AI News Roundup Today (Dec. 18, 2025): OpenAI’s $750B Funding Talks, Google’s Gemini 3 Flash, Amazon’s AI Shakeup, and the Data Center Backlash

Thursday, December 18, 2025 — Today’s AI news cycle had one clear throughline: scale is getting expensive, and the industry is now wrestling with the consequences of building at “national infrastructure” size. OpenAI is reportedly exploring a funding round that would be almost without precedent; Google is pushing faster, cheaper models deeper into Search; Amazon is reorganizing its AI org around chips and frontier research; and the political fight over data centers is escalating from local zoning meetings to Capitol Hill. ^[1]

Below is a detailed roundup of the biggest AI stories shaping headlines on 18.12.2025 across product launches, funding, chips, regulation, and the booming—and increasingly contested—data center economy.

The day in AI, in 10 seconds

OpenAI is reportedly in early discussions to raise up to $100 billion at a valuation around $750 billion, highlighting how compute demand is rewriting the rules of fundraising. ^[2]
Google rolled out Gemini 3 Flash as the default model in the Gemini app and pushed it into AI Mode in Search, framing “speed + reasoning” as the next mainstream battleground. ^[3]
Amazon is consolidating “AGI,” custom silicon, and quantum work under AWS veteran Peter DeSantis as AI chief Rohit Prasad prepares to depart. ^[4]
Oracle’s Michigan mega–data center effort tied to OpenAI’s infrastructure push is facing fresh scrutiny after funding talks reportedly shifted—another sign investors are watching the AI capex boom with sharper skepticism. ^[5]
Washington politics is moving closer to the power meter: U.S. senators are probing whether data centers are driving up household electricity bills, while Sen. Bernie Sanders is calling for a pause on AI data center construction. ^[6]

OpenAI’s reported $100B raise talks show how “compute” became the new currency

The most attention-grabbing number in today’s roundup: OpenAI has reportedly held preliminary talks with investors about raising funding at a valuation of roughly $750 billion, and the round could reach tens of billions—potentially up to $100 billion, according to reporting referenced by Reuters. ^[7]

The reported valuation matters less as a trophy and more as a signal: frontier-model development is now inseparable from industrial-scale infrastructure. Reuters notes the company’s appetite for computing power and points to OpenAI’s major partnerships across the AI stack—both chips and cloud—as the industry races to build systems that can match or exceed human capabilities in broader domains. ^[8]

At the same time, the tone around AI finance is shifting from “growth at all costs” to “growth with receipts.” Reuters also flags investor caution: markets are eager to see whether massive buildouts can translate into durable revenue—not just headlines and benchmark wins. ^[9]

The “circular deal” debate: Amazon + OpenAI chatter keeps bubbling

Separately, Amazon is reportedly in talks about investing $10 billion or more in OpenAI, in a deal that would also involve OpenAI buying some of Amazon’s custom AI chips—an arrangement Investopedia frames as part of a broader pattern of “circular deals” (investments that loop back into supplier/customer relationships). ^[10]

These structures can accelerate scaling—especially when compute supply is the bottleneck—but they also raise a familiar late-cycle question: are valuations being pulled upward by financial engineering as much as by user demand? Investopedia explicitly connects the trend to “AI bubble” anxieties, even as it notes the strategic logic for the companies involved. ^[11]

Google launches Gemini 3 Flash and pushes it straight into Search

Google’s biggest consumer-facing AI move today: Gemini 3 Flash is now the default model in the Gemini app, replacing Gemini 2.5 Flash, and it’s also headed into Google Search—specifically AI Mode, which previously ran on 2.5 Flash. ^[12]

Google is positioning Gemini 3 Flash as a “best of both worlds” model: Gemini 3 Pro–grade reasoning with Flash-level latency, efficiency, and cost, according to statements reported by The Verge. ^[13]

Two details stand out for anyone tracking where mainstream AI is going next:

Efficiency is becoming a product feature. Google says Gemini 3 Flash can outperform its prior flagship (Gemini 2.5 Pro) while operating at a fraction of the cost—a direct shot at the idea that capability must always mean higher inference bills. ^[14]
Distribution is the advantage. By routing the model not only into the Gemini app but into Search AI Mode, Google is effectively turning model upgrades into a default experience—no new app required, no prompt-engineering hobbyist badge needed. ^[15]

Google is also bringing Gemini 3 Flash to developers through a wide set of channels including Google AI Studio, the Gemini API, Gemini CLI, Android Studio, and Vertex AI, reinforcing that this is both a consumer and enterprise push. ^[16]

Amazon reshuffles its AI org: one leader over models, chips, and quantum

Amazon’s AI story today is partly leadership, partly strategy.

Reuters reports that Amazon is restructuring its AI efforts as Rohit Prasad—a key leader connected to Alexa and recent AI initiatives—plans to leave at year’s end. Peter DeSantis, a long-time AWS executive, will lead a new unit combining Amazon’s most advanced AI work: models, custom silicon (including Graviton and Trainium), and quantum computing initiatives. ^[17]

In Amazon CEO Andy Jassy’s framing, the company is at an “inflection point” where these technologies will power future customer experiences—language that underscores how Amazon wants investors and developers to view the reorg: as an attempt to compress timelines and reduce internal fragmentation. ^[18]

The Verge’s reporting adds more color to the competitive context: Amazon is still widely perceived as playing catch-up against Microsoft, Google, Meta, OpenAI, and Anthropic in the foundation-model race, and the new structure is designed to unify focus across models, chips, and compute. ^[19]

A key research detail: Pieter Abbeel will lead Amazon’s frontier model research team while continuing robotics work—another sign Amazon is connecting foundation models to embodied/agentic systems rather than treating them as a standalone chatbot project. ^[20]

Nvidia’s moat faces new pressure—but the fight is really about software

Not all “AI chip competition” is about chip speed. A growing share of the battle is about developer friction—specifically, whether teams can move workloads without rebuilding their entire stack.

Reuters reports Google is working on an internal initiative known as “TorchTPU” to make its TPUs work better with PyTorch, the most widely used AI software framework—and it’s doing so with Meta’s help (Meta is the steward and primary backer of PyTorch). ^[21]

Why this is a big deal:

Nvidia’s dominance has been reinforced by CUDA, deeply embedded in how many teams train and deploy models. If TPUs become meaningfully easier to adopt inside PyTorch-heavy stacks, switching costs drop—and Nvidia’s software lock-in weakens. ^[22]
Reuters notes Google has also been expanding access to TPUs beyond internal use, including selling TPUs into customers’ data centers—not just offering them through Google Cloud. ^[23]

This doesn’t mean Nvidia is suddenly “in trouble.” But it does show the industry’s new playbook: don’t just build chips—build the path of least resistance for developers.

The AI data center boom runs into financing stress, electricity scrutiny, and a rising political backlash

If 2024 was the year AI “needed more GPUs,” 2025 is the year AI “needed more power plants, debt packages, and permitting.”

Oracle’s Michigan mega–data center faces new questions

Reuters reports Oracle said talks remain on track for an equity deal supporting a more than 1-gigawatt data center project in Saline Township, Michigan, and that the talks do not include Blue Owl Capital after reports about stalled negotiations hit the stock. The project is described as part of the Stargate AI infrastructure push involving Oracle and OpenAI, with construction slated to begin in early 2026. ^[24]

The episode matters because it sits at the intersection of AI ambition and balance-sheet reality. Reuters notes investors have been scrutinizing Oracle’s AI build-out as debt climbs and Oracle’s fortunes become increasingly tied to OpenAI’s infrastructure plans. ^[25]

Markets wobble on “AI capex reality checks”

A Reuters markets wrapout highlights how these infrastructure headlines are spilling into broader sentiment: tech jitters and AI infrastructure worries helped pressure shares, with notable moves including declines for major AI-linked stocks. ^[26]

Hut 8’s $7B lease shows the other side: mega-builds are still getting signed

Even with rising scrutiny, large AI infrastructure deals continue. Hut 8 announced a 15-year, 245 MW AI data center lease at its River Bend campus valued at $7.0 billion over the base term (with renewal options that could raise the total), with Google providing a financial backstop for the base lease obligations. ^[27]

Senators probe whether households are subsidizing Big Tech’s AI power bill

In one of the most consequential policy signals today, The Verge reports Sens. Elizabeth Warren, Chris Van Hollen, and Richard Blumenthal sent letters to major tech companies and data center developers asking whether residential customers are effectively subsidizing the electricity costs of AI infrastructure. The Verge also cites U.S. Department of Energy expectations that data centers’ share of national electricity use could rise sharply in the coming years. ^[28]

This is the emerging political pressure point: AI isn’t just an app anymore—it’s a load on the grid.

Bernie Sanders calls for a national pause on AI data center construction

Business Insider reports Sen. Bernie Sanders is pushing for a national moratorium on AI data center construction, arguing the buildout is moving faster than democracy can respond, and raising concerns about power concentration, job displacement, and social impacts. ^[29]

Whether or not such a proposal goes anywhere legislatively, it signals a new phase of the debate: data centers are becoming a national campaign issue, not just a local permitting fight.

“Who’s the tenant?” becomes the next big question

Business Insider also reports that Amazon was the prospective tenant in talks tied to a data center developer’s financing drama earlier this month—another reminder that the most important fact in many data center stories is often the one communities can’t see due to NDAs: who is actually buying the compute? ^[30]

Regulation: states push back against federal pressure on AI rules

Alongside the infrastructure backlash, the U.S. is also entering a sharper federal–state conflict over who gets to regulate AI.

Colorado is moving ahead with its AI regulation despite a Trump executive order calling for a moratorium on state action, Axios reports. The law, signed in 2024, focuses on disclosures and limits for “high risk” AI uses that could lead to discrimination, with implementation delayed to June 30, 2026. ^[31]

In Illinois, local reporting highlights state leaders signaling they “won’t back down” in response to the same federal posture, pointing to state-level AI safeguards and arguing the federal approach is unconstitutional overreach. ^[32]

To understand what’s actually in play, the White House executive order text includes a directive for the Attorney General to establish an AI Litigation Task Force tasked with challenging state AI laws deemed inconsistent with national policy. ^[33]

Taken together, today’s headlines suggest 2026 may become the year AI regulation is shaped as much by courts and federal funding levers as by legislatures.

AI goes shopping: DoorDash brings grocery ordering into ChatGPT

One of the clearest “AI meets everyday life” stories today: DoorDash announced a new DoorDash app inside ChatGPTdesigned to turn recipe ideas into grocery orders delivered from local stores—aiming to collapse browsing, list-making, and checkout into a single conversational flow. ^[34]

MarketWatch frames this as the newest battleground in “shopping inside chatbots,” positioning ChatGPT as a new kind of storefront where discovery and conversion can happen in the same window. ^[35]

Why this matters beyond groceries:

It’s a real-world test of conversational commerce at scale.
It pressures every retailer and delivery platform to answer a new question: when customers ask an AI what to buy, whose inventory and logistics show up first?

Talent and geography: Dublin expansion, and Yann LeCun’s next big bet

OpenAI and Anthropic look to expand office space in Dublin

Bloomberg reports both OpenAI and Anthropic are exploring expanded office footprints in Dublin, a sign of Europe’s growing role in the AI commercialization race. Bloomberg notes Anthropic is looking at leasing around 25,000 square feet over the next three to five years (with plans described as evolving). ^[36]

Yann LeCun’s new startup aims for €3B valuation

On the frontier-research side, the Financial Times reports Yann LeCun is preparing to launch a new AI startup, Advanced Machine Intelligence (AMI) Labs, targeting a €3 billion valuation and seeking €500 million in funding, with a January 2026 launch. The effort is focused on “world models” that better understand the physical environment—technology with obvious applications in robotics and transport. ^[37]

This is a notable countercurrent to pure “bigger language model” thinking: it’s an attempt to build AI that reasons about the world, not just text about the world.

Copyright and training data: Adobe faces a new lawsuit over AI model training

Today’s legal headline adds to the growing list of AI training-data battles. Reuters reports author Elizabeth Lyon filed a proposed class action alleging Adobe misused copyrighted books (including her own) to train its SlimLM small language models without permission. Reuters describes it as the first such case against Adobe in the wave of copyright suits targeting AI training practices. ^[38]

These cases continue to matter for the entire ecosystem because they shape:

how training data is sourced and documented,
what licensing markets emerge,
and whether “clean room” datasets become a competitive advantage.

What to watch next

Three questions coming out of today’s AI news cycle:

Will OpenAI confirm or clarify fundraising plans—and what terms will define the next mega-round? ^[39]
Can Google (and Amazon, and others) meaningfully reduce reliance on Nvidia by lowering software switching costs—not just by building hardware? ^[40]
Does the political center of gravity shift from “AI safety” to “AI infrastructure costs”—power bills, water use, tax abatements, and who pays? ^[41]

The AI era is increasingly being negotiated not only in product demos and research papers, but in utility commissions, financing term sheets, and courtrooms.

Wall Street Closes Lower as AI Funding Jitters Hit Tech Stocks