September was a rough month for crypto investors, in particular for those betting big on ether, the token tied to the ethereum blockchain.
Etherdropped 13% for the month, its second-biggest monthly decline in the past year, behind only a 16% slide in June. Bitcoin fell 7% in September.
It’s difficult to link short-term price movements to any specific event, and with the historic rally in crypto over the past 12 months, pullbacks are to be expected. Ethereum, the second most-valuable cryptocurrency behind bitcoin, is still up about 830% in the past year.
Investors are now buying the September dip. On Friday, the first day of October, ether and bitcoin both climbed over 9%.
Ether 12-month price chart
CNBC
But the September roller-coaster reflects a particularly rocky stretch for the ethereum ecosystem, which has given investors and developers reasons for concern.
The speed of the network and high transaction fees continue to be a problem. The “London” upgrade in August was supposed to make transaction fees less volatile, but it’s had a limited effect.
Meanwhile, rival blockchains dubbed “ethereum killers” are taking advantage of ethereum’s challenges.
Ethereum also unexpectedly split into two separate chains in late August, after someone exploited a bug in the software that most people use to connect to the blockchain. That exposed the network to an attack, and not for the first time.
“All these factors could be having some impact on the speculation side, no doubt,” said Mati Greenspan, founder and CEO of Quantum Economics, in an interview. “But don’t forget that ethereum has appreciated quite handsomely so far this year and the entire market seems to be in consolidation at this time. So I wouldn’t try to read too deeply into these short-term movements.”
Still, ethereum, which serves as the primary building block for all sorts of crypto projects, like non-fungible tokens (NFTs), smart contracts and decentralized finance (DeFi), has some major hurdles to overcome to fend off the emerging competition.
Ethereum’s unexpected split
A central premise of ethereum’s security stems from the existence of only one set of virtual books, meaning you can’t create coins out of thin air. That ledger has to work, because the decentralized nature of the blockchain means there’s no rule keeper or bank that sits in the middle of transactions to act as accountant.
Ethereum developers were rightly alarmed in August when the chain split because of a bug.
“This fork temporarily created two separate records of transactions on the ethereum network – like parallel books,” said Matt Hougan, chief investment officer at Bitwise Asset Management, which created the first cryptocurrency index fund.
For a while, it was unclear whether the split would lead to a “double-spend attack,” where the same token can be spent more than once and transactions can be reversed, Hougan said.Smart contracts overseeing billions of dollars in assets could have also been at risk. Smart contracts allow people to build applications on top of ethereum with self-executing code, eliminating the need of third parties to handle transactions.
Such an attack would have been difficult to execute, since it was clear which nodes were on the correct side of the split and which were not. “But in theory, there was a risk,” Hougan said.
The good news for miners and exchanges is that most of them upgraded their software as recommended and the issue was resolved relatively quickly, said Tim Beiko, the coordinator for ethereum’s protocol developers.
Auston Bunsen, co-founder of QuikNode, which provides blockchain infrastructure to developers and companies, said it was a “responsibly disclosed vulnerability.”
“This is a reminder that blockchains in general and ethereum specifically are new and disruptive technologies,” Hougan said. “They can do amazing things – settle $1 billion transactions in minutes and program money like software – but they are not fully mature.”
Bugs keep happening
The longer-term problem for ethereum is that random glitches like this keep happening.
In April, the ethereum blockchain was hit with a bug in one of the software programs used to access it. And in November, many of ethereum’s DeFi apps temporarily went down after a Geth upgrade debacle, which led to the chain splitting in two.
Geth is short for for Go Ethereum. To access the ethereum blockchain, operators and miners have their pick of software. Most use Geth, which accounts for 64% of the network.
When the ethereum blockchain broke in half a few weeks ago, it was becauseGeth had a bug in its consensus mechanism. That’s what creates the single source of truth for transactions so everyone sees the same thing regardless of what software they’re using.
Developers discovered the bug, put out a new release with a fix and publicly told everyone to update. A lot of users upgraded, but others didn’t. When an unknown actor exploited the bug, ethereum forked, meaning that it broke into two separate chains: one for those who had updated their software and one for those who had not.
Ethereum “sought the veneer of decentralization by having many clients, but as a consequence, they have incompatibilities,” said Nic Carter, co-founder of blockchain data aggregator Coinmetrics.
When the software programs don’t talk to one another, it creates problems for the network.
Bitcoin takes a very different approach. It relies on a highly secure software program for nodes to access the blockchain. Bitcoin developers have long sought to avoid hard forks at all costs, so all changes in the core software tend to be opt in rather than pushed out to users,according to Carter.
“Ethereum prioritizes faster development, but that comes at the cost of a more fragile set of software implementations,” Carter said.
Some crypto experts attribute ethereum’s success to its first-mover advantage. Most NFTs and 78% of DeFi apps, or dApps, run on ethereum, according to the website State of The Dapps.
That’s starting to change, thanks to the growing popularity of rival blockchains.
Even before this latest split in the blockchain, users were complaining about ethereum’s heavy congestion and high transaction fees, which touched a record of $70 earlier this year, and just this week, bounced from $20 to $46 and back down to $32.
‘Ethereum killers’
At current prices, fees continue to drive some users away.
They’re turning to blockchains like Cardano, a platform used to build dApps, and Solana, whose native coin has risen nearly 4,800% since September 2020. Launched last year, Solana is gaining traction in the NFT and DeFi ecosystems because it’s cheaper and faster to use than ethereum.
Investors who had been largely focused on ethereum “have been increasingly diversifying their holdings to other cryptocurrencies, fueling alternative blockchains like Algorand, Solana and Cardano,” said Mark Peikin, CEO of Bespoke Growth Partners.
Bunsen tells CNBC that while Solana is making good strides in terms of being a usable blockchain, it’s not yet decentralized enough to satisfy the larger crypto community.
It’s also not immune to bugs. Last month, Solana suffered a 17-hour outage following a denial-of-service attack, which took the form of a flood of transactions caused by bots.
The list of so-called ethereum killers is long, and includes blockchains like Matic and Polygon, which are complementary to ethereum, according to Bunsen, as well Cardano, which is known for its security.
“I think some of those ethereum killers will make it,” said Bunsen. “But they won’t kill ethereum.”
Ethereum also has its own upgrade in the works. For several years, it’s been building ethereum 2.0, which is expected to be ready by the first quarter of 2022.
The makeover will move ethereum to a less energy-intensive mining process and, according to network founder Vitalik Buterin, could boost speed by over 7,000-fold to 100,000 transactions per second.
If it’s successful, Bunsen said, ethereum 2.0 will be a “huge upgrade in terms of throughput to the ethereum network and a huge win for the environment generally.”
Neptune and OpenAI have collaborated on a metrics dashboard to help teams that are building foundation models. The companies will work “even more closely together” because of the acquisition, Neptune CEO Piotr Niedźwiedź said in a blog.
The startup will wind down its external services in the coming months, Niedźwiedź said. The terms of the acquisition were not disclosed.
“Neptune has built a fast, precise system that allows researchers to analyze complex training workflows,” OpenAI’s Chief Scientist Jakub Pachocki said in a statement. “We plan to iterate with them to integrate their tools deep into our training stack to expand our visibility into how models learn.”
OpenAI has acquired several companies this year.
It purchased a small interface startup called Software Applications Incorporated for an undisclosed sum in October, product development startup Statsig for $1.1 billion in September and Jony Ive’s AI devices startup io for more than $6 billion in May.
Neptune had raised more than $18 million in funding from investors including Almaz Capital and TDJ Pitango Ventures, according to its website. Neptune’s deal with OpenAI is still subject to customary closing conditions.
“I am truly grateful to our customers, investors, co-founders, and colleagues who have made this journey possible,” Niedźwiedź said. “It was the ride of a lifetime already, yet still I believe this is only the beginning.”
A person walks by a sign for Micron Technology headquarters in San Jose, California, on June 25, 2025.
Justin Sullivan | Getty Images
Micron said on Wednesday that it plans to stop selling memory to consumers to focus on meeting demand for high-powered artificial intelligence chips.
“The AI-driven growth in the data center has led to a surge in demand for memory and storage,” Sumit Sadana, Micron business chief, said in a statement. “Micron has made the difficult decision to exit the Crucial consumer business in order to improve supply and support for our larger, strategic customers in faster-growing segments.”
Micron’s announcement is the latest sign that the AI infrastructure boom is creating shortages for inputs like memory as a handful of companies commit to spend hundreds of billions in the next few years to build massive data centers. Memory, which is used by computers to store data for short periods of time, is facing a global shortage.
Micron shares are up about 175% this year, though they slipped 3% on Wednesday to $232.25.
AI chips, like the GPUs made by Nvidia and AdvancedMicro Devices, use large amounts of the most advanced memory. For example, the current-generation Nvidia GB200 chip has 192GB of memory per graphics processor. Google’s latest AI chip, the Ironwood TPU, needs 192GB of high-bandwidth memory.
Memory is also used in phones and computers, but with lower specs, and much lower quantities — many laptops only come with 16GB of memory. Micron’s Crucial brand sold memory on sticks that tinkerers could use to build their own PCs or upgrade their laptops. Crucial also sold solid-state hard drives.
Micron competes against SK Hynix and Samsung in the market for high-bandwidth memory, but it’s the only U.S.-based memory supplier. Analysts have said that SK Hynix is Nvidia’s primary memory supplier.
Micron supplies AMD, which says its AI chips use more memory than others, providing them a performance advantage for running AI. AMD’s current AI chip, the MI350, comes with 288GB of high-bandwidth memory.
Micron’s Crucial business was not broken out in company earnings. However, its cloud memory business unit showed 213% year-over-year growth in the most recent quarter.
Analysts at Goldman on Tuesday raised their price target on Micron’s stock to $205 from $180, though they maintained their hold recommendation. The analysts wrote in a note to clients that due to “continued pricing momentum” in memory, they “expect healthy upside to Street estimates” when Micron reports quarterly results in two weeks.
A Micron spokesperson declined to comment on whether the move would result in layoffs.
“Micron intends to reduce impact on team members due to this business decision through redeployment opportunities into existing open positions within the company,” the company said in its release.
Microsoft pushed back on a report Wednesday that the company lowered growth targets for artificial intelligence software sales after many of its salespeople missed those goals in the last fiscal year.
The company’s stock sank more than 2% on The Information report.
A Microsoft spokesperson said the company has not lowered sales quotas or targets for its salespeople.
The sales lag occurred for Microsoft’s Foundry product, an Azure enterprise platform where companies can build and manage AI agents, according to The Information, which cited two salespeople in Azure’s cloud unit.
AI agents can carry out a series of actions for a user or organization autonomously.
Less than a fifth of salespeople in one U.S. Azure unit met the Foundry sales growth target of 50%, according to The Information.
In another unit, the quota was set to double Foundry sales, The Information reported. The quota was dropped to 50% after most salespeople didn’t meet it.
In a statement, the company said the news outlet inaccurately combined the concepts of growth and quotas.
Read more CNBC tech news
“Aggregate sales quotas for AI products have not been lowered, as we informed them prior to publication,” a Microsoft Spokesperson said.
The AI boom has presented opportunities for businesses to add efficiencies and streamline tasks, with the companies that build these agents touting the power of the tools to take on work and allow workers to do more.
OpenAI, Google, Anthropic, Salesforce, Amazon and others all have their own tools to create and manage these AI assistants.
But the adoption of these tools by traditional businesses hasn’t seen the same surge as other parts of the AI ecosystem.
The Information noted AI adoption struggles at private equity firm Carlyle last year, in which the tools wouldn’t reliably connect data from other places. The company later reduced how much it spent on the tools.