Anthropic’s IPO and Meta’s (META.US) $95 Billion Fundraising: Cryptocurrency Undercurrents Amid the AI Capital Surge
Anthropic has filed to go public, Meta has announced a fundraising plan worth $95 billion, and Cadence has introduced an AI-powered model for chip design. CoinDCX has revealed a wallet security problem, Bitmine has bought back its own shares and increased its cryptocurrency holdings, and the U.S. Office of the Comptroller of the Currency (OCC) has encouraged banks to issue stablecoins.
According to Woofun AI, the rapid growth of investment in artificial intelligence (AI) is changing the market on an unprecedented scale. Anthropic's plan to go public and Meta (META.US)'s large fundraising plan show the direction of the industry. Meanwhile, the cryptocurrency industry is trying to find a new balance between managing business risks and using money efficiently.
On October 26, Anthropic filed an S-1 registration statement with the U.S. Securities and Exchange Commission, officially beginning the process of going public through an initial public offering (IPO). The company plans to list on the Nasdaq under the ticker symbol ANT. This means that a leading company that builds advanced AI models is preparing to enter the public stock market, where investors and regulators will examine its business more closely. According to its prospectus, the company reported revenue of $16.286 billion in the first half of this year and a net loss of $80.748 billion. Of that loss, $72.388 billion came from changes in the estimated fair value of convertible notes and other financial instruments, rather than actual cash payments. Its operating loss was $693 million, and it spent about $8.945 billion on computing power and infrastructure. The company has not yet announced how many shares it will issue or the price range for those shares. Now that its financial information is public, investors can begin reviewing the figures. However, the listing date, size of the offering, and company valuation will depend on later filings. Media predictions should not be treated as confirmed results. When assessing the company, investors need to consider growth in business customers alongside the cost of running AI models, commitments to computing resources, and expected cash returns. Anthropic's advanced models may help it win contracts, but having strong models alone does not guarantee that each contract will make a lasting profit.
Together, the two parts of Meta's plan are worth about $95 billion, but the company has not received the full amount. The fundraising plan shows that expanding AI infrastructure requires companies to use several sources of money at once, including cash from their businesses, debt, and shares. Raising money by selling shares reduces the need to take on new debt with fixed interest payments, but it can also reduce existing shareholders' ownership percentages. Mandatory conversion arrangements delay some of this effect. Like companies that raise money to build cryptocurrency reserves, these businesses need to explain how the additional funding will improve earnings per share. However, their businesses and sources of cash are different. Investors should focus on whether the cash invested in expansion will earn more than the cost of obtaining that capital, rather than simply comparing how much money each company raises.
Data compiled by Woofun AI shows that Cadence (CDNS.US) introduced a specialized group of AI models called Maestro on October 26 for chip and system design. The company plans to give selected customers early access in the fourth quarter of this year. Cadence said Maestro is partly built on NVIDIA's open-source Nemotron model weights. It can work with design agents, electronic design automation (EDA) tools, and reinforcement learning systems. Customers can run it in the cloud or on their own computers and servers. According to Cadence, early tests showed up to a 30% improvement in result quality and up to five times better token efficiency compared with general-purpose foundation models. However, the company did not disclose the exact test conditions or comparisons used, so it is too early to assume these improvements will apply to every design project. This approach brings AI competition into a field where results can be checked against clear engineering requirements. Chip designers can use tools to test whether a design meets timing, power consumption, and other limits. AI agents can then improve their suggestions based on actual engineering results instead of relying only on whether their answers sound like those of an expert. The option to run Maestro on a company's own systems also helps address concerns about protecting design data and confidential information. For the industry, how well these models work with existing tools may matter more than their rankings on general-purpose tests. More real-world evidence is needed to determine whether they can reduce testing costs, improve error handling, and shorten delivery times.
Regarding business risks, CoinDCX reported on October 26 that it had confirmed a loss of about 15.99 million USDT. The company said some users continued sending funds to old wallet addresses that had already been taken out of service. The system responsible for collecting these funds had also stopped working for a period, allowing assets to build up. These addresses had been affected by a security incident in July 2025. The admission came only days after the company first denied that there was a problem. CEO Sumit Gupta had described the activity recorded on the blockchain as "a false alarm." Blockchain researcher ZachXBT later pointed out that at least one of the addresses was still sending funds to the exchange's hot wallet between April and May 2026. Following this finding, the company published a blog post acknowledging the situation. The announcement did not fully explain how the funds were taken or who was responsible. The main problem involved how the exchange had shut down its older systems. Simply removing a deposit address from view does not stop that address from receiving funds on the blockchain. If a system continues recording incoming deposits, the company must keep checking the total balances and making sure that access permissions are correct. CoinDCX maintains that user balances and active systems have not been affected, in line with its public statements. For businesses that hold customers' assets, monitoring old wallets, investigating unusual activity, and reporting problems on time are just as important as protecting new wallets. If past security risks have not been fully addressed, they could cause further losses. Meanwhile, on October 26, Bitmine (BMNR.US) announced that it had bought back about 2.04 million of its own shares over the previous week. The average price was $22.48 per share, bringing the total cost to about $45.9 million.
During the same period, Bitmine also bought 10,452 ETH. According to its disclosures, the company held 6,051,933 ETH as of October 25. It spent about $24.6 million on these ETH purchases during the period, less than half of the amount used for share buybacks. Its weekly ETH purchases fell from 12,535 to 10,452, marking a slowdown for the second week in a row. This suggests that the company is directing more of its money toward buying back its own shares rather than dividing its spending evenly between share repurchases and cryptocurrency purchases. The figures offer a clear example of the funding pressures faced by companies that hold cryptocurrency reserves, as discussed yesterday. Management argues that when a company's shares are undervalued, buying them back may be a better use of money than directly purchasing more cryptocurrency. However, the value of this approach depends on the price paid for the shares, the company's net asset value, and where the money comes from. Share buybacks also leave less cash available for future token purchases and other spending. To judge the value of a company that holds cryptocurrency reserves, investors need to look not only at the number of tokens it owns but also at the number of shares outstanding and its financial obligations. Looking only at the growth of its asset holdings may hide whether the value of each share has improved and whether the company's cash reserves have become too small.
On regulation and security, the Office of the Comptroller of the Currency (OCC) has encouraged community banks to take part in stablecoin payments. On October 26, the OCC invited these banks to apply for permission to issue stablecoins, either directly or through subsidiaries. This shows that regulators are encouraging banks to enter the stablecoin market, but each bank must still complete the required applications and compliance checks. In South Korea, officials are discussing whether to delay taxes on cryptocurrency-related gains further, until 2029. Korean media reports on October 26 said that limited access to data on cross-border transactions remains one of the main problems being considered. Any delay would still require changes to the law by the National Assembly, so it cannot yet be treated as an approved policy. Meanwhile, the National Basketball Association (NBA) and the Commodity Futures Trading Commission (CFTC) have signed a memorandum of understanding to share information about prediction markets. On the same day, they named Kalshi and Polymarket the league's "preferred prediction market exchanges." They also required league officials and team employees to avoid trading contracts linked to NBA events.
On October 26, the U.S. Commodity Futures Trading Commission announced that the NBA and the agency would communicate regularly and share information to help prevent insider trading, fraud, and similar problems. However, this information-sharing agreement does not automatically settle disagreements between state and federal authorities over how prediction markets should be regulated. Separately, Electrum has fixed a security weakness in its transaction-acceleration process that could allow funds to be stolen. Reports on October 26 explained that malicious servers could create fake information about pending transactions. When users tried to speed up their transactions by increasing fees, they could be tricked into signing payments that sent money to an attacker instead. Developers publicly disclosed the problem on October 24 and said that version 4.8.3 contains the fix. They also confirmed that at least one user had lost money because of the vulnerability.