ClickHouse Hits $15B, Gartner Predicts $2.5T AI Spending Surge

So, What Actually Happened?

We scanned 190,000 articles this week so you don't have to, and the AI infrastructure arms race just got a new $15 billion player. ClickHouse hit a $15 billion valuation, positioning itself as a serious Snowflake challenger in the AI data boom. Meanwhile, Gartner forecasts worldwide AI spending will hit $2.5 trillion in 2026—a staggering number that puts every ”AI is a bubble” argument in perspective. And Riot just landed a $1 billion data center deal with AMD, proving that the compute gold rush isn't slowing down.

The Bottom Line: The infrastructure layer is where the real money is moving—and the companies that control data storage and processing will shape what's possible in AI.

The Tracks That Matter

1. ClickHouse Hits $15B Valuation, Eyes Snowflake

ClickHouse Valued at $15 Billion Amidst AI Data Boom, Eyes Snowflake.

ClickHouse just became the most valuable open-source database company in the world. The $15 billion valuation—up from $2 billion just two years ago—reflects something bigger than one company's success: the AI data infrastructure layer is becoming as valuable as the AI models themselves.

The Snowflake angle is significant. ClickHouse is positioning its real-time analytics capabilities as an alternative for AI workloads that need speed over flexibility. While Snowflake excels at complex queries across massive datasets, ClickHouse's column-oriented architecture handles the high-throughput, low-latency requirements that AI inference demands.

For data teams, this creates options. The ”all in on Snowflake” strategy that made sense in 2023 may need revisiting. Not because Snowflake is failing—it isn't—but because AI workloads have different performance profiles that specialized tools handle better.

Here's what works: If your AI workloads are bottlenecked by query latency, evaluate ClickHouse for your real-time analytics layer. The architectural differences from Snowflake aren't bugs—they're features for specific use cases.

2. Gartner: AI Spending to Hit $2.5 Trillion in 2026

Gartner Says Worldwide AI Spending Will Total $2.5 Trillion in 2026.

Gartner's latest forecast puts 2026 worldwide AI spending at $2.5 trillion—a number so large it's hard to contextualize. For reference, that's roughly the GDP of France, or about 10% of US GDP, flowing into AI development, infrastructure, and implementation in a single year.

The breakdown matters more than the headline. Infrastructure (compute, data centers, chips) accounts for the largest share, followed by software and services. Enterprise AI adoption isn't the bottleneck anymore—infrastructure capacity is. Companies have the budget and the use cases; what they're waiting for is the compute and storage to execute.

This validates the infrastructure-first thesis. The companies that seemed overvalued for building data centers, chip fabs, and cloud infrastructure now look correctly priced—or even undervalued—relative to the demand curve.

Here's what works: If you're planning AI initiatives for 2026-2027, secure your infrastructure commitments now. The capacity constraints that defined 2025 aren't resolving—they're intensifying.

3. Runpod Surges to $120M ARR from Reddit Origins

AI Cloud Startup Runpod Surges to $120M ARR, Fueled by Reddit Origins.

Runpod—the GPU cloud company that started as a Reddit community project—just hit $120 million in annual recurring revenue, proving that the AI infrastructure opportunity isn't limited to hyperscalers. The company provides on-demand GPU access to developers who don't want to commit to AWS or Azure long-term contracts.

The Reddit origin story matters. Runpod built its product by listening to what indie AI developers actually needed: flexible pricing, no long-term commitments, and access to consumer-grade GPUs that the enterprise clouds don't offer. That feedback loop produced a service that enterprise customers now also want.

For the cloud market, this signals fragmentation. The assumption that AWS, Azure, and GCP would dominate AI infrastructure is proving incomplete. Specialized providers that focus on specific pain points—like GPU availability and flexible pricing—are capturing meaningful share.

Here's what works: Evaluate specialized GPU providers alongside hyperscaler options. The total cost of ownership often favors specialized platforms for variable workloads.

4. Riot Lands $1B Data Center Deal with AMD

Riot Lands $1 Billion Data Center Deal With AMD—200-Acre Acquisition Expansion.

Riot Platforms just signed a $1 billion data center deal with AMD, including a 200-acre expansion that signals Bitcoin mining companies are pivoting hard toward AI infrastructure. The deal provides AMD with dedicated compute capacity while giving Riot a more diversified revenue stream.

The strategic logic is straightforward. Riot already has power infrastructure, cooling systems, and land—the same resources AI data centers need. Converting Bitcoin mining capacity to AI compute repurposes existing assets for a market with more stable (and growing) demand.

For the AI infrastructure market, this means additional capacity coming online faster than greenfield development would allow. The Bitcoin mining industry built significant infrastructure that can be retrofitted for AI workloads at lower capital cost than new construction.

Here's what works: If you're sourcing data center capacity, consider former mining operators as potential partners. They have infrastructure and power agreements that take years to establish from scratch.

5. AI Model Pairing: 10% Accuracy Boost, 2x Speed

AI Model Pairing Achieves 10% Higher Accuracy and 2x Faster Latency.

Researchers have demonstrated an AI model pairing technique that achieves 10% higher accuracy while cutting latency in half—a combination that usually requires tradeoffs. The approach pairs a small, fast model with a larger, more accurate one, routing queries dynamically based on complexity.

The architecture addresses a real production problem. Most enterprises don't need their most capable model for every query—simple questions can be handled quickly and cheaply, reserving expensive compute for complex cases. Model pairing automates that routing decision.

The 10% accuracy improvement comes from ensemble effects. When both models agree, confidence is high. When they disagree, the system can request human review or additional processing. This is the kind of marginal gain that compounds in production systems serving millions of queries.

Here's what works: If you're running AI in production, evaluate multi-model architectures. The cost savings from routing simple queries to smaller models often justify the added complexity.

6. AI and Nuclear: Small Modular Reactors Find Their Market

Will AI kickstart a new age of nuclear power?.

Small modular reactors (SMRs) have found their killer app: AI data centers. The power requirements of AI infrastructure are so massive and consistent that they justify the capital costs of dedicated nuclear generation. Microsoft, Google, and Amazon are all exploring nuclear power agreements for their data center buildouts.

The economics are straightforward. AI data centers need reliable baseload power—they can't ramp down when the wind isn't blowing. Nuclear provides consistent output with near-zero carbon emissions, satisfying both reliability and sustainability requirements.

The regulatory path is clearing faster than expected. New SMR designs have streamlined approval processes, and the AI industry's lobbying power is accelerating that timeline further. The first AI-dedicated SMR installations could be operational by 2028-2029.

Here's what works: For long-term data center planning, track nuclear power developments. The energy cost advantage of dedicated generation becomes significant at hyperscale.

7. Australia's OAIC Launches First Privacy Compliance Sweep

OAIC Launches First Privacy Compliance Sweep in 2026.

Australia's Office of the Australian Information Commissioner (OAIC) is launching its first comprehensive privacy compliance sweep, targeting organizations across sectors to assess Privacy Act adherence. This isn't a response to specific incidents—it's proactive enforcement designed to establish baseline compliance expectations.

The sweep approach signals a shift in regulatory strategy. Rather than waiting for breaches and complaints, OAIC is auditing organizations before problems occur. This creates pressure to have compliance documentation ready—not just policies, but evidence of implementation.

For organizations operating in Australia, the timing is significant. The sweep coincides with proposed Privacy Act reforms that would strengthen individual rights and increase penalties. Being found non-compliant now may affect how future enforcement actions are treated.

Here's what works: If you process Australian data, conduct a self-audit against OAIC guidelines before they audit you. Proactive compliance documentation is your best defense.

Signal vs. Noise

🟢 Signal: OpenAI's momentum is real and measurable. Our knowledge graph shows OpenAI with +79% PageRank growth this week—not just mentioned more, but increasingly central to how the AI conversation is structured. Anthropic follows at +18%. The gap between leaders and followers is widening, not narrowing.

🔴 Noise: ”AI spending projections” without implementation context. The $2.5 trillion Gartner number is real, but many articles citing it ignore that most of that spending is infrastructure, not AI applications. A company buying servers isn't the same as a company deploying AI successfully.

From the 190K

We scanned 190,000 articles this week. Here's what no one's talking about:

The Agentic Advertising Inflection

The IAB Tech Lab just released an agentic roadmap for digital advertising—and it's more significant than the muted coverage suggests.

The thesis: AI agents will increasingly make purchasing decisions on behalf of consumers. When your customer is an algorithm, traditional advertising doesn't work. The IAB is proposing technical standards for how advertisers can communicate with AI agents, essentially creating a protocol for machine-to-machine commerce.

This isn't science fiction. Amazon Alexa already makes reordering decisions. AI shopping assistants are becoming mainstream. The question isn't whether agents will influence purchases—it's how advertisers will reach them.

The implications for digital advertising are profound. Creative designed for human attention may be irrelevant when agents are making decisions based on structured data and API queries. The advertising industry built for impression-based models faces an existential shift.

The implication: If your business depends on advertising to consumers, start thinking about how AI agents evaluate your offerings. The rules are being written now.

By The Numbers

• $15B — ClickHouse's valuation, up from $2B two years ago
• $2.5T — Gartner's 2026 worldwide AI spending forecast
• $1B — Riot's data center deal with AMD
• +79% — OpenAI's PageRank growth this week
• 55 — GDPR article mentions this week, up from 35 last week
• $120M — Runpod's ARR, up from near-zero in 2024

Deep Dive: When Infrastructure Becomes the Product

Like a DJ who realizes the PA system matters more than the playlist, the AI industry is discovering that infrastructure determines what's possible. The $15 billion ClickHouse valuation, the $2.5 trillion spending forecast, the nuclear power pivot—they all point to the same truth.

The Infrastructure Constraint

AI's limiting factor isn't algorithms anymore. The transformer architecture is seven years old. The major model improvements of the last two years have come from scale—more data, more compute, more training time—not fundamental breakthroughs. The companies that can provide scale become the kingmakers.

This explains the ClickHouse valuation. A database optimized for AI workloads isn't just another analytics tool—it's infrastructure that enables what other companies want to build. Control the infrastructure, and you shape the ecosystem.

The Power Problem

The nuclear power angle isn't a tech curiosity—it's existential. AI training runs consume megawatts of continuous power. The efficiency gains from new architectures are more than offset by scale increases. The AI industry's power consumption is growing exponentially, and the grid isn't keeping up.

Small modular reactors offer a path forward, but they require 5-10 year planning horizons. The companies securing nuclear partnerships now will have capacity in 2030; the ones who wait will be competing for scraps.

What Actually Works

1. Treat infrastructure as strategic: Your data platform choice affects what AI you can run. Evaluate infrastructure decisions on capability, not just cost.

2. Plan for power constraints: If you're building or expanding data center capacity, model the power availability five years out. The projects that seemed overbuilt in 2024 look prescient now.

3. Consider specialized providers: The hyperscaler-default assumption is breaking down. Specialized infrastructure providers often deliver better economics for specific workloads.

4. Track the regulatory overlay: Infrastructure investments require long payback periods. Power, data center, and AI regulations can dramatically affect those economics.

The infrastructure buildout of 2026-2028 will determine the AI capabilities of 2030-2035. The companies that understand this are investing accordingly.

What's Coming

Enterprise GenAI: Lessons from 200+ Deployments

Hard Lessons from 200+ Enterprise Generative AI Deployments. Caylent's analysis of what actually works (and doesn't) in production AI. The failure patterns are more instructive than the successes.

SpacemiT Raises 600M Yuan for RISC-V Chips

SpacemiT Raises Over 600 Million Yuan in Series B. The RISC-V architecture is finding funding as chip supply chain diversification accelerates. Watch for RISC-V in edge AI applications.

M&A Renaissance Fueled by AI and PE

AI and Private Equity Confidence Fueling 2026's M&A Renaissance. Consolidation is accelerating as PE firms chase AI capabilities and AI companies seek scale.

For Your Team

Monday's meeting prompt: ”Gartner says worldwide AI spending will hit $2.5 trillion in 2026. What's our share of that spend, and are we investing in the right layers—infrastructure, platforms, or applications?”

The Infrastructure-First Framework:

1. Audit your data layer — Is your data infrastructure optimized for AI workloads, or just adapted from analytics?
2. Model your compute trajectory — Project your AI compute needs 2-3 years out. Are you on track to have capacity?
3. Evaluate power constraints — For major facilities, understand the power availability timeline
4. Consider specialized providers — The hyperscaler default may not be optimal for your specific workloads

Share-worthy stat: ”ClickHouse just hit a $15B valuation—up from $2B two years ago. When the data infrastructure layer becomes as valuable as the AI models, you know where the industry's real bottleneck is.”

Go deeper: Explore AI infrastructure trends in real-time →

The Track of the Day

”The infrastructure buildout of 2026-2028 will determine the AI capabilities of 2030-2035.”

Like a producer who invests in studio equipment before making music, the AI industry is learning that capability starts with infrastructure. The companies that control data, compute, and power will shape what everyone else can build. ClickHouse at $15 billion, nuclear power partnerships, specialized GPU clouds—the real AI race is for infrastructure, not algorithms.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: January 18, 2026 | Curated by Yves Mulkers @ Ins7ghts