Bottom line: The math behind AI subscriptions is starting to look uncomfortable. Flat monthly pricing helped fuel the rapid adoption of tools like ChatGPT and Claude, but new analysis suggests those fees may not come close to covering the actual cost of heavy use. As users push these systems harder and more demanding AI workflows take hold, the gap between revenue and compute costs is becoming difficult to ignore.

SemiAnalysis has calculated how big that gap really is. After testing subscription tiers from both OpenAI and Anthropic – running long-horizon coding and agentic tasks until weekly limits were exhausted – the firm found that the cost of theoretical maximum usage of these plans if priced at standard API rates far exceeds what users actually pay.

A $200 ChatGPT Pro 20x subscription could cost as much as $14,000 in API pricing if fully utilized. Anthropic’s Claude Max 20x plan, also priced at $200 per month, has a comparable ceiling, with potential usage totaling roughly $8,000 in token costs.

Those figures help explain why utilization rates matter so much to the AI companies offering them. According to SemiAnalysis, Anthropic breaks even on Claude Pro and Claude Max 5x at around 20% utilization. OpenAI’s margin is thinner. It begins losing money on ChatGPT Plus and ChatGPT Pro 5x once usage climbs above 11.4%.

– SemiAnalysis (@SemiAnalysis_) June 10, 2026

The economics get tighter at the high end. Anthropic reaches zero gross margin at roughly 10% utilization on its top-tier plans, while OpenAI crosses into negative territory at just 5.7%. It doesn’t take extreme use for these subscriptions to turn unprofitable.

Adjusting pricing or restricting access is not a straightforward fix. Subscription models have been central to user growth, and pulling back risks slowing momentum in a market where capabilities remain a key competitive differentiator.

Part of the pressure comes from how AI is actually being used. Token consumption is rising quickly, especially with agentic systems that can require up to 1,000 times more tokens than a standard prompt. That kind of demand is already forcing large organizations to rethink how freely these tools should be deployed.

Microsoft, Meta, and Amazon have reportedly pulled back from internal efforts that encouraged heavy usage after costs escalated. In one widely cited example, a company burned through $500 million in a single month using Anthropic’s Claude, largely because it failed to put limits on employee access.

That kind of overspending is pushing companies toward more controlled approaches. One strategy gaining traction is to shift workloads between models depending on the task. More complex queries go to expensive frontier models, while routine work is handled by cheaper alternatives.

– Dong Ming (@dming) June 14, 2026

The savings can be substantial. A Wall Street Journal report found that routing tasks this way can cut costs by up to 95%. “You don’t need a model that knows quantum gravity,” Columbia University vice dean Vishal Misra told the publication. “These open-source models are very capable, and the ability to charge a big premium for AI is going to diminish.”

Some companies have already made the shift. Flo Crivello, founder and CEO of AI assistant startup Lindy, announced that the company moved 100% of its traffic to DeepSeek V4, switching entirely away from Anthropic’s models. DeepSeek V4 proved comparable to Claude Sonnet at a fraction of the cost, and the move has “saved the company millions of dollars,” Crivello said.

Others are going further by building their own AI systems on top of open-source models trained on internal data. While that requires more upfront investment, it offers tighter cost control and reduces dependence on third-party providers. In some cases, these tailored systems may even outperform general-purpose frontier models for specific use cases.

There is some expectation that costs will ease over time. As infrastructure expands and newer models replace older ones, the cost of running mid-tier systems should decline. SemiAnalysis suggests that models at the Opus 4.8 level could eventually be delivered profitably for around $20 per month.

That does not apply to the most advanced systems, though. Frontier models, including those still in development, remain expensive to run. Their highest-end capabilities may increasingly be priced via APIs rather than bundled into consumer subscriptions.

For now, AI providers are juggling two forces: users want powerful tools at low, predictable monthly prices, but the infrastructure to run them remains costly and highly sensitive to usage. OpenAI CEO Sam Altman has acknowledged the tension, noting that rising token costs are becoming a serious issue and that the company is working to help users “get more value for less spend” when using ChatGPT.

Watch Ivory Coast vs Ecuador free live streams from anywhere in the world as two dark horse go head-to-head in Philadelphia.

If you want to make a statement ahead of FIFA World Cup 2026, beating one of the tournament favourites in their own backyard isn’t a bad way to go about it.

Canadian Prime Minister Mark Carney said Sunday that the US export ban that forced Anthropic to shut down Fable 5 and Mythos 5 demonstrates the danger of depending on a small number of powerful AI models. Speaking to reporters during a visit to Ireland, Carney framed the suspension as a warning about systemic vulnerability rather than a failure by any single company.

“The situation we’re in collectively right now with Mythos and Fable is something that can happen with over-reliance on certain models,” Carney said. “Nobody’s done anything wrong in this situation, but we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify.”

The former central banker then drew a direct parallel to the 2008 financial crisis. Carney, who led both the Bank of Canada and the Bank of England before entering politics, said “we have similar things in terms of model risk” and called for redundancy and diversity in AI infrastructure, the same principles regulators imposed on the banking system after the collapse of Lehman Brothers.

The analogy carries weight because of who is making it. Carney was governor of the Bank of Canada during the 2008 crisis and subsequently became the first non-British governor of the Bank of England, where he spent six years strengthening financial system resilience. When he warns that concentrated AI dependence mirrors the systemic linkages that nearly destroyed the global banking system, the comparison draws on direct experience rather than rhetorical convenience.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Carney said there is a “good flow of information” between the Canadian and US governments on AI and acknowledged that Washington has identified “some risks” with Anthropic’s latest models. But his emphasis was on the structural lesson, not the specific dispute between Anthropic and the US Commerce Department.

The comments arrive days before the G7 summit in Évian-les-Bains, France, which runs from 15 to 17 June and has AI governance prominently on its agenda. Anthropic’s Dario Amodei, OpenAI’s Sam Altman, and Google DeepMind’s Demis Hassabis are all slated to attend a working lunch with G7 leaders on Wednesday. Carney said he has already discussed AI with French President Emmanuel Macron.

“We need to make progress” on AI, Carney said, but cautioned that “there will not be a mission accomplished banner that comes out of the G7.” The remark suggests Canada will push for substantive commitments on AI diversification at Évian rather than symbolic declarations.

Canada has been positioning itself on this front. On 4 June, Carney launched “AI for All,” a $2.3 billion national AI strategy that includes sovereign computing infrastructure, a national supercomputer, and plans to raise business AI adoption from 12% to 60% by 2034. The strategy explicitly frames dependence on foreign cloud providers as a vulnerability.

The Anthropic suspension has now given Carney a concrete, real-time example to cite at the G7. A US government decision to invoke export controls against a single AI company cut off access to its most capable models for every user outside the United States, including allies. For G7 nations building their economies around AI capabilities they do not control, the implications are difficult to ignore.

Carney’s Ireland visit also produced a bilateral agreement between Canada and Ireland on AI cooperation, tech collaboration, and food security. Ireland currently holds the presidency of the Council of the European Union, making it a strategic partner for Canada as it seeks to deepen ties with Europe on AI, defence, and critical minerals.

The broader pattern is accelerating. The EU published a tech sovereignty package earlier this month curbing US cloud dependence. India has proposed a $5 billion sovereign AI fund in the wake of the Anthropic suspension.

Britain’s Cosine is rallying BT, HSBC, and BAE to build a sovereign frontier model. Carney’s 2008 analogy suggests he sees these as early moves in a structural realignment, not isolated reactions.

Gaming handhelds are great because they are portable (basically small). But that is also one of its biggest weaknesses. I was reminded of that while trying Asus’ new ROG Xbox Ally X20 bundle at Computex 2026. On its own, the Ally X20 is already a more polished version of the ROG Xbox Ally X. It arrives with nice updates that sound minor on paper but make a device feel more complete in your hands. The real surprise, though, was the bundled ROG XREAL R1 Edition 20 Gaming AR Glasses.

I walked in to try the 20th anniversary edition of ASUS’ handheld console, but the massive 171-inch screen trick surprisingly stole the show.

The handheld gets the right upgrades

The ROG Xbox Ally X20 is not a total reinvention of Asus’ Xbox-branded handheld. It still sits in that familiar Windows handheld space, with PC gaming power, Xbox integration, and the usual promise of taking your library anywhere. Everything ASUS changed here is to smooth out the experience, and you can feel it as soon as you actually hold it.

The biggest upgrade is the display. The X20 moves to a 7.4-inch OLED Nebula HDR panel, which is slightly larger than the standard Ally X screen and far more exciting visually. It supports a 120Hz refresh rate, FreeSync Premium Pro, Dolby Vision, and up to 1,400 nits of peak brightness.

The controls have also been cleaned up. ASUS added TMR joysticks, a Transforming D-pad that can switch between four-way and eight-way control, improved face buttons that sit more flush with the chassis, and rubberized grips on the back that definitely offer improved handfeel, which is important for gamers. A few extra fps, thanks to a faster chip, is great, though bad grips, mushy controls, or a weaker display are something you’ll notice every time you start playing on the gaming handheld.

Then ASUS straps a giant screen to the idea

The bundled ROG XREAL R1 Edition 20 glasses are what make ASUS’ portable console a little more interesting than the new Intel G3 Extreme-powered handhelds. Even at their largest, handheld screens are still handheld screens. Admittedly, that is part of the appeal. If you wanted a bigger screen, you’d likely go for a gaming laptop. For many, this is exactly what they want: a smaller window into big games.

Except that the glasses still change that equation. They can project a virtual display up to 171 inches at 4 meters, which sounds absurd until you actually put them on. Now, you’re not relying on the handheld screen anymore, and the Ally X20 becomes an oversized controller, which is powering a much larger screen.

During my hands-on with this ROG Xbox Ally X20 with the XREAL R1 Edition 20, I understood why the bundle exists. ASUS pairing these two gadgets together makes it all the more immersive. This isn’t the first time a pair of gaming AR glasses has been built for handhelds, but the Ally X20 bundle is the exception as it comes with the ROG XREAL R1 Edition 20, which is also optimized for this system in particular.

How AR glasses just work here

Gaming AR glasses have been pretty niche in the overall PC and console gaming segment. However, you are already holding a portable console in your hands, which comes with its own set of cables, chargers, and a case. So adding another peripheral doesn’t sound too inconvenient. The fact that you can stay portable, while the glasses let the display feel massive when you want it to, sounds incredibly fun.

Not everyone would want this bundle, but the ROG XREAL R1 Edition 20 glasses complete it. They turn the Ally X20 from a nicer handheld into a more flexible gaming setup. You get the benefit of both a convenient form factor console with a 171-inch virtual screen when you want a more immersive experience.

When you buy a TV, what motivates the purchase: Picture quality? Value? Screen size? Brand loyalty?

Is brightness in nits, dimming zones and Delta E colour accuracy among those ‘needs’. I’d hazard a guess and say ‘probably not’.

Which is not to say that these areas are not important, but I’d make an assumption that most of the TV buying audience is not au fait with these areas – some won’t know their nits from their candelas, and others just won’t care.

And that’s fine.

But the TV market seems to have developed a fixation with measurements and numbers, and it’s become a question of to what end?

Advertisement

Playing the numbers game

For the most part, you’re unlikely to come across these technical terms during your TV buying experience. Browsing the likes of Currys, Amazon and Richer Sounds, you’ll be exposed to the marketing ramble, most of it words that look like they could be equations or formulas (NQ4 AI Gen3, a9 AI processor).

On our side, the reviewers’ side, we get this as well as insight on the technical side. Of course, we need to know how they work, but I won’t lie in saying that some of this stuff goes over my head and requires educating myself and asking questions.

Advertisement

The more advanced TVs get, the more complex they become to understand. The use of AI has, in a way, accelerated this complexity faster. AI is often used as a blanket term for tech. For some TV manufacturers, as part of their brand story, it’s easier to say AI to get people interested and fit as part of the overall story in tech.

AI has helped TVs achieve a higher level of performance. But it’s also turned TVs into a numbers game – my number is bigger than yours.

But what does it really mean? If I have a TV from Samsung, LG, Sony, Hisense and TCL and they all hit 3000 nits of brightness, which is better? The one with more dimming zones? The one that covers the widest colour spectrum? The one with the best Delta E number?

Advertisement

Focusing on these areas turns it into a specs battle, and for most people, none of these measurements will have much context in their living room. If the picture looks good, then it looks good. Things like Delta E, tone mapping, and colour gamut – they’ll end up just confusing the buyer or making decisions more complicated, not easier.

Advertisement

Welcome to the real world

The TV that prompted this discussion was the Hisense UR9.

On paper, this TV has stonking specs. It’s an RGB Mini LED, brand new technology to the TV market that allows for purer colours, displaying a wider range of colours than more conventional LCD TVs and doing so with more accuracy. The number of dimming zones is nearly 1000, it claims to hit peak brightness levels of 4000 nits. All these specs suggest a Bona fide contender.

So why was I left underwhelmed by the UR9’s picture?

Colours weren’t that punchy or bright. The levels of sharpness and detail were less than those of an LG OLED65G6 sat next to it. The Dynamic mode, which should be the brightest, looked dull. Filmmaker mode, Cinema mode and IMAX Enhanced mode all looked identical to the point where I thought something must be wrong with the TV.

Advertisement

Advertisement

Having reviewed the U8Q in 2025 and hailed that as the best Hisense TV I’d tested, the UR9 felt as if the balloon had popped, leaving me deflated.

Reasons? I don’t think the anti-reflection/anti-glare screen worked in Hisense’s favour, as it seemed to shave off levels of detail, sharpness, and reduce contrast, which I also noted on Hisense’s Canvas TV. Highlights and overall brightness weren’t as high as measurements suggested, black levels not as strong or as deep as I hoped.

Viewing angles weren’t great either, like down to the type of panel used, and colours were a bit off despite many reviews noting the accuracy the UR9 out of the box.

But something just wasn’t quite right with the TV’s picture performance.

Advertisement

Personally, I wonder if this is down to Hisense’s PQ philosophy, or how it views colour. Every brand has a PQ philosophy; Sony wants to reflect the creator’s intent, as do LG and Panasonic, while Samsung wants to do that and offer the brightest, most colourful picture experience with any technology it gets its hands on.

Advertisement

But I can’t quite discern what Hisense’s philosophy is. In the past I’ve found its TVs have relied heavily on Dolby Vision, which is not their PQ philosophy but Dolby’s.

Earlier this year, I witnessed a shootout between several TVs: OLED, Mini LED, RGB Mini LED, and SQD Mini LED. The model that fared the worst was the Hisense, with all the test patterns and demos causing some sort of issue.

Advertisement

One pattern in particular showed Hisense’s lack of precision. The pattern was a white rectangle surrounded by green. The rest of the TVs showed the white rectangle as white – the Hisense showed the white rectangle to be a shade of green…

The gist of all this? What the specs say on paper doesn’t always translate to the real world.

Don’t be swayed by the specs, the charts, the stats. Not everything is always what it appears to be.

Advertisement

Advertisement

UFC White House live streams will be an MMA event like no other, as the Octagon takes up residence on the South Lawn of the White House in Washington, DC, to mark the upcoming 250th anniversary of the United States on July 4 — and, more specifically, President Donald Trump’s 80th birthday on Sunday. With headliners from the US, Georgia, Brazil and France, the event also known as UFC Freedom 250 is a surprisingly multinational affair.

While there will doubtlessly be no end of pomp and pageantry taking place over the weekend, there’s the serious matter of two UFC titles up for grabs in a co-main event. At the very top of the bill is Ilia Topuria vs Justin Gaethje, with the former defending his Lightweight title. The unbeaten champion is one of the most dominant fighters in the organization and is sure to be a mainstay for years to come, while for ‘The Highlight’ it’s surely the last opportunity to win the belt outright as Gaethje approaches his 38th birthday.

IQM Quantum Computers, the Finnish maker of superconducting quantum systems, has appointed Barbara Venneman to its board of directors as the company approaches what would be the first Nasdaq listing by a European quantum computing company. Venneman currently serves on the board of Vanguard, one of the world’s largest investment management firms, and previously led Deloitte Digital globally. The appointment signals that IQM is stacking its governance with public-markets experience ahead of a shareholder vote scheduled for June 25.

That vote will determine whether IQM completes its merger with Real Asset Acquisition Corp (Nasdaq: RAAQ), a special purpose acquisition company based in Princeton, New Jersey. The deal, announced in February at a pre-money valuation of $1.8 billion, would give IQM a primary listing on Nasdaq with a potential dual listing on the Helsinki Stock Exchange. The SEC declared the registration statement on Form F-4 effective on June 5.

The transaction has gained momentum since February. IQM and RAAQ announced an upsized PIPE of $146 million in early June, up from an initial $134 million commitment, after Finnish pension insurer Ilmarinen joined existing institutional investors. Combined with cash from RAAQ’s trust account, an earlier €50 million financing from BlackRock, and IQM’s existing balance sheet of $172 million, the company expects to hold more than $450 million in cash at closing.

IQM’s commercial profile sets it apart from the cloud-only quantum providers that have dominated the US public market so far. The company builds full-stack superconducting quantum computers for on-premises deployment, giving customers direct ownership of the hardware rather than API access to shared systems. It has sold 21 quantum systems to 13 customers, including four of the ten largest supercomputing centres in the world, and ships what it describes as 15% of all publicly disclosed quantum systems globally.

The company reported verified 2025 revenue of approximately €31 million ($36 million) and more than $100 million in cumulative bookings and order visibility. Founded in 2018 as a spinout from Aalto University and VTT Technical Research Centre of Finland, IQM now employs more than 400 people across Europe, Asia, and North America. Its technology partners include Nvidia, Hewlett Packard Enterprise, and AWS, which made IQM’s Garnet processor the first quantum system available via AWS cloud in the European Union.

Venneman brings more than 30 years of experience in digital transformation and enterprise technology commercialisation. She also serves on the board of advisors at Decagon.AI, an enterprise AI software company, and holds an MBA from McGill University and a computer science degree from the Université de Montréal. IQM’s board chair Sierk Poetting said her background in “enterprise technology commercialization, AI, and governance” would be valuable as IQM scales its commercial presence in the US.

The board also reshuffled its founder representation. CEO and co-founder Jan Goetz will replace co-founder Juha Vartiainen as the founder representative on IQM’s board, consolidating the company’s executive and governance leadership under the same person as it transitions to public-company reporting requirements.

IQM is not the only quantum company taking the SPAC route to public markets. Infleqtion listed on the NYSE earlier this month, and Horizon Quantum Computing has pursued a similar path. The SPAC structure carries well-documented risks, with the 2021 wave of blank-cheque listings producing widespread underperformance. IQM’s deal will test whether a European hardware company with real revenue, institutional backing from BlackRock and Finnish pension capital, and a $1.8 billion valuation can break that pattern.