 73% had hit problems with vibe-coded output, 70% said AI made compliance management harder, and only 37% would let AI handle daily tasks without human review.

The volume climbs regardless. GitHub’s Octoverse 2025 counted 180 million developers, 630 million repositories and 43.2 million pull requests merged per month, up 23% year over year. And RuntimeWire reported the internal figure that best explains Origin’s existence: 35% of pull requests merged inside Cursor were opened by agents running autonomously in cloud virtual machines.

A forge built for humans assumes a pull request represents human intent, opened by someone you can ask what they meant. Once a third of merged changes come from software, the queue stops being a conversation and becomes a scheduling problem. That is a real architectural argument, and it is the strongest thing Cursor has going for it.

GitHub’s reliability crisis handed Cursor an opening it did not have to earn

The supply-side case for an alternative is simpler: GitHub has been unreliable, and its own executives have said so.

An analysis by LeadDev counted 257 incidents between May 2025 and April 2026, 48 of them major — roughly one significant disruption per week. February was the worst month on record with 37. GitHub Actions alone accounted for 57 outages in twelve months. Chief technology officer Vlad Fedorov has said the platform “wasn’t built for the scale it’s now being asked to handle” and must design for 30 times today’s load. In an April engineering post covered by InfoQ, the company acknowledged it “failed to meet its own reliability standards,” citing rapid growth, tight architectural coupling and inadequate load shedding. Monday’s outage was the seventh incident on GitHub’s status page in fifteen days.

The fatigue is audible. “GitHub really doesn’t feel built for the agent era,” one developer wrote on X as Origin went live. “It goes down way too often, but until now there haven’t been many real alternatives.”

The defections started before Origin existed. The Zig programming language moved to Codeberg in November 2025, citing Actions failures among its reasons. In April, Mitchell Hashimoto announced that Ghostty — a terminal emulator with more than 52,000 stars — would leave too, pointing to near-daily outages that blocked reviews and CI for hours. And The Information reported in March that OpenAI, a company Microsoft holds a large stake in, began building its own GitHub alternative partly because outages left its engineers unable to commit for hours at a time, as Tom’s Hardware relayed.

Microsoft’s structure has not helped. Thomas Dohmke resigned as GitHub chief executive in August 2025 and was never replaced; the unit’s leadership was absorbed into Microsoft’s CoreAI organization under executive vice president Jay Parikh. In a May report, The Information wrote that Parikh had warned deputies that coding tools from Cursor and Anthropic could eventually make GitHub obsolete. GitHub’s own answer to the agent era, Agent HQ, lets customers orchestrate third-party agents from Anthropic, OpenAI, Google, Cognition and xAI inside GitHub — a coherent strategy that concedes the agent layer and keeps the substrate underneath. Origin attacks precisely that substrate.

Now that SpaceX owns Cursor, who actually holds your source code?

Cursor’s rise has been extraordinary even by the standards of this cycle. Founded in 2022 by four MIT students, Anysphere raised $8 million from the OpenAI Startup Fund in October 2023, per TechCrunch, then $100 million at $2.5 billion, $900 million at $9.9 billion, and $2.3 billion at $29.3 billion last November. In May, Bloomberg reported annualized revenue of $3 billion and more than 3,000 customers paying at least $100,000 a year.

Then, three days before Origin shipped, Bloomberg reported that SpaceX completed its $60 billion all-stock acquisition of Cursor — an agreement TechCrunch covered in June, days after SpaceX’s record IPO and six months after it absorbed xAI. Cursor now operates inside a division called SpaceXAI. The vendor asking to hold your proprietary source code became, last Friday, a unit of a rocket company with its own frontier-model division and a founder not known for institutional caution.

Jason Andersen of Moor Insights & Strategy raised the model-routing question to Tech Times in June, before the deal closed: “xAI’s models and treatment of guardrails are very different than what Cursor has stood for.” That piece framed the question a chief information security officer now has to answer. When one company controls the editor where agents write code, the host where that code lives and the model those agents run on, what governs what it does with the code?

Cursor has not published an answer. RuntimeWire noted before launch that Origin’s pricing, security architecture, data-handling terms and migration tooling were all unpublished, and Monday’s changelog adds none of them. It says only that Origin reaches “all paid plan users starting today, except enterprise orgs whose admins opt out.” Opt-out, not opt-in — a sentence administrators should read twice.

There is also a track record to weigh. In July, researchers at Mindgard disclosed that Cursor would execute a malicious git.exe planted in a Windows project’s root the moment a user opened it, with no prompt — a repository-poisoning flaw they first reported in December 2025. The Hacker News reported that Cursor declined to patch it, calling the issue out of scope under a shared-responsibility model while conceding it had not “closed the loop with the researcher in a timely manner.” No CVE was issued. The same flaw class turned up unpatched in GitHub Copilot CLI, Google’s Gemini CLI and OpenAI’s Codex — but a vulnerability the vendor declined to fix makes an awkward footnote for a product whose pitch is basically “let us hold your repositories.”

What engineering leaders should settle before they let Origin into the toolchain

Origin is a beta, not a migration, and treated as one it is worth evaluating. The sync mode gives platform teams a low-risk way to measure whether an agent-native review surface shortens cycle time, without touching a single branch protection rule. But three things deserve resolution before anything authoritative moves.

The first is the default. Origin switches on for paid users unless an enterprise administrator opts out, which means an organization that has not made an affirmative decision about whether proprietary code may be mirrored to a new host has effectively had that decision made for it. Confirming your posture is a Monday-morning task, not a next-quarter one.

The second is the paperwork. Retention, residency, training use, subprocessors and what changes now that Cursor reports into SpaceX are all unpublished, and a product page is not a contract. Until those terms exist in writing, the defensible position is to treat Origin as a convenience layer over GitHub rather than a system of record — which is, conveniently, exactly what its architecture already is.

The third is the exit. Origin’s Actions compatibility and its GitHub-as-source-of-truth design are the properties that make it safe to adopt. They are also the ones most likely to erode as Cursor’s incentives shift toward owning the substrate rather than borrowing it. Ask what egress looks like now, while the mirror is still a mirror.

None of which makes Cursor’s argument wrong. GitHub earned its incumbency by being boring, dependable infrastructure, and it has spent eighteen months being neither while a third of the code arriving at its front door stopped being written by people. Origin is a serious answer to a real problem, built by a team that bought the right company to build it.

But GitHub’s failure and Cursor’s are different in kind, and enterprises should not confuse them. Monday’s outage resolved at 20:22 UTC. Availability is an engineering problem, and engineering problems close. The question of who holds your source code, what they may do with it and who they ultimately answer to carries no such timestamp — and on that one, the company that spent Monday selling trust has yet to publish its terms.

If you have ever wondered why a specific song made its way to a playlist, Spotify’s new Playlist Notes feature is here to help. The feature lets editors and users add context to their songs, podcasts, and audiobook picks. 

The feature joins the long line of playlist improvements, including custom cover art, the ability to organize playlists into folders, and smooth transitions between tracks, that make it easier to personalize your listening experience and make your playlists feel more like your own.

What exactly are playlist notes?

Playlist Notes let you attach a quick thought next to any song, podcast episode, or audiobook inside a playlist you made or collaborate on. Maybe a track reminds you of a road trip, or an episode changed how you see something. Now you can share that context without ever leaving the app.

The feature grew out of an earlier idea called User Notes, and Spotify’s editors are getting first access. They will use it to explain why a track made the cut on some of Spotify’s biggest playlists, covering everything from the reasoning behind a pick to why it feels culturally relevant right now.

Alongside the notes, you also get Editor Profiles, which show an editor’s favorite tracks and albums, plus every playlist they work on. It’s a fun way to actually know who is picking the songs you’re vibing to.

Where can you find these notes?

Editor notes are currently rolling out to Free and Premium users aged 16 and up in the US, Canada, the UK, Ireland, Australia, and New Zealand. For now, look for them on playlists like Today’s Top Hits, RapCaviar, Hot Country, All New Pop, Mint, and Fresh Finds Hip-Hop. 

Tap Notes at the top of these playlists to start reading, and tap an editor’s name to browse their profile. You’ll also spot these notes in Now Playing, even when you’re listening outside the playlist where the note first appeared.

You can also add your own notes. Open a playlist you own or collaborate on, tap the three dots next to a track, and hit Add Note. Just remember that anyone who can view your playlist can also read your notes, so keep that in mind before getting too personal.

Playlist Notes for personal playlists are already rolling out on iOS and Android in over 100 markets, so update your app if you don’t see it yet.

Nik Storonsky, the man behind Revolut, has closed a $500m second fund for QuantumLight, the algorithmic venture-capital firm he co-founded to let software, rather than partners, decide where the money goes.

The fund is double the size of QuantumLight’s $250m debut vehicle from last year, a swift escalation for an outfit still trying to prove that data can out-pick the humans.

Nik Storonsky, whose personal fortune has climbed with Revolut and who is reportedly in line for a vast share award, is the firm’s most conspicuous backer.

QuantumLight’s pitch is a direct challenge to the clubby traditions of venture capital. Instead of a bench of star partners trading on instinct and network, the firm runs a systematic, data-driven model that screens companies at scale and generates investment decisions quantitatively, something closer to a quant hedge fund than a Sand Hill Road partnership.

The approach is of a piece with Nik Storonsky’s reputation as fintech’s most relentless optimiser. He built Revolut into Europe’s most valuable private tech company on a culture of aggressive targets and data over sentiment, and last year QuantumLight published a hiring playbook distilling the management methods behind that rise, a set of blunt tips that read like an operating manual for high-growth companies.

QuantumLight applies the same conviction, that most decisions are better made by system than by instinct, to the business of picking winners.

Doubling the fund in barely a year is a statement of confidence, though whose confidence is the interesting question.

QuantumLight has not detailed its outside backers, and Storonsky’s own wealth gives him ample means to seed his own experiments.

An oversubscribed close suggests limited partners are buying the thesis too, wagering that algorithmic sourcing can surface growth-stage companies that traditional funds overlook, and that the model deserves twice the firepower after a single year.

QuantumLight’s first fund is only a year old, far too young to have produced the exits that would show whether the machine actually beats the market, and venture returns take the better part of a decade to judge.

Plenty of firms have promised to “quant-ify” venture capital before, and the discipline’s best returns still tend to come from a handful of outlier bets that are notoriously hard to model, precisely because they look unreasonable at the moment they are made.

Still, the timing is apt. As AI reshapes every knowledge industry, the people who allocate capital are hardly exempt, and a wave of funds now claim to use machine learning to source deals, score founders and time markets.

QuantumLight is among the most committed to the idea, and the most credibly funded, which makes its record a useful test case for whether software can genuinely industrialise a trade that has always prized taste and relationships over spreadsheets.

There is also the matter of a chief executive’s attention. Storonsky is steering Revolut through a pivotal stretch, having lately won a French banking licence and pushed into business banking, with an IPO reportedly a couple of years away and likely to list in the US, all while running a second act in venture capital.

Founders are rarely one-company people, but the split focus is the kind of thing Revolut’s eventual public-market investors may come to weigh.

For now, $500m is a serious sum with which to test a serious idea: that the qualities venture capitalists have always sold, judgment, instinct, a good eye, can be replaced, or at least bettered, by code.

If Nik Storonsky is right, QuantumLight will look prescient. If he is wrong, it will be an expensive reminder that some bets resist being reduced to a model. Either way, the experiment has just doubled in size.

In context: A new Mac Mini assembly line at Foxconn’s Houston facility is expected to begin production later this year, expanding Apple’s US manufacturing footprint for AI-related hardware. The project is part of Apple’s effort to increase domestic production while continuing to rely on contract manufacturers and an international supply chain.

The line will operate in a new 170,000-square-foot space at the plant, which already makes AI servers for Apple. Foxconn is funding the assembly-line buildout, while Apple has committed to purchasing the products made there.

The Mac Mini has seen stronger demand over the past year, particularly among customers using the compact desktop to run AI models at home. Its planned Houston production also comes after Apple discontinued the Mac Pro, which was assembled in the US at a Flextronics plant that President Trump visited with Apple CEO Tim Cook in 2019.

Cook and Commerce Secretary Howard Lutnick toured the Houston site last week. The visit included a new manufacturing school that Apple is opening to provide hands-on training for small and midsize US companies. Apple previously launched a classroom-based manufacturing program in Detroit.

“Advanced manufacturing is where the puck is going,” Lutnick told workers taking an Apple manufacturing course. “The problem is we need to train Americans.”

Cook said Apple’s US investment plans reflect its broader commitment to domestic production. “We believe in the promise of this nation and we’re proud to put our money where our mouth is,” he said in a speech.

Apple’s investment strategy relies heavily on its purchasing power rather than on building and owning factories. The company’s capital spending remains far below that of Amazon, Microsoft, and Google, which have made major investments in AI infrastructure and semiconductor capacity. Apple’s total capital expenditures since 2023 are less than what each of those companies spent in its most recent quarter.

That model is evident in Houston. Foxconn is paying to build the production lines, while Apple is supporting the project through product orders. The arrangement preserves Apple’s long-standing reliance on manufacturing partners while allowing it to use its supply-chain scale to influence where production and component sourcing take place.

Apple has made similar commitments in semiconductors. The company is among the businesses that have agreed to buy chip wafers from TSMC’s Arizona plant, a project expected to cost more than $200 billion. Apple has also reached a preliminary agreement for Intel to manufacture some of its device chips and said it plans to spend $30 billion on US-made chips from Broadcom.

The company has pledged to invest $600 billion in the US over four years. The total includes spending already planned for its American business, including employee pay, retail and corporate operations, and purchases from domestic suppliers.

The administration has encouraged Apple to expand its US manufacturing operations. President Trump has pressed the company to make iPhones domestically, but Apple has not announced such plans. Instead, it is expanding iPhone assembly in India.

The Houston event was among Cook’s final public appearances as Apple’s chief executive. He is expected to become chairman next month, with John Ternus, Apple’s longtime hardware engineering chief, expected to succeed him as CEO.

Image credit: The Wall Street Journal