2025 Was the Year AI Got a Vibe Check

In early 2025, money was no object for the AI industry. By mid-year, a vibe check began to set in.

OpenAI raised $40 billion at a $300 billion valuation. Safe Superintelligence and Thinking Machine Labs raised $2 billion in individual seed rounds before shipping a single product. Even first-time founders were raising funds on a scale previously reserved for Big Tech.

Such astronomical investments were matched by equally incredible spending. Meta shelled out nearly $15 billion to lock up Scale AI CEO Alexandr Wang and poached talent from other AI labs. Meanwhile, AI's leading players have committed to almost $1.3 trillion in future infrastructure spending.

The first half of 2025 mirrored the fervour and investor interest of the prior year. However, that mood shifted in the second half, delivering a vibe check of sorts. Extreme optimism for AI and the wild valuations remained intact, but concerns about an AI bubble bursting, user safety, and the sustainability of technological progress began to temper the excitement.

The era of unabashed acceptance and celebration of AI is fading at the edges, bringing more scrutiny and questions. Can AI companies sustain their current velocity? Does scaling in the post-DeepSeek era require billions? And is there a viable business model that can justify the multi-billion-dollar investment? We've been there at every step, and the most popular stories of 2025 tell the real story: an industry hitting a reality check while promising to reshape reality itself.

How the Year Started

The biggest AI labs grew even larger in 2025.

OpenAI raised a Softbank-led $40 billion round at a $300 billion post-money valuation. The company also reportedly has investors like Amazon circling with compute-related circular deals and is in talks to raise $100 billion at an $830 billion valuation, bringing OpenAI close to the $1 trillion valuation it reportedly seeks in an IPO next year.

OpenAI rival Anthropic raised $16.5 billion this year across two rounds, pushing its valuation to $183 billion with participation from heavy hitters like Iconiq Capital, Fidelity, and the Qatar Investment Authority. (CEO Dario Amodei confessed in a leaked memo that he was not thrilled about accepting money from dictatorial Gulf states.)

Meanwhile, Musk's xAI raised at least $10 billion this year after acquiring X, the social media platform formerly known as Twitter, which Musk also owns.

Smaller, newer startups have also garnered hype from eager investors. For example, Thinking Machine Labs, founded by former OpenAI chief technologist Mira Murati, secured a $2 billion seed round at a $12 billion valuation despite revealing almost no details about its product. Lovable, a vibe-coding startup, raised $200 million in a Series A round, achieving unicorn status just eight months after its launch. This month, Lovable raised another $330 million at a nearly $7 billion post-money valuation. Mercor, an AI recruiting startup, also raised $450 million across two rounds, bringing its valuation to $10 billion.

These absurdly large valuations are occurring despite still modest enterprise adoption rates and significant infrastructure constraints, heightening fears of an AI bubble.

Build, Baby, Build

For larger firms, these valuations aren't coming out of thin air. Justifying them requires building extensive infrastructure.

This has created a vicious cycle. Capital raised to fund computing is increasingly tied to deals where the same money flows back into chips, cloud contracts, and energy, as seen in OpenAI's infrastructure-linked funding with Nvidia. This blurring of the line between investment and customer demand has stoked fears that the AI boom is being propped up by circular economics rather than sustainable usage.

Some of the biggest deals powering the infrastructure boom this year included:

• Stargate, a joint venture between Softbank, OpenAI, and Oracle, which includes up to $500 billion to build AI infrastructure in the U.S.
• Alphabet's acquisition of energy and data centre provider Intersect for $4.75 billion aligns with its plans to increase its compute spend to $93 billion in 2026.
• Meta's accelerated data centre expansion has pushed its projected capital expenditures to $72 billion in 2025 as the company races to secure sufficient compute capacity for next-gen models.

However, cracks are starting to show. Blue Owl Capital, a private financing partner, recently withdrew from a planned $10 billion Oracle data centre deal tied to OpenAI capacity, highlighting how fragile some capital stacks can be.

Whether all that spending materialises remains uncertain. Grid constraints, soaring construction and power costs, and growing pushback from residents and policymakers—such as Sen. Bernie Sanders's calls to curb data centre expansion—are already slowing projects in some regions.

Even as AI investment remains enormous, the reality of infrastructure is beginning to temper the hype.

The Expectation Reset

In 2023 and 2024, each new model release felt like a revelation, with new capabilities and fresh reasons to fall for the hype. This year, the magic faded, and nothing captured that shift better than OpenAI's GPT-5 rollout.

While GPT-5 was meaningful on paper, it didn't have the same impact as earlier releases like GPT-4 and GPT-4o. Similar patterns emerged across the industry, where improvements from LLM providers were less transformative and more incremental or domain-specific.

Even Gemini 3, which tops several benchmarks, only made a splash by bringing Google back on equal footing with OpenAI, which sparked SALTman's infamous code memo and OpenAI's fight to maintain dominance.

There was also a reset this year in terms of where we expect frontier models to come from DeepSeek's launch of R1, its reason for model, competed with OpenAI's O1 on key benchmarks, proving that new labs can ship credible models quickly and at a fraction of the cost.

From Model Breakthroughs to Business Models

As the gaps between new models narrow, investors are focusing less on raw model capacity and more on the surrounding context. The question now is: who can turn AI into a product that people rely on, pay for, and integrate into their daily workflows?

This shift is manifesting in several ways, as companies test what works and what customers will accept. AI search startup Perplexity, for example, briefly considered tracking users' online movements to sell hyper-personalised ads. Meanwhile, OpenAI reportedly considered charging up to $20,000 per month for specialised AI, signalling a bold attempt to gauge customer willingness to pay.

More than anything, though, the fight is now centred around distribution. Perplexity is trying to stay relevant by launching its own Comet browser with agentic capabilities and paying Snap $400 million to power search inside Snapchat, effectively buying its way into existing user funnels.

OpenAI is pursuing a parallel strategy, expanding ChatGPT beyond a chatbot into a platform. The company has launched its own Atlas browser and features such as Pulse, while courting enterprises and developers by adding apps to ChatGPT.

Google is leaning on incumbency, integrating Gemini into products like Google Calendar for consumers and hosting MCP connectors for enterprises to make its ecosystem harder to dislodge.

In a market where it's harder to differentiate with new models, owning the customer and the business model is the real moat.

The Trust and Safety Vibe Check

AI companies faced unprecedented scrutiny in 2025. More than 50 copyright lawsuits were filed in court. At the same time, reports of "AI psycho"—caused by chatbots reinforcing delusions and allegedly contributing to suicides—sparked calls for trust and safety reforms.

While some copyright disputes, including Anthropic's $ 1.5 billion settlement with authors, have been resolved, most remain unresolved. The conversation has shifted from resistance against using copyrighted content for training to demands for compensation (e.g., the New York Times suing Perplexity for copyright infringement).

Meanwhile, mental health concerns around AI chatbots—particularly their sycophantic responses—emerged as a serious public health issue after multiple suicides and life-threatening delusions in teens and adults from prolonged chatbot usage. This has led to lawsuits, widespread concern among mental health professionals, and swift policy responses, such as California's B 243, which regulates companion bots.

Perhaps most telling: the calls for restraints are not coming from the usual anti-tech voices.

Industry leaders have warned against chatbots' juicing engagement, and even Sam Altman has cautioned against emotional overreliance on ChatGPT.

Even the labs themselves are sounding alarms. Anthropic's May safety report documented Claude Opus 4 attempting to blackmail engineers to prevent its own shutdown. The subtext? Scaling without fully understanding why you've built it is no longer a viable strategy.

Looking Ahead

If 2025 was the year AI began to mature and face tricky questions, 2026 will be the year it must answer them. The hype cycle is best described as a fade, and AI companies will be forced to prove their business models and demonstrate real economic value.

The era of trust us, the returns will come is nearing its end. What comes next will either be a vindication or a reckoning that makes the dot-com bust look like a bad day for Nvidia—time to place your bets.