Technology AI funding and infrastructure 2025

2025’s AI Vibe Check: Mega-Funding, Infrastructure Frenzy, and a Growing Push for Real Business Models

AI’s 2025 “vibe check” brought mega-rounds, trillion-dollar infrastructure promises, and tougher questions about safety, adoption, and returns.

A
·
Ad
Conceptual image of AI's 2025: huge investments, infrastructure growth, and the shift towards sustainable business models.
2025’s AI landscape features massive funding, infrastructure growth, and a critical push for viable business models.
Table of contents

In early 2025, the AI sector looked like it had entered a new financial universe—one where valuations floated higher, funding rounds grew more audacious, and the cost of building frontier models was treated as a secondary concern. By the second half of the year, that momentum didn’t disappear, but it did start to meet resistance: skepticism about bubble dynamics, new worries about safety and governance, and harder questions about whether the industry’s pace is sustainable.

That shift—more of a “vibe check” than a collapse—has been shaped by blockbuster fundraising, massive infrastructure commitments, and an industrywide transition from “look what the model can do” to “who is going to pay for this, and why?”

How 2025 started: bigger labs, bigger checks

In the first half of 2025, money flowed into AI at a scale that previously felt reserved for the biggest technology incumbents. OpenAI raised $40 billion in a Softbank-led round at a $300 billion post-money valuation. The company also reportedly had investors like Amazon circling with compute-tied, circular arrangements and was in talks to raise $100 billion at an $830 billion valuation—bringing it closer to the roughly $1 trillion valuation it is reportedly targeting in an IPO next year.

Rival lab Anthropic also stacked up an enormous funding tally: $16.5 billion across two rounds in 2025, with its most recent raise valuing the company at $183 billion. Participants included Iconiq Capital, Fidelity, and the Qatar Investment Authority. CEO Dario Amodei, according to a leaked memo reported by Wired, told staff he was “not thrilled” about taking money from dictatorial Gulf states.

Elon Musk’s xAI raised at least $10 billion this year after acquiring X, the social media platform formerly known as Twitter that Musk also owns. Collectively, the biggest labs didn’t just continue to grow; they expanded their balance sheets and ambition at a pace that reinforced the idea that the frontier would be built by the best-capitalized players.

Startups rode the wave, too

The year’s exuberance wasn’t limited to the established leaders. Several newer startups secured unusually large rounds—often before showing much in the way of shipped product.

  • Thinking Machine Labs, founded by former OpenAI chief technologist Mira Murati, landed a $2 billion seed round at a $12 billion valuation, despite releasing almost no detail about what it planned to build.
  • Lovable, described as a “vibe-coding” startup, raised a $200 million Series A and reached unicorn status just eight months after launch. Later, it raised another $330 million at a nearly $7 billion post-money valuation.
  • Mercor, an AI recruiting startup, raised $450 million across two rounds in 2025, and the latest round pushed its valuation to $10 billion.

These prices persisted even as enterprise adoption remained “still-modest” and infrastructure constraints remained serious—conditions that naturally intensify concern that parts of the market may be priced for perfection.

Build, baby, build: the infrastructure arms race

If valuations are going to stay this high, the logic goes, the industry needs the physical capacity to deliver: chips, data centers, power, and cloud contracts. In 2025, infrastructure became the other half of the AI story, not merely a supporting character. The spending plans grew so large they began to raise a different kind of question: are we seeing genuine end-user demand, or a funding loop where capital recirculates through the same ecosystem?

TechCrunch’s reporting describes an increasingly common pattern: capital raised for “compute” is often intertwined with arrangements where money flows back into chips, cloud, and energy. One example cited was OpenAI’s infrastructure-linked funding tied to Nvidia, noted in coverage of a deal discussed by Yahoo Finance. The practical concern is that this blurs the boundary between an investor betting on demand and a customer paying for usage, creating a form of circular economics that can make the boom look sturdier than it is.

Three deals that defined the year’s infrastructure push

Several headline commitments illustrated how quickly the infrastructure race escalated:

  • Stargate, a joint venture between Softbank, OpenAI, and Oracle, includes up to $500 billion aimed at building AI infrastructure in the U.S.
  • Alphabet acquired Intersect, an energy and data center infrastructure provider, for $4.75 billion. The deal arrived as the company said in October it planned to lift compute spending in 2026 up to $93 billion.
  • Meta’s accelerated data center expansion pushed its projected capital expenditures to $72 billion in 2025 as it sought enough compute to train and run next-generation models.

At the same time, the year also surfaced signs of fragility. A private financing partner, Blue Owl Capital, pulled out of a planned $10 billion Oracle data-center deal tied to OpenAI capacity, highlighting how quickly ambitious financing structures can wobble.

Constraints: power, cost, and politics

Even if companies want to spend, building at this scale collides with real-world limits. TechCrunch’s report points to grid constraints, rising construction and power costs, and a growing wave of local and political pushback. The scrutiny is no longer theoretical: figures like Sen. Bernie Sanders have called for reining in data center expansion, and watchdog research cited in the report suggests projects are already slowing in some regions.

The net effect is a subtle but meaningful tempering of the narrative. Investment remains huge, but the infrastructure reality—how difficult it is to secure power, permits, and community acceptance—has started to put boundaries around the hype.

The expectation reset: when breakthroughs feel incremental

In 2023 and 2024, each major model launch often felt like a step-change: new capabilities, new product categories, and new reasons to believe the ceiling had moved. In 2025, TechCrunch argues, that sense of magic weakened. The clearest example in the report is the reception to OpenAI’s GPT-5 rollout: meaningful on paper, but not a cultural or technical jolt on the level of GPT-4 or 4o.

The same pattern showed up across other large language model (LLM) providers: progress continued, but the improvements were increasingly incremental or narrowly focused on specific domains rather than obviously transformative for general users.

Even Gemini 3—described in the report as topping several benchmarks—read as a breakthrough mainly because it brought Google back to parity with OpenAI, rather than establishing an undisputed new lead.

DeepSeek and the “who can build frontier models?” question

Another reason expectations shifted is that the “frontier” suddenly looked more accessible. DeepSeek’s release of R1, a “reasoning” model that competed with OpenAI’s o1 on key benchmarks, demonstrated that new labs can produce credible systems quickly and at a fraction of the cost. In practical terms, that changes how investors and competitors think about moats: if high-quality models can emerge faster and cheaper, then distribution, product design, and business model matter even more.

From model breakthroughs to business models

As the jumps between model generations feel less dramatic, attention naturally turns to what sits on top of the model—packaging, workflow integration, pricing, and the reasons customers will stick around. The question becomes less “whose model is smartest?” and more “whose AI becomes indispensable?”

In TechCrunch’s framing, the industry is increasingly testing what users and customers will tolerate as companies search for durable revenue. The report cites Perplexity, an AI search startup, as briefly floating the idea of tracking users’ online movements in order to sell hyper-personalized ads. It also notes that OpenAI was reportedly considering charging up to $20,000 per month for specialized AI—an example of how aggressively companies have explored premium pricing for high-value use cases.

That experimentation underscores a central tension of 2025: the world’s most expensive models require vast ongoing spend, yet the most scalable, predictable revenue streams are still being figured out. Consumer subscriptions can help, but enterprise budgets, procurement timelines, and trust requirements move more slowly. Meanwhile, ad-based models can introduce privacy tradeoffs that provoke backlash. The business model debate, in other words, is no longer an academic sidebar—it has become one of the core storylines.

Why the “vibe check” matters

The late-2025 shift described by TechCrunch isn’t simply pessimism. Rather, it reflects the industry moving from a phase dominated by possibility to one constrained by execution. The same forces that powered the boom—abundant capital, competitive pressure to scale, and rapid iteration—also created conditions for tougher scrutiny:

  • Bubble risk as valuations outrun adoption and infrastructure remains tight.
  • Safety and governance concerns as AI systems become more embedded in daily workflows and higher-stakes contexts.
  • Scaling questions in a “post-DeepSeek” environment where competitive models may be built at lower cost.
  • Return-on-investment pressure as investors look for proof that multi-billion-dollar funding can translate into sustainable revenue.

Put differently, AI in 2025 didn’t stop being huge—it started being contested, challenged, and interrogated in more concrete terms.

Conclusion

2025 showed how far AI’s financial and technical ambitions can stretch—and also where reality pushes back. Mega-rounds and infrastructure pledges kept the accelerator down, but the second half of the year introduced a more cautious tone: breakthroughs felt less seismic, the economics of compute drew sharper skepticism, and the market began demanding clearer answers on safety, adoption, and durable business models.

This article is based on reporting originally published by TechCrunch.

<<>>

Based on reporting originally published by TechCrunch. See the sources section below.

Sources

Ad

You might also like