When someone asks me for gadget buying advice, I normally tell them to stick with their current device. In 2025, most new tech products aren't a worthwhile upgrade over even something that was released a few years ago — and with the price of everything going up, that new iPhone can wait. But things aren't normal right now. 

On December 3, The Wall Street Journal reported memory manufacturer Micron would wind down Crucial, its consumer business, to focus on components for the AI industry. The PC I'm writing this article on has an SSD and RAM from Crucial. Overnight, Micron decided to end a business it spent decades building, and from a certain perspective, I guess it makes sense. In recent months, OpenAI has signed more than $1.4 trillion worth of infrastructure deals, creating unprecedented demand for server-grade solid-state storage and RAM. 

To meet the moment, manufacturers have been allocating more of their production capacity and wafers to high-margin commercial customers. For consumers, the result has been skyrocketing RAM prices, with some DDR5 kits now costing as much as two or three times as much as they did a couple of months ago. Recent analysis from TrendForce shows the price of some consumer-grade SSDs increased between 20 and 60 percent in November for the same reason. Then there's LPDDR5X memory, which is used in both smartphones and NVIDIA's Grace Blackwell and Vera Rubin platforms. In 2026, it's expected to increase in price as well. The demand for AI infrastructure is such that all consumer electronics may cost more in the coming months.

That gets me to the purpose of this article. If you've been thinking about upgrading to a new graphics card, I would recommend you buy one sooner rather than later. The AI boom came for RAM first, and there are already signs it will come for GPU pricing next. A recent report suggests AMD is considering raising the MSRP of its 8GB models by $20 and 16GB models by $40 due to the price of GDDR6 memory. NVIDIA, meanwhile, is rumored to have recently told its board partners it would no longer supply them with VRAM for their cards.     

Neither NVIDIA nor AMD responded to comment requests from Engadget requesting they share how they plan to work with their board partners to ensure GPU prices remain stable. NVIDIA also did not comment on reports the company will stop providing VRAM to its board partners. 

Separate from the memory shortage, neither NVIDIA nor AMD are expected to release new GPUs soon. According to recent rumors, the earliest a Super refresh of the Blackwell line could arrive is sometime in the middle of 2026 — not at CES in January as the 40-series Super cards did in 2024. The memory crunch could complicate things there too, since the company has typically relied on more and faster VRAM to offer better performance on its Super cards. With 50-series Super GPUs, it might not be the case that NVIDIA announces them at the same MSRP as their non-Super predecessors, which was the case with the 40-series. 

As for AMD, the company debuted its RDNA 4 cards at the start of the year. We know it's already working on RDNA 5, and if a recent chat with Sony's Mark Cerny is any indication, the new architecture will be a major step change for AMD. However, right now rumors indicate the earliest RDNA 5 could arrive is sometime in 2027.

In other words, with nothing new on the horizon and pricing of existing stock likely to increase, there might be only a short window where you can get a new GPU at a reasonable price. It's impossible to predict the future, but if you're in need of an upgrade and have the means to purchase, there might not be a better opportunity before the end of 2026.  

Recommendations

The recommendations in Engadget's recent GPU guide are still as relevant today as they were a few months ago. Once again, the best advice I can give is to buy a card with at least 12GB of VRAM, and preferably 16GB if your budget allows for it. Unless you mostly plan to play older games on a 1080p monitor, it's not worth considering a model with 8GB of VRAM — it won't last you long enough to warrant the purchase price. 

Our recommendations are grouped from most affordable to most expensive. Where possible, I've tried to find options from both Newegg and Amazon. You won't find any high-end picks like the RTX 5080 since if you can afford that card, this guide isn't for you.  

Intel Arc B580

Devindra Hardawar for Engadget

For those on a tight budget, I would start and end my search with the Intel Arc B580. Newegg has models from ASRock and Onix at or under the card's $250 MSRP. I can't speak to the quality of ONIX cards, but ASRock is well-regarded. Over on Amazon, you can find the B580 for $300. With Intel cards you sometimes need to put up with odd driver issues, but as far as budget options go, the B580 offers value that's hard to beat. The one thing about budget cards like the B580 is they’re likely to face the most pricing pressure from the memory crunch due to the smaller margins manufacturers are making on them.   

NVIDIA RTX 5060 Ti 16GB

Devindra Hardawar for Engadget

If you have more than $250 to spend on a GPU, the RTX 5060 Ti is the GPU to buy. Avoid the 8GB model and go straight for the 16GB variant. NVIDIA announced the 5060 Ti at an MSRP of $429, and luckily as of the writing of this article, you can still find one close to that price.

Newegg, for instance, is selling the MSI Ventus Black Plus version of the card for $440. Amazon has the silver colorway of that same GPU listed for $460 currently. The retailer also has models from Gigabyte and Zotac in and around that same price. 

If I had to pick between the 5060 Ti and 5070, which NVIDIA only offers with 12GB of VRAM, I would pick the former. The 5060 Ti is a safer bet, and offers nearly as much performance, particularly in games that include ray tracing as an option.    

AMD Radeon RX 9070 and RX 9070 XT  

Devindra Hardawar for Engadget

For a mid-range option, the Radeon RX 9070 and 9070 XT offer excellent value. Of the two cards, the 9070 is the better purchase for most people due to its less demanding power requirements, but if you got a PSU that can handle the 9070 XT, go for it. 

Right now, Newegg has a few 9070 models from ASRock and Sapphire just under the card's $549 MSRP. My friend recently bought the Sapphire card linked above, and has had nothing but good things to say about it. You'll pay more going through Amazon, but the company has a couple of options around $600 from XFX and Gigabyte. 

When it comes to the 9070 XT, Newegg has an ASRock model priced right at the card's $599 MSRP. Many of the other options from Sapphire and XFX are unfortunately priced between $650 and $700. The same is true on Amazon, where the cheapest model I could find was $630.   

NVIDIA RTX 5070 Ti

Devindra Hardawar for Engadget

For our final recommendation, consider the RTX 5070 Ti. It's a great option if you want to play games at 4K for less than what the 5080 and 5090 cost. Newegg has MSI and Zotac models on sale for $750, the card's recommended price. There are also a handful of other options from ASUS and Gigabyte that are just over $800. Amazon, meanwhile, is selling one Gigabyte variant for $749. 

I sat down to write this story hoping to point people to deals on RAM kits for their gaming PC builds, but after an extensive search, I'm sorry to say there aren't any promotions worth even considering. Sure, if you visit Newegg or Amazon, you'll find plenty of bundles listed as Black Friday specials, but following a quick visit to PCPartPicker or CamelCamelCamel, you'll find many of those aren't deals at all. 

Take this 32GB kit of 6,000MHz DDR5 RAM from G.Skill I found on Newegg. It's one of the few "good" deals I found, but there's a catch. It's listed at $20 off its current $400 MSRP, with an additional $30 off if you use a promo code. To sweeten the deal, Newegg is even throwing in a NZXT all-in-one liquid cooler valued at $160. But here's the thing, according to PCPartPicker, that same kit cost $155 a couple of months ago. Unless you can make use of the free cooler, you don't need me to tell you $50 off a RAM kit that used to cost less than half of what it costs now is not a great buy. 

And it's not just that one set of sticks from G.Skill — nearly every kit of DDR5 RAM I could find has increased in price in recent months. For instance, Amazon has listed this 32GB bundle from Crucial at $322. A little more than a month ago, you could get that same kit for $175. 

If it feels like the pandemic all over again, when it was impossible to buy a GPU at MSRP, you're not far off the mark. Once again, there's a component shortage, but this time around, it's not cryptomining causing an insatiable demand for parts. Instead, it's the booming AI industry buying up every RAM stick it can for their data center builds. Unless you've been living under a rock, it's been hard to ignore the amount of money that's been thrown around by NVIDIA, Microsoft and others.  recently. Much of it has been of a seemingly circular nature, but that hasn't done anything to dent demand for both short and long-term memory. 

The problem is that many of the companies that produce consumer-grade memory, including heavyweights like SK Hynix and Samsung, also make memory for AI servers, and most of the RAM coming off those production lines is going straight to high-volume clients like OpenAI and Anthropic. In fact, demand for RAM has been so strong that even the price of some DDR4 memory kits has gone up — if you can even find the older format in stock.      

A statement CyberPowerPC posted earlier this week gives a sense of just how dire the situation is right now. "Recently, global memory (RAM) prices have surged by 500 percent and SSD prices have risen by 100 percent," the company said on X, adding it would be forced to increase the price of its pre-built PCs.

As a consumer, there's little you can do at the moment. You can either buy now and pay extra, or wait in hopes the price of RAM will stabilize sooner rather than later. Unfortunately, with no signs of the AI boom cooling down in the immediate future, it's hard to know when things might change. Save a bubble pop, the price of RAM is likely to remain high into early 2026, with the possibility of a trickle effect to SSD and GPU pricing.

If you were hoping to create some silly images this long holiday weekend with Google's new Nano Banana Pro model, I have some bad news: the company is restricting free usage of the AI system. In a support document spotted by 9to5Google, Google notes free users can currently generate two images daily, down from three per day previously. "Image generation and editing is in high demand," the company writes. "Limits may change frequently and will reset daily." 

It would appear Google is also limiting free Gemini 3 Pro usage, with the document stating non-paying users will get “basic access — daily limits may change frequently" as well. When the company first began rolling out Gemini 3 Pro on November 18, it guaranteed five free prompts per day. That was in line with Gemini 2.5 Pro. If you pay for either Google AI Pro or AI Ultra plan, your usage limits have not changed. They remain at 100 and 500 prompts per day, respectively. 

Google isn't the first company to enforce stricter usage following a popular release. You may recall OpenAI delayed rolling out ChatGPT's built-in image generator to free users after the feature turned out to be more popular than anticipated. However, OpenAI eventually brought image generation to free users. 

Well, I suppose it was only a matter of time, but Google is making AI Mode harder to avoid. In the US, the company has begun rolling out an update for Chrome on Android and iOS that adds an AI Mode shortcut to the browser's new tab page. It's predominantly featured, appearing right below the browser's signature search bar.

"This will let you ask more complex, multi-part questions, and then dive even deeper into a topic with follow-up questions and relevant links," the company said of the update. In the near future, Google plans to bring the shortcut to 160 additional countries, with support for other languages — including Hindi, Indonesian, Japanese, Korean, Portuguese — on the way as well. 

Google introduced AI Mode at the start of March when it previewed the feature through its Labs program. Since then, it has been aggressively rolling out AI Mode in nearly every market it operates, beginning this past May at I/O 2025 May when the company made the chatbot available to all US users.

Amazon is doubling its investment in Anthropic. The e-commerce giant will provide Anthropic with an additional $4 billion in funding on top of the $4 billion it committed last year. Although Amazon remains a minority investor, Anthropic has agreed to make Amazon Web Services (AWS) its “primary cloud and training partner.”

Before today’s announcement, The Information had reported that Amazon wanted to make any additional funding contingent on a commitment from Anthropic to use the company’s in-house AI chips instead of silicon from NVIDIA. It appears Amazon got its way, with both companies noting in separate press releases that Anthropic will use AWS Trainium and Inferentia chips to train future foundation models.

Additionally, Anthropic says it will collaborate with Amazon’s Annapurna Labs to develop future Trainium accelerators. “Through deep technical collaboration, we’re writing low-level kernels that allow us to directly interface with the Trainium silicon, and contributing to the AWS Neuron software stack to strengthen Trainium,” the company said. “Our engineers work closely with Annapurna’s chip design team to extract maximum computational efficiency from the hardware, which we plan to leverage to train our most advanced foundation models.”

According to another recent report, Anthropic expects to burn through more than $2.7 billion before the end. Before today, the company had raised $9.7 billion. Either way, it’s bought itself some much-needed runway as it looks to compete against OpenAI and other companies in the AI space.