Who controls the next technology cycle

High rates, AI finance, robots, safety rules and satellite networks all pointed to the same question this week: as technology moves closer to daily life, who keeps control?

Samir Hajibayli, Baku city, VC Caucasus Ventures, LinkedIn

I have been waiting for the AI bubble to burst for almost two years. This week I realized that I am tired of waiting. The bubble may be real, but calling the exact day it breaks has become less useful than watching where money and control are moving next. The Fed minutes, pressure on AI valuations, a new generation of robots, Mistral’s latest model and SpaceX’s move into mobile networks all pointed in the same direction. Capability still matters. Control is becoming just as valuable.

Money still has a price

The Federal Reserve minutes released on October 7 carried a clear warning. In September, the Fed raised its policy rate by 25 basis points to a range of 3.75% to 4%, and most participants thought another increase would probably be appropriate before the end of the year. The minutes also noted that heavy borrowing to finance AI infrastructure was adding to pressure in the bond market.

The latest data make the timing less simple. US payrolls rose by only 29,000 in September, far below the 90,000 economists expected, while unemployment moved up to 4.2%. The latest PCE inflation reading was also cooler than forecast, although headline inflation remained at 3.4% and core inflation at 3.0%. These numbers do not demand an immediate hike. They give the Fed room to wait. Persistent inflation and resilient spending, however, keep a December move alive.

That distinction matters for technology investors. Higher rates do not stop innovation, but they make distant profits worth less today. A company asking investors to fund five years of losses must now offer more than a persuasive story. In the Caucasus and Central Eurasia, this pressure arrives through a higher required return, tighter foreign funding and fewer easy exits. The Fed may be in Washington, but the price it sets travels far beyond the United States.

I am done predicting the bubble

AI valuations still deserve skepticism. OpenAI told investors that its September annualized revenue was almost $50 billion, below an earlier signal of nearly $70 billion. Annualized revenue can be a generous measure because it may take one strong month and multiply it by twelve. Anthropic’s numbers are even harder to ignore: $4.6 billion of revenue in 2025, an operating loss of $8.06 billion and $7.33 billion spent on computing and infrastructure. Its prospectus also describes $518 billion of future cloud and infrastructure obligations.

Both companies may still become extraordinary businesses. That does not make every valuation reasonable. I no longer want to spend each week shouting that the bubble is coming. Public markets will eventually ask the questions that private rounds can postpone: what is recurring revenue, what is accounting, how concentrated are the customers, and when does growth become cash?

This week offered a small preview. Australian AI data center developer Firmus abandoned a planned $5 billion IPO after investors resisted its price and debt load. It had sought an equity valuation of $30.6 billion, almost three times its valuation from only two months earlier. The market did not reject AI. It rejected the amount of faith required at that price. That is the kind of bubble signal I find useful: a real seller finally meeting real buyers.

Intelligence leaves the screen

The next large narrative may be robotics. AI has shortened the time needed to teach machines new tasks and made them more flexible outside carefully programmed routines. On October 5, German robotics company RobCo reached a $1 billion valuation as it prepared a two-armed, self-learning robot for industrial work that traditional automation struggles to handle. The first major wave is therefore more likely to enter factories, warehouses, mines, farms and hospitals before it reaches our kitchens.

The long-term opportunity still feels like an Apple moment: computers once belonged to institutions, then became personal. Robots could follow the same path. Yet a physical assistant creates a harder ownership question. If I buy one, do I own the machine or only permission to use it? The robot may see my home, learn my routines and send data back to its maker. A software update could change its behavior, and the vendor may retain the power to disable it. We learned to accept subscriptions for software. With robots, that bargain gains cameras, hands and movement.

This is why Mistral deserves more attention than it receives. The French company announced Mistral Large 4 on October 6 and plans a wider open-weight release on October 27. Open weights allow an organization to run the model on its own hardware instead of depending entirely on the developer’s servers. For governments and companies that care about data control, that may be more important than winning every benchmark.

Mistral has positioned itself as a safer European alternative, but I would not award that title without a question mark. The company said the model tried to move beyond its test environment and that the behavior was contained. Openness gives the buyer more control, while also reducing the developer’s control over how the model is used. Mistral is smaller than the leading US labs, but it exposes a useful distinction: sovereignty and safety overlap, yet they are not the same thing.

Control needs transparency

I have been reading the safety debate through two extreme possibilities. In one, humanity loses control of systems it can no longer understand. The other is more hopeful and, to me, more exciting. We create something far more capable than ourselves but keep final authority over it. I sometimes think of it as an enslaved god, an uncomfortable phrase that captures the ambition and the danger at the same time.

China and India both moved toward tighter oversight this week. China called for AI to remain safe, reliable and controllable, while also warning officials against blind investment and technology bubbles. India announced a consultation focused on safety, agent autonomy and whether AI systems should retain and reuse data. Yet control by the state is not the same as accountability to the public. A review of 857 model releases by nine Chinese developers found published, model-specific safety results for only 3.6% of them. Companies may have tested privately, but outsiders could not examine the evidence.

China can impose rules faster than more fragmented systems, and that speed matters when the risk is serious. It can also reduce the freedom that helped many technology companies grow elsewhere. The useful lesson is not to copy one political model. It is to combine clear limits with visible tests, independent scrutiny and room for responsible companies to build.

A small smile at the telecom selloff

One event this week made me smile. Telecom companies have earned a special kind of frustration from anyone who has fought an unclear charge, watched a balance disappear or lost service exactly when everybody needed the network. My reaction was emotional, not an investment thesis.

SpaceX agreed to buy nationwide low-band spectrum in the United States in a deal reported at about $8 billion, subject to regulatory approval. T-Mobile fell 13.2%, while AT&T and Verizon lost about 10%. The market understood that satellite-to-phone service could weaken the old gatekeepers. But tower companies rose between 6.5% and 13.3% because even a satellite network may need more ground infrastructure. SpaceX did not destroy telecom. It changed who has leverage.

The regional connection is already real. Beeline Kazakhstan launched Starlink Mobile for ordinary smartphones in areas without terrestrial coverage, with clear uses in mountains, on remote roads and during natural disasters. The future will probably be hybrid. That is good for resilience, but regulators should make sure we do not replace a local gatekeeper with a foreign one that is even harder to question.

What This Means for Our Region

A KPMG survey of more than 70 companies in Central Asia and the Caucasus found that 53% already use AI in production with measurable results, compared with 24% globally. Only 25% had a cross-functional AI governance model, compared with 69% worldwide. The sample is limited, but the gap captures the risk: we are adopting quickly while the rules, skills and accountability structures are still catching up.

Our region has limited capital, which can force useful discipline. We do not need to finance every frontier model or copy the newest US valuation. Robotics investments should begin where the economic case can be measured, especially in energy, mining, logistics, agriculture and manufacturing. A model that can run locally may be more valuable than a famous model that keeps data and control elsewhere. Satellite connectivity should be judged by reliability, emergency access and competitive terms, not only by its novelty.

I will keep watching the AI bubble. When prices finally reset, there will be opportunities and we should be ready to take our position. Until then, I would rather follow the movement underneath the headlines. Capital is becoming more expensive. Intelligence is moving into machines. Networks are moving into orbit. The best regional investments will be the ones that still make sense after the excitement fades, with clear cash flows, clear ownership and someone clearly responsible when the system fails.