2025 was the year AI got the vibe check

Money was not against the AI ​​industry at the beginning of 2025. The vibe check went through in the second half of the year.

OpenAI earned $40 billion at a cost of $300 billion. Safe Superintelligence and Thinking Machine Labs raised $2 billion in seed before shipping a single product. Even first-time startups were growing at a rate once reserved for Big Tech.

Such astronomical sums were followed by equally incredible sums. Trim they spent about $15 billion to shut down Scale AI CEO Alexandr Wang and spent millions to poach talent from other AI labs. So far, the big AI players have pledged about $1.3 trillion in future use.

The first half of 2025 matched the interest, and investor interest, of the previous year. That attitude has changed in recent months to provide a check of sorts. The high expectations of AI, and the absurd calculation, have not changed. But these ideals are now tempered by concerns about the explosion of AI, user safety, and the sustainability of technological progress at its current pace.

The era of unabashed acceptance and celebration of AI is fading at the edges. And with that, more research and questions. Can the AI ​​industry hold its own? Does upgrading in the post-DeepSeek era cost billions? Is there a business model that brings in billions of dollars in revenue?

We’ve been there every step of the way. And our popular stories of 2025 reflect reality: companies that reflect reality even as they promise to reinvent reality.

What a start to the year

Major AI labs have grown this year.

In 2025 alone, OpenAI raised a Softbank led the $40 billion round at a value of $300 billion after investment. The company also says it did money like Amazon round and round agreements, and are in negotiations to raise $100 billion out of $830 billion to calculate. This could bring OpenAI closer to the $1 trillion valuation it is reportedly seeking in an IPO next year.

OpenAI competitor Anthropic also closed $16.5 billion this year on two lines, its own The latest upgrade brought its value to $183 billion with heavy hitters like Iconiq Capital, Fidelity, and Qatar Investment Authority participating. (CEO Dario Amodei admitted to the staff in a memo leaks that he was “not interested” in taking money from the hostile Gulf countries).

Then there’s Elon Musk’s xAI, which he raised at least $10 billion this year after buy Xthe social network formerly known as Twitter that Musk also owns.

We’ve also seen small, new startups get a lot of pushback from froth-mouthed perfumers.

Former OpenAI chief engineer Mira Murati founder of Thinking Machine Labs discovered a $ 2 billion seed round at a cost of $ 12 billion even sharing almost no information about its products. Basic vibe-coding Lovable’s $200 million Series A found a unicorn horn just eight months after launching it; this month, Lovable also raised $330 million about $7 billion after accounting. And we can’t leave out AI recruitment Mercor, which raised $450 million this year in two rounds, the latest one bringing its valuation. $10 billion.

These absurdly large numbers are still happening even in the face of the growing number of adoption-based businesses and major infrastructure challenges, fueling fears of an AI bubble.

Hang on, baby, hang on

For large companies, these figures do not come from anywhere. Validating this calculation requires a lot of construction.

The result has created a vicious cycle. Funds raised for financing are closely linked to how those same funds are returned to chips, cloud contracts, and power, as seen in the funding linked to the OpenAI infrastructure. and Nvidia. In fact, it is blurring the line between money and customer needs, leading to fears that the AI ​​boom is fueled by the circular economy rather than sustainable use.

Some big ones agreements this year are encouraging the growth of construction were:

• Stargatea partnership between Softbank, OpenAI, and Oracle, which includes up to $500 billion to build AI infrastructure in the US.
• Alphabet is gaining strength and data center infrastructure Cut it for $ 4.75 billion, which comes as the company said in October that it plans to raise its revenue in 2026 to $ 93 billion.
• Shave and data center expansionwhich has resulted in the loss of his income to $72 billion in 2025 as the company scrambles to find enough money to train and run the next generation models.

But the cracks are starting to show. Its private equity partner, Blue Owl Capital, soon release of Oracle’s $10 billion data-center investment associated with OpenAI, which shows how some of the big things could be.

Whether all that money ends up is another question. Grid constraints, rising infrastructure and electricity costs, and backlash from residents and policy makers – including calls from figures such as Sen. Bernie Sanders restoring the expansion of the data center – it is already there delaying work in some areas.

While the investment in AI is still huge, the reality of the infrastructure is starting to temper the hype.

Hope is renewed

In 2023 and 2024, every major release feels like a revelation, with new possibilities and new reasons for hype. This year, the magic disappeared, and nothing caught the change better OpenAI release of GPT-5.

Although it made sense on paper, it didn’t land with the same punch like earlier releases such as GPT-4 and 4o. Similar trends emerged across the industry as the transition from LLM providers was less frequent and more or less regional.

Although Gemini 3which tops several benchmarks, it only succeeded in bringing Google into parity with OpenAI – which sparked Sam Altman’s infamous ‘code red’ memo and OpenAI’s fight for continued dominance.

There was also a revision this year in terms of where we expect the border models to come from. The implementation of DeepSeek for R1, his “hypothetical” model. which competes with OpenAI’s o1 on benchmarks, it proved that new labs can deliver reliable models quickly and at low cost.

From modeling to business

As the size of each jump between new brands narrows, investors focus less on raw product information and more on packaged information. The question is: who can turn AI into something that people rely on, pay for, and integrate into their daily lives?

This change is reflected in a number of ways as companies see what works, and what customers let fly. AI Perplexity research, for example, briefly floated the idea of ​​tracking users’ movements on the Internet sell hyper-personalized products. Meanwhile, OpenAI says it’s considering charging up to $20,000 per month for exclusive AIa sign of how hard companies test the water that customers are willing to pay for.

More than anything else, the war has become divided. Paradox is trying to be relevant in its introduction Comet Browser and the power of agents and paying Snap $400 million to explore the search within Snapchat, and buy its way in the available channels.

OpenAI is following a similar path, expanding ChatGPT beyond chatbots and becoming a platform. OpenAI has launched its own Atlas browser and other consumer oriented ones like HittingI’m dating again businesses and producers in start programs within ChatGPT only.

Google, for its part, is leaning responsibility. On the consumer side, Gemini is directly involved in things like Google Calendarwhile on the business side, the company is hosting MCP connectors making its creation difficult to remove.

In a market where it is difficult to differentiate by leaving a new model, having a customer and a business model is a real strategy.

The trust and safety vibe check

AI companies received countless investigations in 2025. More than 50 copyright cases went through the courts, while reports of “AI psychosis” – the effects of chatbots reinforcing fraud and which are thought to be contributing to multiple suicides and other life-threatening events – led to calls for trust and security reform.

While some copyright wars ended – like Anthropic settlement of $ 1.5 billion for authors – many are still unsolved. Although the discussion seems to be shifting from refusing to use copyrighted material in teaching, to demands for compensation (See: The New York Times is suing Perplexity for copyright infringement).

Meanwhile, the psychological concerns about AI chatbots are their own sycophantic answers – emerged as a major public health problem several deaths due to suicide and life threatening scams in young people and adults after using the chatbot for a long time. The result has been lawsuits, much concern among psychiatrists, and quick responses like California’s SB 243 to regulate AI companion bots.

Perhaps even more telling: calls for bans are not coming from tech skeptics.

Industry leaders warn against chatbots”the opposite of juicing,” and even Sam Altman has warned against over-reliance on ChatGPT.

Even the labs themselves started sounding the alarm. May’s Anthropic Defense Report listed Claude Opus 4 as an experiment blackmail experts to prevent its closing itself. Small words? Expanding without understanding what you have built is not a viable option.

Looking ahead

If 2025 was the year that AI began to grow and face difficult questions, 2026 will be the year that it must answer. The cycle of power has begun to slow down, and now AI companies will be forced to prove their businesses and demonstrate real economic value.

The era of ‘trust us, returns will come’ is nearing its end. What comes next may be the confirmation or valuation that makes the dot-com bust look like a bad day for Nvidia’s stock. Time to place your bets.