Four of every five dollars raised this week went to AI infrastructure. Here is what happened in the other 20 percent.

On the morning of July 1, Together AI announced it had closed $800 million in Series C funding led by Aramco Ventures, with NVIDIA, Vista Equity Partners, and General Catalyst also in the round. The next day, Crusoe reported it was in talks to raise $3 billion -- a round that would triple its valuation to roughly $18 billion. That same morning, Switch, the Las Vegas-based data center operator, disclosed it was seeking $2 billion in fresh capital. And sitting atop all three: Kling AI's $2 billion close, backed by General Atlantic at an $18 billion valuation for the Chinese video AI company.

In two days, four companies raised or sought $7.8 billion. Three build the physical layer AI runs on: compute, power, data center fabric, inference cloud. The fourth generates AI video at scale. Application companies, the ones building software products on top of that infrastructure, were not part of this conversation. Their checks averaged $25.8 million at the Series A level and $12.3 million at seed. The infrastructure-to-application funding ratio this week was approximately 4:1 by dollars, the widest it has been in any single week this year.

This is not a temporary distortion. Strip the three mega-rounds and the week shows $2.9 billion across 29 rounds -- a perfectly ordinary summer pace. The $9.9 billion headline is accurate but misleading. The important number is $7.8 billion in 48 hours for infrastructure, and the fact that the capital sources behind those rounds -- Aramco Ventures, NVIDIA, General Atlantic, and unnamed sovereign-adjacent funds -- are not traditional venture. They are growth equity, corporate strategic, and energy-sector capital treating AI compute like a utility asset class.

The numbers

The 958% funding increase is real but driven entirely by concentration. The top three rounds (Crusoe $3B, Switch $2B, Kling AI $2B) account for $7 billion of the week's $9.94 billion. Strip those and the underlying market runs at $2.94 billion across 29 rounds, close to the four-week trailing pace. The spike in new startups is also misleading: the database this week absorbed a large batch of AI-generated startup submissions from what appears to be a single automated submitter, most of them Israeli-labeled and nearly blank, inflating the raw count significantly.

Crusoe, Switch, and Together AI: one thesis, three entry points

Crusoe is in talks to raise $3 billion at a valuation that would reach approximately $18 billion, according to Bloomberg's July 2 report. The company's model is built on energy arbitrage: it identifies stranded power assets -- electricity that cannot be economically transmitted to population centers because it was generated far from demand -- and co-locates AI compute with that power source. The result is data center electricity costs well below what any hyperscaler pays at scale, at the cost of remote geography and high capital intensity for buildout. Investors backing this round are implicitly betting that AI's electricity consumption, which the IEA estimates will triple global data center consumption by 2030, cannot be satisfied by grid infrastructure near existing cities. Crusoe's pitch is that the constraint is power, not silicon.

Switch, the Las Vegas-based hyperscale data center operator (listed in the database as "Switch Mobility" due to a cataloging error -- the Bloomberg article's URL reads "data-center-firm-switch"), is separately seeking $2 billion. Switch operates campuses across the US and internationally and has been a beneficiary of hyperscaler co-location demand. The Crusoe and Switch rounds, both surfacing July 2, together represent $5 billion on a single day for AI's physical container -- power and space.

Together AI's $800 million Series C (SiliconAngle, July 1) is a different layer of the same infrastructure stack: inference cloud. The company runs open-source AI models on dedicated compute, providing a serverless environment that it claims runs at roughly twice the performance of the fastest alternatives. The round was led by Aramco Ventures, with NVIDIA, Vista Equity Partners, and General Catalyst participating. Aramco's presence is notable: the Saudi state oil company's venture arm has been systematically building positions in AI infrastructure companies whose compute appetites depend on stable, large-scale power supply -- which Aramco understands better than most investors. Together AI disclosed that its annual bookings crossed $1.15 billion in Q2 2026. It plans to use the new capital to grow its public cloud capacity by a factor of 50 over five years.

The round follows Groq's $650 million raise the prior week. Groq and Together AI now occupy the same market position with similar-sized balance sheets: both pitch dedicated AI inference at lower cost than hyperscalers, both target the same enterprise developers. One of them will consolidate the other's customers. The announcement timeline -- Groq on June 22, Together AI on July 1 -- suggests both companies were running parallel processes that closed within nine days of each other, which will shorten the window before the competitive dynamics become visible in sales data.

Kling AI at $18 billion: what Chinese video AI capital looks like at scale

Kling AI closed $2 billion in a round led by General Atlantic, valuing the Kuaishou-backed video AI company at $18 billion (Bloomberg, July 2). The company had previously disclosed a Series A process in June seeking the same $2 billion at the same valuation -- the July 2 announcement represents the formal close of that round rather than an additional raise.

Kling generates video from text and image prompts and has achieved meaningful international traction despite the US-China AI technology environment that has constrained some Chinese AI companies' global ambitions. The $18 billion valuation places it above Runway (which raised at a $4 billion valuation in 2024) and in territory normally associated with foundation model companies, not video generation applications. The gap between Kling's valuation and its US-based peers in video AI is large enough to represent either a mispricing or a genuine difference in the companies' competitive positions that is not yet reflected in Western investor sentiment.

The General Atlantic lead is a signal worth reading carefully. GA is a growth equity firm that has historically bet on companies with strong emerging-market fundamentals alongside global growth ambitions. Their presence -- rather than, say, a dedicated AI fund or a US strategic -- suggests the investment thesis for Kling is partly a China-market story (where short video AI has regulatory and distribution advantages that US companies cannot replicate) rather than a pure global video AI bet.

Schneider Electric's $3.1 billion for Cognite: why incumbents are buying the data layer

The week's most instructive transaction was not a funding announcement. Schneider Electric, the French energy and industrial automation company with 150,000 employees and approximately $40 billion in annual revenue, acquired Cognite for $3.1 billion (date announced June 30, per database exit records). Cognite built industrial DataOps software: tooling for connecting, contextualizing, and querying operational data from oil platforms, factories, and infrastructure assets. The company had raised approximately $200 million in venture capital over its lifetime before this exit.

Schneider did not acquire Cognite to build an AI product. It acquired Cognite because Schneider's existing customers -- utilities, manufacturers, oil and gas operators -- cannot run AI on their operational data without first solving a data integration and contextualization problem that Cognite addresses. The asset is not the AI; it is the plumbing that makes the operational data accessible to AI. This is the classic pattern of industrial software acquisition: the incumbent buys the data layer because it owns the customer relationship and distribution, then packages the capability into its existing platform.

The $3.1 billion purchase price against roughly $200 million in venture funding represents approximately 15x invested capital. For other industrial data and DataOps startups in the same position -- Seeq, Sight Machine, and similar -- the comp set has been updated. The buyer class is mature industrials, not AI companies. And the acquisition logic is defensive: Schneider's competitors are structurally motivated to buy the same category of company before Schneider's Cognite integration gives it an advantage in AI-enabled industrial services.

Straiker's $64 million defines what AI agent security actually means

Among the week's smaller rounds, Straiker's $64 million Series A (June 29, SiliconAngle) deserves more attention than its size suggests. The Mountain View company builds security software specifically for AI agents -- autonomous systems that can take actions across connected enterprise systems. The round was led by Marathon Management Partners, with Citi Ventures, Illuminate Ventures, Workday Ventures, Bain Capital Ventures, and Lightspeed participating. Revenue has grown more than 15-fold in under a year since a $21 million seed in March 2025.

The company's published research provides context for the growth rate: 36% of successful attacks on AI coding agents result in remote code execution, and 91% of successful attacks on productivity agents lead to undetected data theft. These numbers, if they hold at scale, represent a category of enterprise risk that didn't exist 18 months ago. The security surface of an AI agent is fundamentally different from traditional software: agents reason on the fly, take autonomous actions, and connect to data systems with broad permissions. The attack vectors -- prompt injection, context manipulation, privilege escalation through tool use -- are not well covered by existing enterprise security tooling.

Straiker's CEO previously led Palo Alto Networks' Prisma Cloud division. The leadership profile, combined with Workday Ventures' participation, suggests the company is targeting the same enterprise buyer who is deploying coding assistants and HR workflow agents at scale -- and discovering they need a security layer they did not plan for when they started the deployment. The 15x revenue growth in a year is consistent with a new category that is past the "are enterprises willing to pay for this" stage and into the "how fast can we close" stage.

Three startups built the same product this week without knowing it

Among the new startups added to the database this week, three are building tools to help businesses appear in AI search results -- the recommendations that ChatGPT, Claude, Gemini, and Perplexity surface when users ask product or service questions rather than perform keyword searches. None of them appear to know the others exist, and none have settled on the same name for the category.

Foundin.ai (Netherlands) runs live AI checks showing local businesses whether they appear in ChatGPT, Gemini, Claude, and Perplexity recommendations, then provides action plans to improve AI search visibility. GeoSurge (London) raised $10 million in seed funding (eu-startups, July 1) to help brands understand and influence their presence in AI-generated outputs -- a slightly broader brief that includes brand monitoring alongside optimization. Visiblie (Belgium) closed a 500K convertible loan (tech.eu, July 2) to build AI search visibility tools for businesses.

The timing is not coincidental. Through 2025, most SEO practitioners treated AI search optimization as a secondary concern. By mid-2026, with ChatGPT search reaching meaningful scale and Perplexity active as a product discovery layer, businesses that built customer acquisition on traditional search rankings are finding their presence in AI-generated answers does not correlate with their Google positions. The standard SEO playbook -- domain authority, backlink profiles, structured data -- does not transfer cleanly to answer engine optimization. These three startups are early movers in a category that has been called AEO (Answer Engine Optimization), GEO (Generative Engine Optimization), and AISO (AI Search Optimization) depending on who is naming it.

The category is real. It is also likely overcrowded before it has been formally defined. The SEO tool incumbents -- Semrush, Ahrefs, Moz -- are structurally motivated to absorb this capability into their existing suites rather than allow a new standalone category to mature. The interesting question is whether any of the early AEO startups builds enough proprietary data or defensible methodology to make acquisition the only viable path for the incumbents, or whether the capability gets replicated in-house within 18 months.

Microtrends worth watching

• Quantum Systems at $8 billion: The European drone intelligence company raised $1.2 billion on July 3 (Bloomberg video coverage), doubling its valuation to $8 billion. A European aerospace and defense AI company at hyperscaler-adjacent scale is a new data point. The round suggests European defense AI is now fundable at sizes that were only available to US and Israeli defense tech companies 24 months ago.
• Omen AI's cooling intelligence thesis: $31 million for continuous monitoring of liquid coolant systems in data centers (SiliconAngle, June 29). Unsexy infrastructure, but liquid cooling failure in an AI training cluster is catastrophic and hard to detect before it becomes expensive. As Crusoe, Switch, and the rest of the infrastructure buildout scales up, the operations intelligence layer -- predictive monitoring, anomaly detection for physical systems -- becomes a meaningful market.
• Venice.ai's $1 billion valuation for private AI: $65 million at a $1 billion valuation for a platform offering private, uncensored AI without content moderation policies (SiliconAngle, July 1). The billion-dollar number is a data point about how large the market for AI output that does not comply with the major providers' content policies actually is -- and how seriously investors are taking it.
• Oxmiq Labs, Stathera, Luxonis -- custom silicon and vision AI picking up pace: Three hardware-oriented companies raised this week below the radar: Oxmiq Labs ($35M, custom AI silicon design cost reduction), Stathera ($35M, vacuum-sealed silicon oscillators for AI chips), Luxonis ($14M, vision layer for intelligent automation). The hardware stack below the GPU is becoming investable again.
• Higharc's $95M Series C for homebuilding AI: Homebuilding is a sector where software digitization is still incomplete -- floor plans, permitting workflows, and construction scheduling are poorly digitized across most production homebuilders. Higharc is raising at a level that suggests the market is more willing to pay for AI applied to physical construction workflows than many assumed.

What might happen next week

Two predictions, both falsifiable:

Together AI and Groq will each announce a named enterprise anchor before the end of July. Two inference cloud companies raised $1.45 billion combined in nine days with effectively the same pitch: dedicated AI compute, cheaper and faster than hyperscalers, purpose-built for open-source model inference. The capital is now in both companies' accounts. The next phase of competition is customer announcements, not fundraising. One of them -- most likely Together AI given Aramco's strategic interest in enterprise AI deployments in Gulf markets -- announces a named partnership or preferred provider agreement before the end of July. The other follows within two weeks.

The AI search visibility category gets a canonical name and an acquisition inquiry before Q4. GeoSurge raised $10 million this week on the thesis that brands need help understanding and shaping their AI search presence. Foundin and Visiblie are operating in the same space. When a category has three independent startups building nearly identical products in a single week's database activity, it is past the "will anyone pay for this" question and into the "who wins" question. Semrush or Ahrefs will make an acquisition inquiry in the AEO space before October 2026. The category's canonical name -- AEO, GEO, AISO -- will be settled by whoever the first major trade press outlet uses consistently, which should happen within 60 days.