Is AI open yet—and can it make money? – October 7 technology and markets briefing

I look past the biggest headline numbers to separate what is known from what is still an expectation. Investment comments are personal observations, not recommendations to buy or sell.

Information cutoff: October 7, 2026, Korea time. I am not repeating yesterday’s coverage of data-center power shortages, the CFTC’s crypto-trading proposal, or the Bank of Japan’s discussion of AI and interest rates. Google’s new power-supply agreement is real, but it extends the same data-center power story, so I left it out. Today I focus on what “open” means for a model, the quality of an earnings forecast, a rule withdrawal, and a change in how investors view software.

1. Mistral Large 4 preview: “open” does not yet mean you can download it

France-based Mistral AI announced an API preview of Large 4 on October 6. The launch post describes it as a roughly one-trillion-parameter model; its model page gives a more specific total of 1.05 trillion, with 49 billion active at a time. In plain terms, it is a very large organization of specialist parts, but only some of those parts are used to answer each prompt. The complete model weights are not public yet. Mistral says it plans to release them at the end of October, so this is not yet a model anyone can download and install on a private server. Mistral announcement, model documentation

Mistral says it trained the model from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European data centers. That could appeal to organizations that want to host AI under European rules. But being trained in Europe does not by itself guarantee that customer data stays there or that a company can operate the model independently. We need the weights, license, operating costs, and independent evaluations before judging how useful that choice will be. I would also distinguish company-reported benchmarks from results that other evaluators can reproduce.

2. Samsung’s Q3 profit forecast: a record estimate is not a reported result

Samsung is scheduled to release preliminary third-quarter results on October 8. Reuters reported an LSEG consensus of KRW 106.1 trillion in operating profit, based on 21 analysts’ forecasts. That is nearly nine times the KRW 12.17 trillion reported for the same quarter last year—but it is not a result Samsung has announced. The consensus has also been cut 7.7% since late August. When expectations are already this high, what matters is how the actual release compares with them. Reuters report

Demand for high-bandwidth memory (HBM), used in AI accelerators, remains strong, but memory prices are rising more slowly. Reuters cited a forecast for conventional DRAM contract prices to rise 10–15% quarter over quarter in Q4, down from roughly 60% in Q2. In other words, profits can remain high even if the profit margin stops improving. I will be watching HBM shipments, memory-price guidance, and the effect of a stronger won—not just the headline profit figure.

On October 6, the KOSPI fell 0.89%, and foreign investors were net sellers of about KRW 1.77 trillion. Local reporting also cited concern about the end of Samsung and SK hynix share buybacks and the burden of high interest rates. So I would not read one day’s share-price move as a single verdict on the earnings forecast; several pressures were being priced at once. KOSPI market close

3. FinCEN withdraws proposed rules on self-custody wallets and mixers

The U.S. Financial Crimes Enforcement Network (FinCEN) announced on October 5 that it was withdrawing two crypto-related proposals; the withdrawals appeared in the Federal Register on October 6. One proposal would have required banks and money transmitters to verify and record customer information for certain transactions involving self-custody wallets, including records for transactions over $3,000 and reports for those over $10,000. The other would have treated international crypto-mixing transactions as a special money-laundering concern and required additional reporting. Neither proposal had taken effect. FinCEN announcement, self-custody proposal withdrawal, mixing proposal withdrawal

This did not remove an existing $3,000 or $10,000 rule. It removed additional requirements that had been proposed but never implemented. FinCEN also said it would continue watching mixers for illicit finance, so this is not an end to anti-money-laundering or sanctions checks. To me, the important point is that the agency stepped back from a broad proposal that could have swept in legitimate privacy tools; it still needs to decide how to target actual risks more narrowly.

4. U.S. software stocks rebound: AI may replace a product—or improve it

Reuters reported that the S&P 500 software and services index rose 1.3% on October 6, reaching its highest level since November 2025. Investors had worried that AI would quickly replace established software products. More recently, companies have been adding AI to existing tools, and customers are beginning to use those features in real work. That has revived the view that AI can also help software companies sell and improve their products. The sector’s expected 2026 earnings growth has risen from 13.8% at the end of March to 20.6%. Reuters market report

An index rally does not mean every software company is winning. The questions are whether AI features improve renewals and add-on sales, and whether revenue still exceeds the computing cost of those features. Analysts also warn that, as more data-center capacity and coding tools arrive, direct competition from AI could grow again in 2027. I see this rebound not as proof that AI is no threat, but as a test of whether companies with existing products and customers can turn AI into lasting value.

My read of the day

On October 6, the S&P 500 rose 0.6% to a record and the Nasdaq gained 0.45%, while South Korea’s KOSPI fell. Markets can move in different directions even when both are reacting to AI: investors were pricing U.S. software earnings expectations differently from Korean semiconductor flows and rate concerns. U.S. market close

The common question is shifting from “How big and new is the AI?” to “Can it be operated under workable terms, and does it create real profit or customer value?” Mistral’s weights are not public yet, Samsung’s figure is still a forecast, and a software-stock rebound does not settle the long-term competitive picture. I will keep announcements and results, and expectations and evidence, separate as the next data arrives.

Investment comments are a personal record for understanding the news, not a recommendation to buy or sell any asset. Each reader is responsible for their own decisions.