Blackstone Real Estate has closed on a 134,200-square-foot research building at 350 Cobalt Way in Sunnyvale, California — a footprint SiliconValley.com’s September 2026 coverage independently verified — paying $95 million for a property leased entirely to Nvidia through August 2036, a lease term GlobeSt also confirmed in its own September 2026 reporting, according to The Business Journals. County records reviewed by The Registry put the actual transfer price at $99.6 million, a gap that reporters attribute to rounding in the initial headline figure. Either way, the deal lands as one of the clearest signals yet that electrical capacity, not square footage, now sets the price on Silicon Valley real estate tied to artificial intelligence.
The building isn’t a data center in the conventional sense. It’s a former Fujitsu facility built in 1978 and renovated in 2022 — a timeline Connect CRE corroborated in its own September 2026 coverage of the sale — carrying roughly 10 megawatts of electrical capacity that Nvidia has been using for research and development work with data-center-grade infrastructure attached, per The Mercury News. That power hookup is the entire reason a 1970s-era R&D building just traded for nearly $100 million.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The Deal, in Numbers
Blackstone bought the property from PSAI Realty Partners, which had acquired it for roughly $31 million in 2023 after Fujitsu vacated. Nvidia signed a full-building lease and began occupying the entire property in 2024, a timeline Hoodline’s September 2026 report also confirmed, and the space has sat at the center of the chipmaker’s Sunnyvale campus expansion ever since. Marketing materials from JLL, the brokerage that handled the listing, described 350 Cobalt Way as “one of Nvidia’s few research and development labs with data center facilities, operating as mission-critical infrastructure,” per The Registry’s reporting on the listing language.
| Detail | Figure |
|---|---|
| Address | 350 Cobalt Way, Sunnyvale, CA |
| Buyer | Blackstone Real Estate (affiliate) |
| Seller | PSAI Realty Partners |
| Reported headline price | $95 million |
| Recorded transfer price | $99.6 million |
| Building size | 134,200 square feet |
| Built / renovated | 1978 / 2022 |
| Electrical capacity | ~10 megawatts |
| Tenant | Nvidia (full building) |
| Lease term | Through August 2036 |
| PSAI’s 2023 purchase price | ~$31 million |
Do the math on that spread and the property’s value came in at more than 3 times its 2023 sale price, Hoodline reported in September 2026, mostly because Nvidia moved in and the market’s appetite for anything with committed power capacity exploded. At the recorded $99.6 million transfer price, that works out to roughly $742 per square foot, according to a September 2026 calculation from The American Developer, a figure that would be hard to justify for a 1970s-era building almost anywhere outside a power-constrained submarket like Sunnyvale. A vacant office-slash-lab building with electrical infrastructure sitting idle in 2023 became a trophy asset once a tenant with a 12-year lease and a name like Nvidia attached itself to the meter.
Why 10 Megawatts Changes Everything
Ten megawatts sounds modest next to the gigawatt-scale campuses hyperscalers are chasing across Texas, Ohio, and Virginia. But in Sunnyvale, where land is scarce and utility interconnection queues stretch for years, existing capacity is worth more than acreage. Getting new power to a site in Silicon Valley today means filing with the local utility, waiting on substation upgrades, and often waiting years for transmission work that has nothing to do with construction schedules. A building that already has 10 MW live and metered skips that queue entirely.
The Mercury News captured the dynamic directly: the property “has nearly 10 megawatts of electrical capacity, a scarce advantage as AI and advanced manufacturing companies demand more power,” according to reporting carried by The Business Journals. That single sentence explains the entire valuation logic behind the deal. Nvidia isn’t paying for office space. It’s paying, through its lease, for guaranteed electrons in a market where electrons are the bottleneck.
This is a broader shift playing out across the semiconductor and cloud industry, one that mirrors the capacity crunch documented in AMD’s 107% jump in data center revenue and the power-constrained competitive positioning behind Arm’s push into AI-optimized CPU designs. When every major chip vendor is racing to ship more silicon, the real ceiling on deployment isn’t fabrication, it’s the grid.
Blackstone’s Bigger AI Infrastructure Play
The Sunnyvale purchase didn’t happen in isolation. On the same day the deal was reported, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR aimed at mobilizing more than $500 billion in third-party capital for AI-compute infrastructure, according to Nvidia’s official newsroom. That announcement is a financing framework, not a single transaction, but it puts the Sunnyvale deal in context: Blackstone isn’t dabbling in one building. It’s positioning itself as one of the primary capital partners behind the entire AI buildout.
Nvidia also made a strategic investment in Lancium, a Blackstone portfolio company, alongside a collaboration to deploy Nvidia AI-factory technology across Lancium’s campuses. Lancium has said it has approximately 4 gigawatts of capacity under lease and a development pipeline exceeding 15 gigawatts of powered land, according to industry tracking cited in the same reporting cycle. Set against that backdrop, a 10-megawatt Sunnyvale R&D building is a rounding error in capacity terms, but it’s a symbolically important one: it shows Nvidia and Blackstone transacting directly on real estate, not just financing vehicles.
Blackstone has also been reported to be pursuing a joint venture with Google, internally referred to as Project Braid, targeting 500 megawatts online by 2027 with roughly $5 billion in Blackstone equity capital committed through its funds. None of that is confirmed at the same evidentiary level as the Sunnyvale purchase, but the pattern is consistent: Blackstone is stacking data-center-adjacent bets across multiple deal structures, from direct property acquisition to joint ventures to portfolio-company investment.
How the Deal Compares to Other 2026 AI Infrastructure Bets
| Deal / Vehicle | Parties | Reported Scale |
|---|---|---|
| 350 Cobalt Way acquisition | Blackstone / PSAI Realty (Nvidia tenant) | $95M–$99.6M, 10 MW capacity |
| AI-compute financing platforms | Nvidia, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR | $500B+ capital mobilization target |
| Brookfield AI Infrastructure Fund | Brookfield, with reported Nvidia commitment | $10B fund target, $2B Nvidia commitment reported |
| Google–Blackstone joint venture (reported) | Google, Blackstone | 500 MW target online by 2027, ~$5B equity reported |
| Lancium powered-land pipeline | Nvidia investment, Blackstone portfolio company | ~4 GW under lease, 15 GW+ pipeline |
Lined up this way, the Sunnyvale deal is the smallest number on the page by a wide margin, and that’s precisely the point. It shows that the same capital pursuing gigawatt-scale campuses is also willing to pay a premium for a single building with a fraction of that capacity, as long as the tenant and the power hookup are already locked in. Scarcity, not scale, is driving price at the small end of the market just as much as at the large end.
A Building’s Journey: From Fujitsu to Nvidia to Blackstone
The property’s ownership history is a compact case study in how fast valuations have moved. Fujitsu occupied the building for years before vacating it. PSAI Realty Partners picked it up in 2023 for around $31 million, at a moment when the site sat empty and the AI infrastructure boom hadn’t yet reached full throttle. Nvidia then leased the entire building in 2024, converting an underused asset into a fully occupied, long-term-leased property with a Fortune 500 tenant attached to it.
That single tenancy change is what allowed the building to sell for roughly triple its 2023 price within about two years. It’s a pattern that’s shown up elsewhere in the AI hardware ecosystem too: assets that looked ordinary before the AI buildout, from used GPUs to powered land parcels, have repriced sharply once demand caught up with supply. The resale premiums now showing up on Nvidia B200 accelerators trading above their original launch price reflect the same underlying scarcity dynamic playing out in silicon rather than real estate.
What Named Sources Are Actually Saying
The reporting trail on this deal runs through commercial real estate outlets rather than Blackstone or Nvidia press releases, and the record is worth laying out plainly. The Registry broke the transaction with the headline “Blackstone Buys Nvidia-Leased 134,200 SQFT Sunnyvale R&D Building for $99.6MM,” a figure that matches county transfer records more closely than the rounded $95 million headline that circulated elsewhere, according to The Registry.
NewsBreak’s aggregation of the local coverage summarized the core mechanics of the deal: “Blackstone Real Estate, through an affiliate, acquired the 134,200-square-foot building at 350 Cobalt Way for $99.6 million,” according to NewsBreak’s summary of local reporting.
On the power angle specifically, reporting carried by The Business Journals noted that the former data center “has nearly 10 megawatts of electrical capacity, a scarce advantage as AI and advanced manufacturing companies demand more power,” according to The Business Journals. And JLL’s marketing language, cited in The Registry’s coverage, framed the building as “one of Nvidia’s few research and development labs with data center facilities, operating as mission-critical infrastructure,” according to The Registry.
No Blackstone executive, Nvidia executive, or PSAI representative appears on the record with a direct quote in the available coverage. That’s worth flagging rather than papering over: this is a transaction reported through property records and brokerage marketing material, not a deal either company has publicly commented on.
Historical Context: Real Estate Has Always Followed Infrastructure
Commercial real estate has repriced around infrastructure scarcity before. Fiber-lit buildings commanded premiums during the dot-com buildout. Rail-adjacent warehouses spiked in value once e-commerce logistics needed same-day delivery radii. What’s different about the current cycle is the speed. PSAI bought this building for $31 million in 2023, before the generative AI capital cycle had fully matured, and sold it for more than triple that less than three years later, once a single-tenant AI lease attached itself to the asset.
That compressed timeline echoes what’s happened across the broader AI supply chain in 2026. Component and infrastructure assets that once traded on conventional depreciation curves are now trading on capacity scarcity instead. The same logic underpins Qualcomm’s $60 billion AI chip agreement with AWS and the financing structures behind SoftBank’s $11 billion bond raise tied to its OpenAI stake. Capital is chasing anything that removes a bottleneck, whether that bottleneck is chip supply, financing, or, in this case, an electrical hookup in Silicon Valley.
Market Impact: What This Means for Data Center Investors
For institutional real estate investors, the Sunnyvale deal reframes what counts as a “data center asset.” It’s no longer just purpose-built hyperscale campuses with raised floors and redundant cooling loops. Any property with existing utility capacity, sufficient structural load, and proximity to a chip company’s campus is now a plausible AI infrastructure play, even if it was originally built as office or light-industrial space in the 1970s.
That reframing has knock-on effects for appraisal methodology. Cap rates for properties with confirmed electrical capacity are compressing relative to comparable buildings without it, because buyers are pricing in the avoided cost and delay of a new utility interconnection. Brokers marketing similar properties in power-constrained submarkets, from Sunnyvale to Santa Clara to parts of Austin, are likely to start leading with megawatt figures rather than square footage, following the JLL playbook used on 350 Cobalt Way.
It also raises the stakes for chip and cloud tenants signing long leases. A 12-year commitment, like the one Nvidia holds through 2036, gives a landlord bankable, investment-grade cash flow, which is exactly the kind of asset institutional capital like Blackstone wants to hold. Expect more chip companies to lock in decade-plus leases on power-secured buildings specifically because it makes the underlying real estate more attractive to buy and finance.
The Silicon Valley Power Squeeze
Silicon Valley’s grid wasn’t built for this. PG&E’s interconnection queue in parts of Santa Clara County has stretched into multi-year waits for large new loads, pushing companies toward exactly the kind of retrofit-and-reuse strategy Nvidia and Blackstone just executed. Buying a building that already has a transformer and a service agreement in place is often faster than permitting a new one from scratch, even accounting for the acquisition premium.
This dynamic is compounding a broader hardware supply squeeze that’s already showing up in component pricing. Google’s push to put TPUs into orbit to sidestep terrestrial power and cooling constraints is one extreme response to the same underlying problem. Buying a power-secured office building in Sunnyvale is a far more conventional one, but it’s driven by the identical scarcity.
Predictions: Where This Trend Goes Next
- More single-tenant, chip-company-leased buildings with legacy electrical infrastructure will trade at multiples of their pre-AI valuations over the next 12 to 18 months, following the pattern set at 350 Cobalt Way.
- Institutional buyers will increasingly disclose megawatt capacity alongside square footage in acquisition announcements, treating power as a headline metric rather than a footnote.
- Expect more direct Nvidia involvement in real estate and infrastructure financing, building on the Lancium investment and the $500 billion financing-platform framework announced with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR.
- Secondary Silicon Valley submarkets with older industrial and R&D stock, not just Sunnyvale, will see renewed investor interest specifically for buildings with existing utility service.
- Lease terms on power-secured properties will keep stretching toward the decade-plus range, since longer commitments make the underlying real estate more financeable for buyers like Blackstone.
The Skeptic’s Case
Not everyone is convinced this is a durable trend rather than a pricing anomaly tied to one high-profile tenant. A single-tenant building carries concentration risk: if Nvidia’s Sunnyvale footprint needs ever shrink or consolidate elsewhere, an owner is left holding a 1978-era structure whose value was priced almost entirely on the strength of one company’s lease. The $99.6 million recorded price also sits well above what comparable vacant industrial space in the same submarket would fetch without a marquee tenant attached, meaning the premium is fragile if AI capital spending cools.
There’s also a financing-market question hanging over the entire sector. The $500 billion figure tied to the Nvidia-backed financing platforms is a mobilization target, not committed capital already deployed, and the Google-Blackstone joint venture details remain unconfirmed at the same level of certainty as the Sunnyvale transfer record. Investors chasing power-secured real estate should treat the headline numbers in this space as directional rather than final until deals close and are recorded.
What Comes Next for Blackstone and Nvidia
Neither Blackstone nor Nvidia has publicly detailed further plans specific to 350 Cobalt Way. But given the timing, with the Sunnyvale closing landing the same week as the $500 billion financing-platform announcement and the Lancium investment, it’s reasonable to expect Blackstone to keep hunting for similar power-secured, single-tenant assets across the Bay Area and other AI-hardware-adjacent metros. Nvidia, for its part, has signaled through both the lease structure and its infrastructure financing partnerships that it intends to keep expanding its physical footprint well beyond chip design offices, into the kind of mission-critical, power-heavy facilities that used to be the exclusive domain of cloud hyperscalers.
For now, the clearest takeaway is the price tag itself. A 48-year-old building without a single new square foot of construction just changed hands for nearly $100 million, and the reason had almost nothing to do with the walls and everything to do with what’s plugged into them.
Frequently Asked Questions
How much did Blackstone actually pay for the Nvidia building?
Reports vary slightly: The Business Journals’ headline figure was $95 million, while The Registry and The Real Deal cited a recorded transfer price of $99.6 million based on county records.
Where is the building located?
350 Cobalt Way in Sunnyvale, California, in the heart of Silicon Valley.
Is Nvidia the owner of the building?
No. Nvidia is the tenant, leasing the entire 134,200-square-foot property through August 2036. Blackstone is now the landlord.
Why does the building’s electrical capacity matter so much?
The property carries roughly 10 megawatts of existing electrical capacity, which is valuable because new utility interconnections in Silicon Valley can take years to secure. Buying a building with power already in place lets a tenant skip that wait.
Who owned the building before Blackstone?
PSAI Realty Partners, which bought it for roughly $31 million in 2023 after Fujitsu vacated the site, before Nvidia signed its full-building lease in 2024.
Is this part of a larger Blackstone AI strategy?
Yes. Blackstone is also named among the financial institutions partnering with Nvidia on financing platforms aiming to mobilize more than $500 billion for AI-compute infrastructure, and Nvidia has separately invested in Lancium, a Blackstone portfolio company with a multi-gigawatt powered-land pipeline.
Does this deal include a full data center build-out?
No. The property is described as a research and development lab with data-center-grade infrastructure and mission-critical systems, not a purpose-built hyperscale data center campus.
What does this mean for other AI-adjacent real estate?
It signals that any building with confirmed electrical capacity, even older or smaller properties, can now command significant premiums if it’s leased to an AI-hardware tenant, reshaping how brokers and investors value commercial real estate near major chip and cloud campuses.
Related Coverage
- Adreno Neural Fusion Cuts Mobile Gaming Power 40% [2026]
- Nvidia Smooth Motion Setup Guide: 13 Steps, 40 Min [2026]
- Nvidia's Top Leaker Confirms RTX 60 Delay to 2028 [2026]
- GPU Passthrough Setup: IOMMU/VFIO in 13 Steps [2026]
- Arm's AGI CPU Takes Aim at AMD's $6.7B Surge [2026]
