Here’s what’s on our plate today:

• 🧪 How AI's appetite for memory reached Apple's price tags.

• 📰 AI's real cost warning, NVIDIA speeds up Claude for science, chatbots spread vaccine myths.

• 💡 Roko's Pro Tip: budget for memory staying expensive; the old cheaper-every-year era is over.

• 🗳️ Poll: What happens next with AI-driven memory prices?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

• Apple broke the silence: on June 25, 2026, It raised prices across Macs, iPads, and accessories, citing an unprecedented spike in component costs, while its stock dropped about 6%, its worst single-day decline in over a year.

• HBM is eating the supply: AI accelerators need high-bandwidth memory (HBM), which is far more profitable to produce than standard DRAM, so Samsung, SK Hynix, and Micron are shifting factory capacity toward HBM and starving consumer devices of chips.

• The margins prove it: Micron's gross margin jumped to nearly 85% from 39% a year earlier, while contract DRAM prices rose as much as 98% in the first quarter of 2026 alone.

• Small players are getting crushed: GoPro warned it could go out of business, Sonos shares fell 23%, and budget smartphone makers are getting frozen out, with IDC warning that memory suppliers are "only answering calls of the big players.”

• The stakes: if AI's appetite keeps outpacing new factory supply through 2027, today's price hikes stop being a cycle and become the new floor for what every device costs.

How AI's appetite for memory reached Apple's price tags

Just as the human brain performs tasks in the present while constantly drawing on memories to make decisions, computers also rely on two distinct systems to function. One is the processor, the chip that carries out calculations and executes instructions. The other is memory, which stores the information the processor needs to access instantly while working. Without that second component, even the fastest processor would have nothing to think with.

For most of the personal computing era, memory was the least glamorous part of a computer. The chips that temporarily hold data while a machine is running, known as DRAM (dynamic random-access memory), and the flash storage that preserves files even after the power is switched off, quietly did their job inside every laptop, smartphone, and server. While processors stole the spotlight during product launches and marketing campaigns, memory remained largely invisible, treated as a commodity rather than a selling point. That was partly because memory chips became steadily cheaper and more abundant each year, allowing manufacturers to ship every new generation of devices with more RAM and storage at roughly the same price as the previous generation.

The market itself was equally predictable. Samsung, SK Hynix, and Micron together controlled more than 95% of global DRAM production, supplying virtually every major electronics manufacturer. Smartphone makers, PC companies, cloud providers, and automakers all drew from the same pool of memory, with prices rising and falling according to familiar cycles of consumer demand. For decades, that balance held.

But artificial intelligence has begun to rewrite those rules. As AI systems grow more capable, they also become far more dependent on memory, not simply to store data but to keep enormous models, context windows, and intermediate calculations instantly accessible while generating responses. The result is a fundamental shift in the economics of computing, transforming memory from a commodity into one of the industry's most valuable strategic assets. On June 25, 2026, that shift reached consumers in an impossible-to-ignore way when Apple substantially increased the prices of memory upgrades across its Mac lineup, highlighting how AI's insatiable demand for memory is beginning to reshape the cost of personal computing itself.

What Apple made visible

That morning, Apple's online store briefly went dark and returned with new prices across nearly its entire hardware lineup. The entry-level MacBook Neo rose to $699 from $599, the cheapest iPad to $449 from $349, and the iPad mini, Apple TV, HomePod, and Vision Pro all climbed as well. Apple, which rarely raises prices mid-cycle and rarely without adding features to justify it, told customers it had "never seen a component price increase this much, this quickly."

Following the announcement, the iPhone maker's shares fell about 6%, their worst single-day decline in more than a year, as the new prices took effect globally, with the iPhone notably spared.

Within hours, Microsoft announced that Xbox console prices would increase by $100 to $150 starting August 1, 2026, and that it would discontinue its 2 TB model. The company said the cost of console memory and storage had risen more than 2.5 times and warned that prices could double again by the fall of 2027.

Taken together, the announcements transformed what had largely been an industry story buried inside supplier earnings calls and semiconductor research reports into something every consumer could see. Apple had spent months absorbing rising memory costs to protect the margins it is famous for maintaining, but eventually, the economics stopped working. Days earlier, CEO Tim Cook had described the situation as a "hundred-year flood," saying higher prices had become unavoidable. The increases revealed a market shift that had been quietly building since early 2025, as AI companies buying memory in enormous volumes began competing with the rest of the technology industry for a supply that was no longer sufficient for everyone.

Why has memory stopped getting cheaper?

The reason memory chip prices have surged has little to do with the semiconductor industry's familiar boom-and-bust cycle, where shortages push prices higher, manufacturers build more capacity, and prices eventually fall again. This time, the problem is different. According to IDC, the industry is undergoing a shift that could be lasting in how the world's chipmaking capacity is allocated. Instead of producing memory for smartphones, PCs, and other consumer electronics, manufacturers are increasingly prioritizing the specialized memory needed to power AI accelerators, the chips that train and run today's most advanced AI models.

The shift is being driven by a specific type of memory called high-bandwidth memory, or HBM. Unlike the DRAM found in laptops and smartphones, HBM is built by stacking multiple memory layers and placing them next to AI accelerators inside servers. This allows AI chips to access massive amounts of data at extremely high speeds, something essential for training and running today's large AI models.

HBM is also far more profitable to manufacture than conventional memory. Since the same chipmaking facilities produce both, no production line devoted to HBM can be used to make the DRAM that ends up in consumer devices. Faced with that choice, Samsung, SK Hynix, and Micron have increasingly shifted their manufacturing capacity toward HBM, where demand is stronger, and profit margins are significantly higher. That decision makes business sense, but it also means there is less conventional memory available for PCs, smartphones, and other electronics, pushing up prices across the industry.

The financial results show just how dramatic that shift has become. In its most recent quarter, Micron's revenue more than quadrupled from a year earlier, while its gross margin jumped to nearly 85%, up from 39% a year ago, underscoring how AI has transformed memory from a low-margin commodity into one of the semiconductor industry's most profitable businesses.

Why the smallest makers feel it first

This means that memory chip makers now have the luxury of deciding who gets their products and who has to wait. Because supply is concentrated in the hands of just three companies and much of their future production has already been reserved through long-term contracts, there is little room for buyers to negotiate. Micron alone says it has secured $22B worth of long-term commitments, while analysts do not expect meaningful new supply until additional factories begin operating in 2027 or 2028.

The shortage has sent prices soaring, and market researcher TrendForce estimated that contract DRAM prices rose by as much as 98% during the first quarter of 2026, with another 58% to 63% increase expected in the second quarter. In a market this tight, suppliers naturally prioritize their largest and most profitable customers. As Nabila Popal, senior research director at IDC, described it, the situation has become an "absolute existential crisis" for smaller Android smartphone makers and manufacturers producing devices priced below $100 because "memory suppliers are only answering calls of the big players."

The effects are already rippling through the technology industry. GoPro, the action-camera maker, warned investors it might go out of business after memory costs jumped between 80% and 115% late in the first quarter, and after suppliers told it in April they were cutting production of the memory its products rely on. Sonos, the speaker company, has watched its shares fall 23% this year under the same pressure. The cost is not confined to gadgets sold on shelves: a small communications-equipment firm supplying defense contractors told CNBC that a server it bought for $5,373 in 2020 now runs just under $15,000, with delivery whenever supply allows.

Larger technology companies have been better positioned to absorb some of the shock or pass it on to customers. PC manufacturers such as Lenovo, Dell, and HP have already signaled price increases of between 15% and 20%, while IDC expects global smartphone and PC shipments to decline this year as higher memory costs suppress demand rather than bring the market to a standstill.

When a foundation rearranges itself

The bigger shift lies in what happens when one of computing's most fundamental building blocks begins serving a single customer above all others. For decades, the consumer electronics industry operated on the assumption that memory would become cheaper, faster, and more abundant every year. That expectation shaped product roadmaps, pricing strategies, and the promise that each new generation of devices would deliver more for roughly the same price.

AI infrastructure has turned that assumption on its head. Data center builders and AI model operators have become the priority customers for memory manufacturers, while smartphones, laptops, gaming consoles, and other consumer devices increasingly compete for what remains. Whether this proves to be the industry's new normal or merely a temporary distortion depends on unanswered questions, such as whether AI demand will continue to grow at its current pace. Will the new semiconductor fabs expected later this decade meaningfully expand supply? Or will device makers quietly adapt by shipping products with less memory, replacing visible price hikes with a slower, less noticeable decline in what consumers get for the same money?

For the better part of four decades, memory was the part of a computer that no one had to think about. It quietly became cheaper, denser, and more abundant, enabling every new generation of devices to become more capable without increasing cost. The price increases in late June changed that, bringing one of computing's most invisible components into public view. Infrastructure, after all, tends to attract attention only when it stops behaving as expected.

Tim Cook's description of the situation as a "hundred-year flood" captures the scale of the disruption, but it also raises a more uncomfortable question. If the companies controlling the world's memory supply are earning record profits from a market reshaped by AI demand, what incentive do they have to return to the old normal? Whether this proves to be a temporary flood or the beginning of a permanently higher watermark for memory prices may ultimately determine not just the cost of AI, but the price of every device that depends on it.

Roko Pro Tip

Bite-sized Brains

• AI's real cost warning: New numbers from Google and Amazon are flashing a warning sign about the true, mounting cost of running AI at scale, raising fresh doubts about the economics behind the boom.

• NVIDIA speeds up Claude for science: NVIDIA's BioNeMo is now accelerating Anthropic's Claude for scientific research by pairing frontier models with specialized tools to advance drug discovery and biology.

• Chatbots spread vaccine myths: A new poll finds AI chatbots are amplifying anti-vaccine misinformation, a fresh reminder of how easily these systems repeat and spread health falsehoods.

Monday Poll

Meme Of The Day

The Toolkit

• Assembly AI: Speech-to-text API that handles transcription, speaker detection, and audio intelligence for production apps. 

• Chroma: Open-source vector database built for AI apps, fast to set up and easy to scale for RAG and embeddings. 

• Continue: Open-source AI code assistant that integrates with VS Code and JetBrains, with full control over models and context.

Rate This Edition