Invest1 distinct publisher3 min readPublished
The colocation incumbent is now worth $100bn, roughly 47% more than Digital Realty, on the strength of an Nvidia and Together AI inference program that does not open until the first quarter of 2027 and whose invoices Together sends.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
The structure is the part worth pricing. Together AI is the seller of record and bills the end customers who use the program [4], so the token spend from a customer running one of its 200 open-source models [5] lands in Together's ledger, and what Equinix collects is what it has collected since the 1998 dot-com build [23]: space, power, cooling and security for the box [10]. Maryam Zand, the vice president running Equinix's AI ecosystem strategy, calls it an inference platform as a service that lets customers optimize their tokenomics, meaning reduce costs [19]. Cutting a tenant's cost per token fills cabinets, but it does not hand Equinix the meter.
The repricing, meanwhile, has already happened. A 33% gain this year on a $100bn market cap [6] works back to roughly $24.8bn of value added, since $100bn less $100bn divided by 1.33 leaves $24.8bn [1], and that increment alone is about 36% of Digital Realty's entire $68bn [9][2]. Equinix now carries about 47% more equity value than the rival that leaned into large-scale hyperscale facilities [18][3]. Inference Exchange opens in the first quarter of 2027 [3], and the financing terms were not disclosed [2], which leaves a price built on a rent curve nobody has invoiced.
The mechanism underneath it is real enough. Equinix runs 281 legacy colocation facilities across 77 metropolitan areas on six continents [11], about 3.6 sites per metro [5], and inference has become more critical than training as apps turn agentic, pushing workloads onto a wider variety of chips including CPUs [16]. Jensen Huang, speaking by video with Equinix chief executive Adaire Fox-Martin, said the facilities put you close to where the action is, where all the sensors are [21]. With over 10,500 customers against that $100bn [10][6], the average tenant carries about $9.5m of market value, which is a broad book rather than a wager on two or three labs.
What the strategy forecloses is the big end. Most xScale facilities sit under 100MW while some AI data centers are measured in gigawatts [13], so it would take at least ten of Equinix's largest to match one campus [7], and Vlad Galabov, the data center analyst, says Equinix was too slow to react to gigawatt-scale demand [14]. Against Goldman Sachs Research's projection of more than $5tn of hyperscaler spend by 2030 [12], Equinix's whole equity is 2% of that number [4]. The capital is going into metro interconnection and cooling instead, with sites optimized for Nvidia's B300 Blackwell Ultra and some liquid-cooled halls for Vera Rubin [20], plus a new multi-cloud connectivity service, Fabric One [22].
The more interesting version of the thesis is that the 33% is paying for scarce interconnection in 77 metros [11], with the Nvidia name as the permission slip, not for AI revenue itself. The counter-thesis is Galabov's [14]: the incumbent missed the demand that mattered and is now selling adjacency. Disclosure would settle it: interconnection and cabinet revenue growing faster than it did before the deal, and Inference Exchange holding its date [3]. If Equinix's next prints look like ordinary colocation growth, the $24.8bn [1] is just a price that has not yet earned the name value.
Ranked by verification strength, evidence, and original report placement.
On Wednesday, Equinix inked a deal with Nvidia that gives customers a flexible way to run their AI models on open-source cloud platform Together AI.
Financing details for Wednesday's Equinix-Nvidia deal were not disclosed.
The program, called Equinix Inference Exchange, will be available in the first quarter of 2027.
Maryam Zand said Together AI will be the seller of record, billing its end customers who use the new program.
Equinix's stock price is up 33% this year, beating all of megacap tech, and lifting the company's market cap to $100 billion.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Nvidia's Perplexity talks move its money one layer further from its own chips1 distinct publisher
product
CrowdStrike's SafeMind pairs attack-emulation and defensive AI models under one harness1 distinct publisher
product
Nvidia's $105bn guarantee, not OpenAI's balance sheet, is what makes Ohio's 8 gigawatts buildable1 distinct publisher
invest
A twenty-point jump in September hike odds reprices the AI trade the earnings just validated1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One newsroom, on the record at the event
CNBC was in the room and names its sources: Maryam Zand for the footprint, the 2027 date and the billing arrangement; Huang and Fox-Martin on video; Galabov and Chanos for the skeptical side. The market figures are the checkable kind. What no one has verified is the deal itself — financing undisclosed, no Nvidia or Together AI voice on terms — and the two biggest numbers in the piece, Goldman's $5trn and McKinsey's inference share, are borrowed projections rather than tested findings.
Announced now, purchasable in 2027
The thing being celebrated cannot be bought for another two quarters, and not one customer, pilot or committed megawatt is named. What is real today is the old business — 10,500-plus tenants, 281 sites, $2.63bn of quarterly revenue growing 16% — none of which is attributable to Inference Exchange. Fabric One arrives alongside it with the same absence of users.
Valuation running ahead of the product
The stock move is a fact; wiring it to an inference platform that opens in 2027 and whose customers Together AI invoices is the leap. Equinix's whole $100bn is about 2% of the spending wave it is being credited with riding, and the sharpest counterpoint sits in the same piece: Galabov's 'too slow' on gigawatt demand, and his observation that being unexposed to AI bubble risk also means missing the boom. CNBC's decision to print the bear case is what keeps this gap from being wider.
Announcement day, and everyone on stage is long
The primary voices all profit from the framing: Equinix's AI ecosystem VP selling 'tokenomics', Nvidia's CEO praising Equinix's proximity to 'where the sensors are' on the day a joint program launches, and Together AI handed the billing relationship. The counterweights carry their own positions too — Chanos is short both Equinix and Digital Realty, and Galabov hosts a data-center podcast. That does not make anyone wrong; it does mean no disinterested party is quoted anywhere in this reporting.
Firm on the numbers, thin on the mechanics
Attribution is clean and the market figures will not move under scrutiny, so the descriptive spine of this story should hold. Confidence stops at the commercial layer: with no pricing, no capacity commitment, no word on how Equinix books revenue when someone else sends the bill, and roughly two quarters before anyone can test the product, the parts that matter most remain unfalsifiable for now.