Disruptive's $10 Billion Blind Pool Just Became the Largest Bidder in the AI Power Bottleneck — and the Same Assets Trade at Half the Price in Public Markets

Disruptive, the Dallas-based investment firm founded by Alex Davis, announced on Oct. 27 that it had closed a $10 billion fund — the largest venture fund ever raised in Texas — to back roughly ten late-stage companies with checks of up to $1 billion each. The structure is the story. For fourteen years the firm raised capital deal-by-deal through single-purpose vehicles, so every dollar an investor committed went into a named company with a disclosed valuation. This vehicle is a blind pool: investors commit before they know which ten assets the money buys.

That concentration is the mechanism. A fund that writes ten checks averaging $1 billion is not diversifying across a portfolio of companies; it is choosing a handful of assets and underwriting them like an owner. The stated focus spans artificial intelligence, AI infrastructure, defense technology and enterprise software, and the named portfolio includes Palantir, Groq, ElevenLabs, Helsing, Shield AI, Databricks, Reflection AI, Airbnb and Spotify. Davis's own framing is that private technology companies are \"growing larger, staying private longer and requiring more capital than ever before.\"

The decisive evidence for what a concentrated $1 billion check buys sits in the suppliers' order books, not in the fund's press release. In the week the flagship fund closed, GE Vernova reported Q3 2026 orders of $29.3 billion, up 96% organically, and a total backlog of $197 billion that grew $20 billion in a single quarter. Its gas equipment backlog plus slot reservations rose from 116 GW to 122 GW, and management said it expects at least 130 GW by year-end. Power segment EBITDA margin reached 21.4%, up 800 basis points organically, and full-year revenue guidance was raised to $46.5–47.5 billion.

Alongside it, Vertiv posted Q3 net sales of $4,313 million, up 61%, with organic orders more than doubling year over year, and lifted full-year 2026 sales guidance to $14,450–14,850 million with adjusted EPS of $7.10–7.20. Eaton reported a 43% year-over-year increase in Electrical segment backlog, with Electrical Americas up 33% and Electrical Global up 103%, and told investors its US data center backlog now stands at 307 GW — about fifteen years of construction at 2025 build rates. These are not software vendors repricing sentiment. These are the physical suppliers of turbines, switchgear and transformers, and they are sold out.

Now the number that should stop a venture investor cold. Groq — the AI inference chip designer Disruptive has backed twice — raised $350 million on Aug. 17, 2026 at a $3.5 billion valuation, roughly half the $6.9 billion it was marked at in September 2025. Disruptive led both rounds. Its own flagship AI infrastructure holding was marked down by about half in the same window in which its public suppliers of power equipment were repricing upward. Private marks on the compute layer are compressing; public marks on the physical layer are not.

The gap is not a contradiction in the data; it is a difference in what each buyer is buying. Groq's private mark prices a chip architecture that NVIDIA has already licensed for $20 billion, with roughly $17 billion of proceeds shared among all backers. The public mark on GE Vernova prices a turbine slot that cannot be manufactured for the better part of a decade. When a single manager writes ten $1 billion checks into AI infrastructure, it is not underwriting chip architectures at all — it is underwriting access to power, grid capacity and industrial equipment, which is precisely the asset class the public market is already repricing.

An AI infrastructure company of the kind Disruptive plans to fund consumes megawatts, not abstractions. Groq now runs 13 data centers and told investors in August it is scaling from 54 MW toward more than 200 MW by 2027. Every incremental megawatt of that build has to be connected, switched and protected, which is where the order books sit. Quanta Services — the transmission and substation contractor — carried a $53.4 billion total backlog at the end of Q2 2026 and raised full-year revenue guidance to $39.3–39.7 billion.

The upstream layer is where the capacity actually gets added, and it is not American. HD Hyundai Electric signed a $721 million supply agreement for a North American data center in July 2026, delivering through 2028, and reported Q3 sales up 18.5% year over year with an order backlog of $8.84 billion. Hyosung Heavy Industries ended Q2 2026 with KRW 17.5 trillion of consolidated backlog, US demand above 50% of volume, and a record KRW 787 billion US power-equipment order earlier in the year. Japanese heavy industry holds the turbine and transformer end: Mitsubishi Heavy Industries said gas-turbine order pace in the first half of 2026 matched or exceeded 2025's record, with slots booked into 2027 and 2028.

The demand side is what makes this durable rather than cyclical. Amazon, Microsoft, Alphabet and Meta together plan roughly $700 billion to $725 billion of capital expenditure in 2026, up about 77% from 2025. On the supply side, Constellation Energy executed 920 MW of long-term nuclear contracts in Q2 2026 alone, and its October agreement with Google covers 890 MW of upgraded nuclear capacity alongside more than $4.3 billion of Constellation investment in uprates. Vistra and Meta are building out 2,609 MW of nuclear capacity in the PJM market on the same template.

The strongest argument for owning the public supply chain is that its multiples have already de-rated even as its order books hit records. GE Vernova trades at 21.0x trailing EV/EBITDA on a business that just grew orders 96% and backlog $20 billion in one quarter. Vertiv sits at 30.8x, Quanta Services at 31.4x, Eaton at 26.9x. Meanwhile the private marks that would anchor a venture portfolio — Groq at $3.5 billion, half its year-ago value — are falling. The public market is paying less per unit of growth than the private market is, which is the reverse of the usual venture-cycle setup.

That de-rating happened for a reason, and it is not a demand problem. On Oct. 28 GE Vernova reported a blowout quarter and the stock fell 1.9% on the day; on the same day Vertiv rose 16.5%. The market is not paying for the order book, it is paying for whether that order book converts into cash without a working-capital disaster. Vertiv ended Q3 with $4.1 billion of customer prepayments before delivery — customers funding the build. Capital intensity, not demand, is what the multiple is discounting.

The knock-on for the venture cycle is that the marginal private dollar now has fewer places to go. Global venture funding reached a record $510 billion in the first half of 2026, and 16 companies raised billion-dollar-plus rounds in the second quarter alone, accounting for $108.6 billion or 53% of all funding in the quarter. With Disruptive now able to write $1 billion single checks, the fund is competing for the same late-stage AI assets that crossover and growth investors have been chasing — and the same power and industrial assets that the public supply chain is selling into. Capital-markets plumbing has become the bottleneck, not capital.

In the days to quarters horizon, the question is conversion, not demand. GE Vernova has guided to at least 130 GW of gas backlog and slots exiting 2026 and to more than $200 billion of total backlog in the fourth quarter, and it has an agreement to acquire Trench Group for grid equipment. Eaton reports Q3 2026 on Nov. 3, and Quanta Services reports on Oct. 29 — watch whether backlog converts to revenue at the guided margin, and whether customer prepayments keep building. NVIDIA reports fiscal Q3 in mid-November, and Palantir on Nov. 2, which matters because Palantir is a named Disruptive holding and a read on the defense and enterprise software leg of the fund.

Over one to three years, the risk moves from valuation to financing. The AI buildout has accumulated over $500 billion of debt by late 2026, roughly a third of 2026 AI capital spending is expected to be debt-financed, and the bond market may face a gap approaching $1 trillion. Behind-the-meter power projects — which private capital has been using to bypass grid queues — carry their own reliability and technology risk. If credit tightens, the physical asset that everyone is buying becomes a financing problem, and the private mega-checks that rely on continued AI capital formation look expensive first.

Disruptive's flagship fund is the clearest signal yet that venture capital has stopped being a patient, diversified early-stage business and become a late-stage owner. A $10 billion blind pool writing ten checks of up to $1 billion each concentrates the same way an infrastructure fund concentrates: few assets, large positions, long duration, no second bite. That is a structural change in who supplies the AI buildout, because a manager that writes $1 billion into a late-stage AI company is, in practice, buying that company access to power, turbines, transformers and grid capacity.

The public supply chain is already selling out that capacity: GE Vernova at 122 GW of gas backlog and slots, Eaton with a 307 GW US data center backlog, Hyosung Heavy Industries with the US above half its volume, Quanta Services with $53.4 billion of work booked. Meanwhile the private marks on the compute layer have fallen — Groq to $3.5 billion, about half of a year earlier, in a round Disruptive led. The trade is not a disagreement about AI. It is a disagreement about which layer of the stack gets paid, and the answer now looks like the layer with turbines and transformers in it.

The disciplined version of the call is narrow. Own the suppliers whose order books are contracted, whose customers prepay, and whose multiples have de-rated while volume doubled. Be careful of the private marks that the new fund will set, because the one observable data point in this cohort went the wrong way. And watch the financing: when roughly a third of AI capital spending is debt-financed and private credit defaults are forecast to reach 8%, the bottleneck stops being equipment and becomes capital itself — at which point every layer of the stack reprices, and the assets with real backlog are what survive.