The ongoing shortage of RAM chips is not only expected to persist into the next year, but will likely intensify in 2027, with tight supply conditions lasting until at least 2028, according to Samsung, which manufactures and supplies roughly a third of the world’s memory chips.

Samsung said in its Q2 earnings call that frontier AI labs, desperate to obtain access to memory infrastructure, have been “sharing their medium- to long-term demand forecasts” directly with the Korean tech giant to secure future supply.

The high demand enables Samsung to prioritize customers willing to sign long-term contracts. This multi-year visibility will allow the company to install equipment and ramp up production without worrying that demand will dry up, helping it avoid the memory industry’s historical boom-and-bust cycles.

Memory shortages driven by the AI boom have also pushed up chip prices in recent months. This has been a double-edged sword for Samsung: While sales at its semiconductor unit hit an all-time high in Q2, profitability in its smartphone and TV divisions shrank, as the higher-priced chips drove up component costs.

Samsung has even started passing some of those increasing component costs on to consumers by increasing the prices of its Galaxy smartphones and tablets. However, as a result, demand for these devices has dropped.

The memory shortage, informally dubbed “the RAMaggedon”, has also forced Apple, Samsung’s archrival, to raise the prices of its Macbooks, Macs and iPads last month. On its latest earnings call, Apple warned that revenue growth for the upcoming quarter is projected to slow to between 9% and 11% year-over-year, down from its recent 16% quarterly growth rate.

With memory manufacturers shifting production capacity toward AI data centers and away from consumer electronics, consumers are facing a new reality: higher device prices. Nvidia is expected to raise its consumer graphics card prices by 20% to 30%, which may further drive up prices of gaming devices, desktop computers, consoles, laptops, and the like.

SpaceX said on Thursday that it will remove the unpermitted turbines powering its xAI data centers near Memphis as it transitions to a permanent, 1.2 gigawatt natural gas power plant. 

The turbines won’t be completely removed until July 2027, though. SpaceX said that it’s currently operating 69 gas turbines to power the Colossus data centers, of which many have been operating for months. The NAACP and Southern Environmental Law Center have sued xAI over the use of unpermitted turbines. SpaceX acquired xAI in February. 

In its IPO filing, SpaceX said it plans to buy $2.8 billion worth of gas turbines for its data centers over the next three years.

SpaceX claims that it is allowed to operate the existing turbines without permits because they remain on the trailers they were shipped on. But federal regulations say that the turbines xAI has been using, regardless of what they sit on, require permits because of their size and how they’re being used.

The turbines are currently located south of Memphis in Mississippi, just over the border with Tennessee. The region is among the most polluted in the U.S., and xAI has been operating gas turbines that have the potential to emit more than 2,000 tons of smog-forming NO_x per year.

Last month, the Department of Justice sided with SpaceX in the NAACP’s lawsuit, saying the unpermitted turbines were a matter of “national, economic, and energy security.”

The new power plant that SpaceX is building will consist of 41 gas turbines ranging in size from 16.48 megawatts to 50 megawatts, according to permit documents issued by the state of Mississippi. They appear to be different from those currently in use, though TechCrunch could not verify the specific models of the 69 existing turbines.

Earlier this year, Elon Musk bought APR Energy, a company that specializes in temporary natural gas power. Based on an archive of the company’s site before it was taken down, the turbines in the company’s fleet also appear to be different from those cited in permits for the new, permanent power plant. 

Given that APR Energy’s turbines aren’t likely to be part of the new Colossus power plant, the new turbine fleet is likely intended for another, unannounced project.

While AI agents are increasingly capable of solving customer support problems, most people can still tell immediately when they’re talking to a machine instead of a human.

Smallest.ai, a startup founded in late 2024, is betting the next leap in voice agents will not come from making large language models faster, but from using smaller, specialized models built for human conversation. Simply put, the company wants to make speaking to an AI agent indistinguishable from talking to a human.

To do so, it’s developing a small voice model designed to mimic how humans process information by listening, thinking, and speaking simultaneously.

“While I’m speaking to you, you’re already thinking, and you might interrupt me if I talk for too long,” Sudarshan Kamath (pictured left), founder and CEO of Smallest.ai, told TechCrunch, adding that this is exactly how the startup’s model is designed to work.

To fuel this mission, Smallest.ai has raised $13 million in a Series A round, led by Seligman Ventures with participation from Sierra Ventures and 3one4 Capital. The fresh capital brings the startup’s total funding to over $21 million.

“The way an LLM works is you give it an entire prompt, and then it starts thinking,” Kamath said. While that latency is acceptable in a text chat, in a voice conversation, even a short pause feels unnatural. “If you think about how we are talking, I’m not giving you like a large clipping of my audio, and then you start thinking.”

The startup’s model serves as a real-time intelligence layer that enables natural customer conversations on specific topics, with virtually zero response lag. But if the model encounters a subject outside its limited knowledge base, Smallest.ai hands off the query to a large foundational model, briefly placing the customer on hold to “research” the issue—just as a real human would do.

Kamath believes that all AI agents will soon rely on two models: a small voice model for real-time interaction, and an “offline” LLM that is called upon as needed to solve complex problems.

Unlike large foundational models, Smallest.ai focuses strictly on voice-specific nuances, such as handling diverse accents, supporting dozens of languages, and operating in noisy environments.

The startup’s existing customers include companies in the voice space, including RingCentral and Truecaller. Kamath said that any customer support company, including newer ones like Sierra and Decagon, is a potential customer for the startup.  

When asked why a well-funded AI customer support company wouldn’t build its own voice model, Kamath said that for customer support startups, becoming “extremely good at doing voice is a distraction from their core business.”

Smallest.ai competes with voice AI leader ElevenLabs, as well as Cartesia and regional players like Sarvam that focus on local languages.

While some competitors apply voice AI to use cases, like audio dubbing and podcasting, Smallest.ai focuses strictly on real-time conversational voice agents for its enterprise customers.

“We want our models to break the Turing test,” Kamath said. “You should speak to our model and not know it’s AI or human. That’s the sole focus of the company.”