While Fluidstack captured headlines in late July with an $830 million funding round—another hyperscaler-infrastructure bet at a dizzying $7.5 billion valuation—a more revealing pattern was taking shape in the weeks surrounding that announcement. Eight startups closed Series A rounds totaling over $200 million between late June and late July 2026, and not one of them builds foundation models. Not one sells GPUs.
Instead, they're tackling what founders now call "second-order AI problems": grid capacity modeling, liquid coolant monitoring, AI agent governance, and robotic labor to build solar plants fast enough to power the whole endeavor. The infrastructure layer beneath the infrastructure, if you will.
The timing tells its own story. The International Energy Agency's 2026 update projects global data center electricity consumption will roughly double—from about 485 terawatt-hours in 2025 to approximately 950 TWh by 2030. U.S. regulators responded in June with explicit orders to fast-track large load interconnections to the grid. Then, on July 14, New York imposed a data center construction moratorium, citing energy strain and community pushback.
The gold rush, it seems, has produced both acceleration and backlash. And in that tension, investors are finding opportunity.
When the Grid Becomes the Bottleneck
The infrastructure reality is stark, perhaps more so than the AI hype cycle suggests. The Department of Energy's July 2026 draft National Transmission Needs Study flags urgent capacity constraints driven by data centers, advanced manufacturing, and other large industrial loads. The Federal Energy Regulatory Commission's June 18 orders aim to "supercharge" grid integration, but that regulatory tailwind is colliding with grassroots resistance. Seattle debated a ban in late July. Noise litigation over data center cooling systems is spreading across multiple jurisdictions.
That's where companies like Axle Energy see an opening. The startup announced a $25 million Series A on July 8, led by Energize Capital, to expand its virtual power plant platform across Europe. Co-founders Karl Bach and Archy de Berker position distributed flexibility—aggregating batteries, EVs, and heat pumps into grid-interactive resources—as essential infrastructure. Their framing is blunt: "AI, war, and weather push grids to resilience limits."
A week later, Gridcog closed a $10 million Series A led by ABB, with participation from utilities Axpo, DNV, and VERBUND X, to scale deterministic modeling software for hybrid renewable and battery projects. The CEO's pitch underscores what's missing in much of the energy-tech stack: "Most modelling tools ask you to trust a number you can't trace. Gridcog does the opposite: every assumption is visible, every result goes down to the interval."
Then there's Gritt, which emerged from stealth on July 21 with a $26 million Series A. The company deploys robots with AI-guided controls to accelerate solar panel installation—claiming throughput improvements of 3-5x over manual crews. It has already contracted to install 2.8 gigawatts over the next 18 months, targeting the EPC labor bottleneck that threatens to delay the power projects AI campuses need. Not the sexiest pitch, perhaps. But necessary.
Liquid Cooling's Hidden Complexity
As liquid cooling shifts from niche to mainstream—Grand View Research pegged the data center liquid cooling market to reach $29.5 billion by 2033, up from single-digit billions recently, at a compound annual growth rate exceeding 20 percent—materials and failure modes are emerging as gating factors. TrendForce estimates the AI optical transceiver market will hit $26 billion in 2026, with co-packaged and near-packaged optics markets expected to surpass $39 billion by 2030.
Omen AI, which closed a $31 million Series A on June 30 led by Nava Ventures, addresses what happens when those systems fail. The company deploys continuous spectroscopic sensors to monitor coolant, oil, and water in real time across data centers managing substantial capacity. The founder's commentary in the announcement is characteristically blunt: "Taking a sample, shipping it to a lab, and waiting days for results is dangerously inadequate when you're protecting billions in GPU infrastructure."
Applied Computing's $20 million Series A, announced July 15 with KBR leading and Databricks Ventures participating, takes a different angle. The startup is building a foundation model for entire oil, gas, and petrochemical plants—part of the broader wave of power-intensive industries preparing for AI-driven optimization. Infrastructure twice removed from the model layer, yet directly tied to energy demands the AI buildout creates.
It's worth pausing on that point. The money flowing into these companies isn't chasing incremental improvements to inference speed or parameter count. It's chasing the unglamorous stuff that determines whether the AI boom scales or stalls.
The Agent Security Problem
The third cluster of July deals addresses what Gartner now calls "agent sprawl."
The firm warned in July 2026 guidance that by 2029, at least 70 percent of organizations running production agentic AI in infrastructure and operations will suffer a material security, service, or cost incident due to insufficient runtime controls. Verizon's 2026 Data Breach Investigations Report, published in May, found that vulnerability exploitation has overtaken stolen credentials as the top breach entry vector, while shadow AI usage tripled to 45 percent of surveyed organizations.
Empirical Security raised $25 million on July 20, led by Brightmind Partners, to build per-customer predictive exploit models amid what the company frames as "AI-accelerated exposures." The founding team previously built Kenna's EPSS scoring system. They're betting that static vulnerability databases won't keep pace with AI-augmented attackers—a reasonable wager, given the trajectory.
Two days later, AegisAI announced a $36 million Series A from Battery Ventures, with Accel and Foundation Capital participating, to deploy autonomous AI agents for email security and spear-phishing defense. The founders led Google's reCAPTCHA and Safe Browsing teams. Their July 23 statement pulls no punches: "The most immediate, catastrophic risk to your organization isn't an AI agent hacking your firewall. It's an AI model manipulating someone into clicking a link or sharing credentials. You cannot patch human trust."
Hush Security closed the wave on July 28 with a $30 million Series A that brought Akamai in as a strategic investor alongside Battery Ventures and YL Ventures. The company's pitch is governance and access control for the "non-human workforce"—secrets, service accounts, and now AI agents. As agent-to-agent communication proliferates, identity sprawl becomes an attack surface regulators and CISOs are only beginning to map.
What's notable here isn't just the money. It's the speed. Three major security rounds in eight days, all betting that the AI agent boom will outpace governance infrastructure. That kind of clustering doesn't happen by accident.
Strategic Investors Show Their Hand
The investor mix signals where incumbents see moats eroding—or where they're hedging against their own obsolescence.
ABB led Gridcog's round. Utilities Edison International and NVentures co-invested in ThinkLabs AI's $28 million Series A earlier this year, targeting AI-native three-phase AC power-flow analysis. Akamai took a strategic position in Hush. National Grid committed $1.75 billion in a July 1 minority investment in Joulent to develop multi-gigawatt power infrastructure for AI and industrial loads. Not a Series A, admittedly, but a data point on the scale of capital required one layer below the data center operators.
Corporate development teams at energy companies are no longer spectating. They're writing checks and taking board seats in companies that didn't exist two years ago, solving problems that didn't exist five years ago.
The tell, perhaps, is in what's not being funded at Series A scale in this window: chatbots, wrappers, consumer AI novelties. The pattern is infrastructure, materials, security, and energy. The second-order bottlenecks.
The Regulatory Paradox
The paradox shaping the next 18 months is regulatory—and it's messier than the headlines suggest.
Federal agencies are accelerating interconnection approvals while states and municipalities slam on the brakes. That bifurcation favors startups solving for flexibility—virtual power plants, grid analytics, novel siting strategies—over those betting on traditional greenfield expansion. It also suggests the coming wave of funding will tilt toward companies that can unlock latent capacity in existing infrastructure rather than those building net-new megawatt hours from scratch.
The materials and cooling stack is less forgiving. Co-packaged optics are still in early volume production, with large-scale deployment expected soon. Liquid cooling adoption is outpacing thermal interface material innovation, creating supply chain pinch points industry veterans are calling "the new chip shortage." If those bottlenecks hold, the foundation model labs slow down not because they run out of ideas but because they literally can't keep the lights on.
For founders, the pattern is instructive, maybe even obvious in hindsight. The question isn't "Can we build a better model?" but "What breaks when everyone else builds better models?"
Grid capacity. Cooling systems. Agent governance. Physical construction throughput for power projects.
Each Series A in this cohort is a bet that the answer to that question is worth more than incremental improvements to the models themselves.
The New Narrative
Eight deals in two months. Over $200 million deployed. No GPUs sold.
The AI infrastructure gold rush has moved from silicon to kilowatts, from prompt engineering to thermodynamics. The smart money isn't betting on who builds the best AI anymore. It's betting on who keeps the power on while they try.
That shift—from what AI can do to what it demands—may define the next chapter of the boom. Or, depending on how quickly these second-order problems get solved, what comes after it.
