Opinion: AI, Hardware and the Future of Consoles

Let's run through a quick news thread: Western Digital sold its entire 2026 hard drive production before February even started. Seagate did the same the following week. The Steam Deck OLED vanished from shelves due to memory shortages. Sony is considering pushing the PS6 to 2028 or 2029. Nintendo is thinking about raising the Switch 2's price before its first year is even up because RAM ended up costing more than expected. Mario Kart World launched at $80 and was a hit. GTA 6 is expected to hit shelves at $80, with premium editions going over $100.

All of these news stories are, in some way, connected by the same — and not so indirect — theme: artificial intelligence. Or more specifically, the Big Tech race for AI infrastructure. Amazon, Google, Microsoft, Meta, and OpenAI have cleaned out the hardware market by building data centers at an unprecedented scale. In October 2025, OpenAI secured up to 900,000 DRAM wafers per month just for the Stargate project. Global production was at 2.25 million — one single project ate up 40% of it.

Samsung, SK Hynix, and Micron manufacture memory for everything: servers, phones, laptops, cameras, TVs, consoles, and, of course, AI. When a sector shows up paying top dollar and signing long-term contracts, production shifts direction, and every component that goes into an AI server — or is produced with Big Tech in mind — is one that isn't going into a console. In a shareholder meeting, the CEO of Silicon Motion put it bluntly: we're seeing something unprecedented in scale — HDs, DRAM, HBM, NAND, everything in severe shortage throughout 2026. IDC estimates data centers will consume 70% of global memory chip production this year. The CEO of Phison expects electronics manufacturers to start shutting down product lines as 2026 rolls on.

The gaming market picks up the tab. Console manufacturers don't have the same bargaining power as OpenAI. By the time they get in line, half the production's already been sold. Fusion Worldwide, a component distributor, summed it up: if you're not a tech giant trying to buy memory right now, you're competing against the best-funded supply chain operation in history.

At the end of the line are the players. The industry's consumers are the ones who will pay the final price for the unchecked rise in hardware costs and how they affect every stage of production and manufacturing for one of the 21st century's most popular hobbies — a hobby that depends, from start to finish, on the very same components the big tech and AI development companies so desperately need to expand.

AI models like GPT, Claude, or Gemini run on servers packed with high-performance processors, industrial-scale RAM, and specialized chips like GPUs and TPUs. A single dedicated AI server consumes fifty times more memory than a high-end home computer. A data center houses thousands of them, running 24/7, 365 days a year — in short, every time someone asks ChatGPT to summarize an email, turns a selfie into a Studio Ghibli-style image, or uses AI as a therapist, a server somewhere is doing the work. A physical machine, stored alongside thousands of others, consuming real energy and processing power.

ChatGPT had 100 million users in January 2023. By January 2025, it was over 300 million. Every interaction burns processing cycles, and as more complex tasks — which demand more hardware n— become doable with AI, its user base grows, straining infrastructure and demanding more data centers be built.

OpenAI, for example, launched Sora in December 2024, a tool capable of generating short videos from text. Processing billions of parameters in seconds requires a much larger infrastructure than a text-based interface like ChatGPT. At the same time, Google expanded Gemini 2.0. Meta integrated generative AI into Instagram and Facebook, reaching billions of users, and ClaudeAI released Claude Code.

If usage expands as AI covers more functions, and that increased usage demands building more data centers, demand chokes supply. When OpenAI locks down a contract for 40% of the world's DRAM wafer production, the rest of the market scraps over what's left, and consumers feel the impact in their wallets — RAM that used to cost $300 hits $800. Graphics cards sold for $500 in late 2023 could top $900 in 2026, and SSDs have shot up 35% in six months.

The shortage will persist as long as the AI infrastructure race continues. And beyond consumers trying to build their own PCs, the console industry is also grappling with the component crisis. Each manufacturer is trying to respond in its own way. None are responding well.

Nintendo launched the Switch 2 at $450 in June 2025 and sold 17 million before the shortage got worse. Now the bill's come due: RAM costs more than planned. Analysts at Niko Partners believe the company might have to drop the standalone version and only sell bundles above $500. They bought themselves some time with the launch, but margins could take a hit.

Without the massive contracts Sony and Nintendo have, Valve is at the mercy of the spot market. The Steam Deck OLED sold out due to memory shortages, and the Steam Machine's launch was delayed, and its price might be "revised" before it was even announced.

Microsoft spent $80 billion on AI data centers in 2025 and is expected to top $100 billion in 2026. In other words: they're part of the problem. At the same time, Xbox is enduring four straight years of declining hardware sales. Console revenue dropped 25% in fiscal 2025, and the company canceled old projects in a wave of layoffs tied to its pivot toward AI. In February, the gaming division got a new CEO: Asha Sharma, former president of Microsoft's CoreAI division.

AMD has confirmed that a new Xbox is in development for 2027, but the company's business no longer hinges on hardware: in 2024, Game Pass had 34 million active users; in 2025, it pulled in nearly $5 billion. By not relying on cheap hardware, Microsoft's current model is the one best equipped to survive the very shortage the company itself is helping to create.

Then there's the PlayStation 6 case. For three decades, Sony followed a pattern: PS3 in 2006, PS4 in 2013, and PS5 in 2020. Expectations were for a PS6 in 2027 — that's probably not happening. In February, Bloomberg reported that Sony is considering pushing the console to 2028 or 2029. Component costs would make any feasible price point unworkable for the average consumer. If 2029 is confirmed, the PS5 will have the longest cycle in history: nine years.

According to TrendForce, memory will account for 21% to 23% of total console manufacturing costs in 2026. If that holds, it kills an old industry strategy: console makers have always relied on gradually lowering prices to expand their user base. The PS4 Slim came out years later, cheaper, bringing in people who hadn't bought it at launch. With memory eating up a third of production costs, that strategy doesn't work anymore without taking a bigger loss on hardware — no price drop means the installed base stops growing at the rate the industry needs to justify current budgets.

The gaming industry grew for decades by getting consoles into more people's hands, in more countries, at accessible prices. For Sony, the PlayStation 2 sold 160 million units because it reached markets its predecessor barely touched. That process didn't repeat in the PS3 era and paved the way for the Xbox 360, but it came back around with similar results in the PS4 generation. The PS5 Pro at $699 had enough backlash that Sony isn't eager for an encore. A PlayStation 6 launching into a shortage market, with costs inflated by billion-dollar data center contracts, would arrive with a price tag few would be willing to pay amid a potential global economic crisis — projections are flirting with the $1,000 mark.

In developed markets, that price is a considerable hurdle. In emerging economies, it becomes an almost insurmountable barrier, where each new generation already arrives as a luxury item under normal conditions. This could be the first generation to actually shrink player base growth in the medium term, on a broad scale, when a new-gen console lands in certain countries at prices approaching five figures.

Brazil is a case in point. The PlayStation 4 arrived in 2013 for R$ 3,999 — the most expensive in the world that year. The minimum wage was R$ 678. Sony said they weren't making a cent: taxes and fees added up to R$ 4,257 per unit. The PS5 arrived in 2020 for R$ 4,999, with the minimum wage at R$ 1,045. The PS5 Pro, in 2024, cost R$ 6,999. If the pattern holds, the PS6 would land dangerously close to R$ 10,000.

Hardware cost is the main barrier, but it's not the only one. Software prices can also rise due to logistical factors: cartridges and discs use NAND memory. That's also in short supply and is projected to jump up to 33% in the first quarter of 2026.

For the Switch 2, this is a huge problem. Nintendo has positioned the console as a physical platform — 57% of games sold in 2025 were physical copies, according to Omdia. Every manufactured copy costs more. The response publishers have found is the Game-Key Card: an empty cartridge that works as a download key. It solves the margin issue for the manufacturers but pushes the infrastructure cost onto the consumer, who pays for physical media and still has to download the game.

A price hike for games has been expected for a few years now, and it's slowly becoming a reality. In June 2025, Mario Kart World hit stores at $80. There were protests and heated debates on social media, but the gamble paid off: 14 million copies sold by December of the same year.

It set the first precedent, but it won't be the only one. Grand Theft Auto VI — the most expensive game ever made, with a development budget confirmed by Take-Two to be over $1 billion — is projected to cost $80 for the standard edition, with premium editions exceeding $100. Once GTA 6 sets that bar, other publishers will watch the first week's sales. If they hold up — and they will; it's GTA — the industry will have its new benchmark, repeating the same process that normalized $70 games.

Swen Vincke, CEO of Larian Studios, stated in December that the RAM crisis is forcing the team to optimize the new Divinity title and will likely require extensive changes as early as the first Early Access build and that the crisis is "wrecking" the developer's plans for the new game. While he wasn't more specific about exactly which plans, we can get a sense of it.

Developing a AAA game takes, on average, four to seven years. What happens to games planned for hardware that gets delayed? Studios that started production in 2022 or 2023 made technical choices assuming that the PlayStation 6 would be on the market in 2027. With a possible delay to 2029, the game launches into a market without the platform it was built for. In that scenario, what are these studios supposed to do?

The options are all bad and force them into gambles that will play out years from now. If they develop and launch for the PS5, they risk the game arriving when the PS6 is already out, making the title immediately dated and less appealing. Waiting for the next-generation means holding the release for two or more years, with the financial cost that entails for teams of hundreds of people. Doing both means an additional port that wasn't in the original budget, with no guarantees the final product will be as well-received as planned.

Larian has the financial runway to make that choice carefully and work with very long development cycles to adapt to economic shifts. Others might not. And if the financial risk is too high, projects will be cancelled. Unfortunately, countless projects seem like risks right now.

The AI industry has a solution. Or claims to: cutting costs by using the very AI that's causing the crisis.

Asset generation, QA automation, dynamic NPCs, procedural environments — 90% of developers are already using some form of AI in their pipelines, according to a Google Cloud survey of 615 studios in 2025. The promise works in parts. AI cuts down time on repetitive tasks and lets smaller teams produce more.

But assistance can turn into a financial threat on the stock market. On January 30th, Google released Project Genie, which generates explorable 3D environments from text. It lasted a minute and was stable enough to change the geometry of a racetrack mid-session.

That same day, Unity plummeted 21%. Roblox fell 12%. AppLovin lost 13%. Take-Two — GTA 6's publisher — tumbled nearly 10%, its lowest value since the game was delayed in November. Investors saw a rudimentary demo and decided studios needed to pay attention. Strauss Zelnick, Take-Two's CEO, said he was "surprised" by the reaction and pointed out that AI tools aren't entertainment experiences but instruments that creators use to make entertainment.

The logic is sound, but the market didn't care. The mere hype of replacement is already moving billions in market capitalization, regardless of what the technology actually does in practice. On top of expensive hardware, the risks for the industry now include funding decisions, layoffs, and strategy being shaped by investor sentiment fed by 60-second demos before any generative AI tool has proven it can produce a minimally complete game. After all, the gap between generating an asset and building a coherent system, with gameplay design, art direction, difficulty balancing, narrative layers, and a hundred other challenges an experienced team solves over years, is enormous.

In the real world, the studio using AI to cut costs is indirectly paying for the same shortage that's making their machines more expensive, with a discount offered by the Big Tech companies that created the problem. Investors don't seem to see it that way.

There's an alternative route that the crisis could accelerate. Long console cycles produce some of the best games of a generation. The more developers understand the hardware, the more they can squeeze out of it: the late stages of the PS2, PS3, and Xbox 360 produced titles that defined their eras. If the PS5 runs for nine years, the next few years could bring more refined games — provided studios stay committed to the current platform instead of shifting resources to the next hardware too early, which the uncertainty around the PS6 makes difficult.

If visual improvement plateaus, more titles will need to focus on maximizing the capabilities of the PlayStation 5 and other consoles and engines that can handle the hardware available on the market. The trend could shift development focus to other differentiators. Done right, it could usher in a new period of high creativity in gameplay, story, and design.

There are strategic reasons to follow this path: the race for photorealism has always been a marketing gimmick for the next generation. Graphics sell easily in a 90-second trailer, but every layer of a game doesn't fit in a trailer and requires a complete product to be properly evaluated.

During the peak of the console wars, developers optimized for what could be shown, and budgets followed suit. Facing this crisis, the promise of "graphics that only run on the next generation" loses meaning when there's no release date, and current hardware lasts longer than expected. The player doesn't have the next console, doesn't know when they'll get it, and doesn't know at what price. The differentiator now is the game as a game, not how well a title replicates cinema or how many individual strands of hair a character has.

Another possibility, perhaps complementary, is focusing on smaller budgets with more emphasis on creative diversity, with cheaper prices for both developers and the end consumer. Clair Obscur: Expedition 33 proves the path works. Sandfall had a modest budget by AAA standards and delivered what most blockbusters can't — if it doesn't become a model by proving that less graphically ambitious games can produce great results, it might become one simply because it's what the market can handle in uncertain times.

There are caveats: the crisis removes one excuse for not pursuing quality through means other than graphics, but it doesn't guarantee anyone will actually pursue it. Not every budget constraint or projection for new generations will make an Expedition 33. It could also result in unfinished games rushed out due to financial necessity or poor planning. It could also lead to titles aiming for a similar budget while charging $70 or even $80 for the final product, instead of the $50 Sandfall asked for at launch.

Sony stated in its 2025 annual report that it intends to increase revenue per user rather than expand its customer base. CFO Lin Tao admitted in August that the company's bet on live service games is facing difficulties: Sony has cancelled eight of the twelve games announced for that model. Concord alone cost over $200 million, grossed one million, and was shut down two weeks after launch.

The company isn't abandoning the model because it already represents too large a slice of revenue, despite execution stumbles. Live services already account for 20% to 30% of first-party game revenue. PS Plus pulled in $1.5 billion per quarter in 2024.

Expensive hardware could push players toward these live services, but also toward subscription services. Microsoft's model — an extensive library and a fixed monthly fee, no console required — seems to work, despite concerning precedents from other markets sparking skepticism in some.

Video streaming emerged as a convenient alternative to video rental stores. A subscription offered broad access, eliminating the need to buy individual titles or rent tapes and DVDs. Over the years, titles left major aggregators like Netflix as studios created their own platforms. Each service started holding exclusives, and anyone wanting to watch everything now pays for multiple subscriptions and still doesn't have full access.

The gaming market already has similar services. Microsoft runs Game Pass. Sony has PS Plus. EA maintains EA Play. All three have raised prices in the last two years. Meanwhile, spending on game subscriptions in the United States grew 20% in 2025 compared to the previous year.

However, there are fundamental differences between subscribing to Netflix and any game subscription service: canceling a video streaming service has a low cost. The content stays on the platform, subscribers come and go, watch what they want, and leave. Nobody rewatches the same movies every day, so the loss is minimal.

Canceling a game subscription in a landscape of diversified services might mean giving up access to an entire library and everything that comes with it. Saved progress, achievements, communities, and hundreds of hours invested in a platform create financial and psychological ties. Leaving costs more as the sense of loss is greater. Staying brings more benefits, even while paying rising prices or losing titles from the catalog.

The incentive to exploit this dynamic grows as hardware becomes less accessible, but diversifying subscription services, as happened with video streaming, isn't as simple for games. Any smartphone or smart TV can stream Netflix or Amazon Prime. None of them require significant graphical capabilities, nor do they need optimized processors in a PC or console.

Games go the opposite direction: developers depend on platforms to sell their games and for people to play them, especially at a time when building your own PC is more expensive. Coexistence with console manufacturers is necessary for their products to be enjoyed by consumers. Trying to create their own catalog requires much more than just commercial interest — publishers don't have bargaining power in this relationship: any service they offer only reaches consoles because the platform allowed it, on the terms the platform dictates.

The good news is that the shortage won't last forever, but estimates are very conservative. The most optimistic horizon is 2027 for a gradual market easing. That means 2026 will be the worst year. In the most pessimistic view, gamers could spend the next four years without upgrading their PCs or see a new console as a distant dream.

SK Hynix plans to invest $13 billion to expand memory production, focused on HBM — the memory used in AI data centers. Micron has allocated $20 billion in 2026, with a factory in Idaho. However, new factories take two to three years to be operational: Micron confirmed that Idaho won't have an impact before 2028, and SK Hynix estimates the shortage should last until the end of 2027. The CEO of Phison predicts supply and demand won't balance until 2030 — and that by then, many electronics manufacturers will have gone bankrupt or abandoned product lines.

So far, there's still no transition fund, no agreement with console manufacturers, and no public acknowledgment that AI expansion has a concrete cost for industries that depend on the same components. For the gaming market, the price of the AI bubble is being paid before it eventually bursts. Startup funding in the sector has fallen to its lowest level in a decade: $627 million in the first half of 2025, compared to $2–3 billion in 2023 — the capital has migrated to AI.

The bubble exists on a scale the financial market hasn't seen since the dot-com era. OpenAI has committed $1.4 trillion to infrastructure over eight years, with $20 billion in annual revenue. The hyperscalers spent over $400 billion on AI infrastructure in 2025. Morgan Stanley estimates $3 trillion in global data center spending between 2025 and 2028, half financed by private credit.

If the bet doesn't pay off at the scale it was financed, the shock will hit every sector, and a market-correction-induced recession wouldn't arrive at a good time for an industry selling entertainment products costing hundreds of dollars. Cheaper chips in a contracting economy also don't mean more affordable consoles when consumers would have less income to buy them.

At the end of the chain, there's the player. Their fate in 2026 was decided in rooms where games were never discussed, by the infrastructure of a technology whose returns haven't been proven at the scale it was financed. And they're already paying the price before knowing if they'll be able to afford to keep up when — and if — the show begins.