AI funding and the price of trust

The funding discussions, earnings reports and Binance attack were different stories. But each exposed something that enthusiasm can make us overlook. A promising future does not pay today’s bills, and a sophisticated system is not necessarily a safe place to keep money. For our region, those distinctions matter before we decide which parts of this technological change to build around.

Samir Hajibayli, Baku city, VC, LinkedIn

I have spent enough time waiting for the AI bubble to burst. This week, I found a more useful question to follow: who can afford to keep going, and who is being asked to prove it?

Growth is no longer the whole story

The week opened with a warning Elon Musk had posted on Sunday, October 25. He said debt markets had largely closed to AI infrastructure companies without substantial revenue from other businesses. In the same announcement, he said SpaceX had secured $40 billion in GPU financing, helped by its Starlink and launch revenues.

That is Musk’s account, not proof that every bank has stopped financing AI. In fact, the financing he described points to a more selective market, rather than one that has simply run out of money. Lenders want to know what will repay them if the optimistic forecast does not arrive on time.

Alphabet’s results brought that question into public markets. Its third-quarter business generated about $42.3 billion in operating cash, but spending on property and equipment reached $51.4 billion. That left negative free cash flow of roughly $9.1 billion. Revenue growth alone could not remove the concern about how much further spending would be needed.

This does not mean Google is running out of cash. Alphabet reported around $265 billion in cash and short-term marketable securities. The question is whether the growing investment bill will produce an adequate return. Having enough money to spend and spending it well are different things.

On October 29, Alphabet shares fell more than 12%, removing roughly $500 billion from its market value. That headline needs careful reading. Half a trillion dollars did not leave the company’s bank account, nor does the decline measure cash withdrawn from the stock market. A lower share price revalues all outstanding shares, including those that did not trade.

The consequences still matter. Lower valuations can make new fundraising more dilutive and weaken confidence. But calling every repricing a withdrawal of liquidity hides the more interesting development: investors may still believe in AI while becoming less willing to finance it on any terms.

Reading beyond the Anthropic headline loss

Anthropic’s October 26 IPO registration filing shows why those distinctions matter.

Compute and infrastructure spending increased, but revenue grew faster. For each dollar of revenue, those expenses fell from $2.01 in the first half of 2025 to 55 cents in the first half of 2026. That improvement does not include every cost of running the business.

The $80.7 billion first-half net loss also needs context. Most was an accounting charge because some investors’ claims became more valuable as the company’s valuation rose. That did not mean the same amount of cash left the company. Anthropic reported an operating profit in the second quarter, although the full first half remained loss-making at the operating level.

I would read neither failure nor settled economics into these numbers. Anthropic warns that capacity commitments can pressure margins. Will paying demand keep catching up with capacity purchased ahead of it? An IPO filing exposes that question; it does not answer it.

Meanwhile, technology keeps getting cheaper to use. OpenAI’s GPT-6.1 Terra launch on October 26 focused on professional work, visual understanding and computer use, with standard API prices half those of its predecessor. The interesting part for me is what becomes commercially possible when completing a useful task costs less.

That can benefit a small team in Baku even while investors question the companies supplying its models. The technology becoming more useful and its producers facing financial pressure can both be true. Our opportunity may be to turn those cheaper capabilities into a service customers pay for, without taking on the cost of building the underlying model.

Security is part of the product

Binance’s roughly $6.3 billion theft was a different kind of loss altogether. These were assets transferred to attackers, not a stock-market repricing. The FBI attributed the October 29 attack to North Korea’s TraderTraitor activity, often associated with the broader Lazarus ecosystem.

Binance’s initial explanation pointed to a compromised engineer’s laptop, access to a service container, and unauthorized withdrawals. Think of the container as a compartment running part of the exchange’s software. Later updates described additional breach points. Binance said it had found no evidence of leaked wallet private keys; the full attack sequence remained under investigation.

The distinction matters. An attacker does not necessarily need to break a blockchain’s cryptography if they can compromise the systems that authorize transfers. The network can accept a technically valid transaction that the rightful owner never intended. A full post-mortem is needed to establish exactly how the attackers crossed those controls.

By November 1, Binance said it had procured approximately 95% of the affected assets and would cover the losses. That meant replenishing assets, not recovering 95% from the thieves. Customer protection depended partly on the company’s financial resources after its security failed. Bitcoin also fell following the news, although the attack should not be treated as the only influence on prices that week.

So is a hot or cold wallet safe? Keeping keys offline reduces exposure, but it does not make every approval safe. Security researchers have documented cold-wallet losses involving misleading transaction approvals. Custody, signing procedures and recovery arrangements matter alongside the device itself.

AI adds another concern. The UK’s cybersecurity authority expects it to make cyber intrusions more efficient and frequent. That is not evidence that AI caused this Binance breach. It is a reason to ask harder questions as we give software more access to money and company systems. A cheap automated task stops being cheap if one mistake can authorize an irreversible payment.

A regional market needs shared rules

This is why I connect the week’s global stories with the Organization of Turkic States summit in Ankara on October 30. Its agenda included trade, investment, transport, energy and digital cooperation.

The case for cooperation is practical: competing separately makes some ambitions unnecessarily expensive. Different markets can contribute customers, industrial demand, energy, capital and talent. Those advantages become more useful when businesses can actually combine them.

Imagine an Azerbaijani software company selling to a Turkish manufacturer and expanding to Kazakhstan. Regional cooperation should make its second and third markets easier to enter: clearer payment arrangements, more compatible compliance requirements and contracts that customers and investors can trust. A summit does not create a seamless market by announcing one.

The Binance story also suggests a practical area for cooperation: sharing threat information and improving security standards for financial infrastructure. We should judge progress by whether a company can trade more easily and operate more safely, not only by how many agreements are signed.

This is different from asking each country to build its own version of every global technology business. A larger accessible customer base could make local companies more investable. Shared infrastructure could reduce duplication, provided its costs, responsibilities and risks are clear. Cooperation should also remain open to neighbours beyond the organization.

My conclusion this week is not that the AI story is ending. It is that financing and trust deserve as much attention as capability. For the Caucasus and Central Eurasia, the useful response is to help businesses earn revenue across borders and protect the money and data entrusted to them. That would leave us better positioned whether the next year brings another boom or a painful correction.