In his position as head of AT&T Ventures, Vikram Taneja leads the company enterprise capital arm of the telecommunications large, managing the company’s portfolio throughout direct fairness investments, warrants and limited-partner fund positions.
His funding mandate primarily focuses on early-stage know-how corporations from seed to Collection B that align with or affect the worldwide telecommunications, community infrastructure and enterprise software program sectors.
Below his management, AT&T Ventures targets investments in software program, {hardware} and infrastructure sectors the place AT&T’s community scale and inside engineering assets present a definite industrial or technical diligence benefit. Portfolio corporations embody enterprise and deep-tech companies akin to Databricks, Apptronik, Cyera, Carbyne, Aira and AST SpaceMobile.
Previous to his present 12-year stint directing AT&T Ventures, Taneja spent greater than 20 years working throughout company growth, enterprise lending and funding banking. He beforehand managed M&A and strategic funding actions for WarnerMedia throughout AT&T’s possession.
Taneja additionally served as a director at Orix Ventures, the place he targeted on growth-capital debt and fairness investments in mid- to late-stage know-how companies, in addition to holding company finance and funding banking roles at J.P. Morgan and PricewaterhouseCoopers.
In an electronic mail interview with Crunchbase Information, Taneja shares why he believes that whereas AI has drastically lowered the barrier to constructing software program, it has additionally shifted the definition of seed-stage technical danger.
The brand new dynamics, in his view, provides AT&T Ventures a chance to distinguish itself by providing fast, real-world technical validation and community integration somewhat than simply capital.
The interview has been edited for brevity and readability.
Crunchbase Information: If startups are constructing totally functioning apps by the seed spherical utilizing AI, what does that imply for the standard definition of technical danger? Is tech danger useless at seed, or has it simply advanced into one thing else?
Vikram Taneja: The outdated definition of technical danger was “can they construct it?” Though not totally absent on the seed stage, I’d say it’s turning into much less related given the dramatically decrease barrier to constructing software program with AI instruments.
However what changed it’s really more durable to reply: “Is the tech defensible?” Not simply “does it work?” however “does it compound?”
Knowledge moats, proprietary coaching units, community results constructed into the structure — that’s the brand new measure of sturdiness.
In prior cycles, technical complexity alone created some pure safety. Because of this, the technical danger dialog has shifted to give attention to how an organization defends itself over the subsequent three to 4 years, particularly as frontier labs transfer down the stack into software layers and begin focusing on complete verticals.
Equally, the distribution query exhibits up a lot earlier. “How will you get this to market?” is more and more requested on the seed stage somewhat than later within the cycle.
We’re additionally seeing elevated competitors for buyers to safe bigger stakes at seed that they’d have beforehand pursued on the A spherical. That is driving buyers to be extra thorough on the seed stage, and founders need to be ready to fulfill larger expectations throughout the board.
When anybody can use AI instruments to spin up a working app in a weekend, product execution occurs quick, however moats could be extremely shallow. On the seed stage, how are you separating a really defensible platform from a fantastically executed wrapper?
Taneja: In early 2025, we noticed a wave of AI wrapper corporations constructed on high of frontier fashions like OpenAI‘s GPT, Anthropic’s Claude or LLaMA, and a whole lot of capital flowed into them. What’s modified is that frontier LLMs have now clearly began to take extra of a platform method — transferring into the appliance layers and starting to select off the low-hanging fruit.
Because of this defensibility turns into important in AI investing. No platforms are completely defensible, however on some stage, you must ask that query now on the seed stage.
We’re in search of platforms utilizing proprietary knowledge that may’t be replicated by AI, corporations which have embedded deep area experience — areas the place general-purpose AI nonetheless lacks business context — into their workflows, or extremely specialised ecosystems or area of interest markets that present one other layer of insulation in classes which might be too focused for frontier labs to pursue immediately.
Are you seeing a change within the precise headcount or make-up of seed groups? If AI handles the heavy lifting of the preliminary code, are these founders spending their seed capital on engineers, or are they shifting assets instantly to distribution and go-to-market?
Taneja: There’s nonetheless an engineering focus within the early stage, as there ought to be, however we’re more and more seeing product, gross sales, or partnership roles turning into wanted sooner than prior to now. And the reason being, as you said, that it’s simpler to construct a working prototype, or perhaps a production-ready software, so the main target in a short time turns to establishing trials with clients or exploring distribution paths to dial within the product options.
For strategic buyers like AT&T Ventures, the place we frequently do proof-of-concepts with potential portfolio corporations, that is very thrilling. We get an opportunity to work with corporations earlier of their formation, can get actual technical validation a lot sooner than in any other case, and may equally attempt to discover a path to collaborate extra shortly.
AT&T Ventures has historically performed closely within the Seed to Collection B area. If institutional VCs are dashing to seed to seize bigger stakes as a result of the tech is mature, how does that change the aggressive panorama for CVCs? Are you discovering your self competing immediately with conventional multistage funds sooner than earlier than?
Taneja: The make-up of seed rounds has undoubtedly modified. Multi-stage funds used to point out up at Collection A or B when there was sufficient traction to underwrite. Now they’re at seed as a result of, as we mentioned, the businesses are mature sufficient, and they’re looking for winners earlier within the cycle. So sure, we’re in the identical rooms as earlier than.
However I’d push again on the concept that we’re competing immediately.
A Tier 1 monetary VC’s seed test and an AT&T Ventures seed test are completely different devices. They’re providing capital, model, steering and sample recognition from backing tons of of corporations.
We’re providing one thing a monetary VC structurally doesn’t: our community groups working along with your product in a manufacturing surroundings, oftentimes earlier than we even write the test, for instance. That’s free diligence working in each instructions. We’re validating the corporate, however it’s additionally receiving a real-world sign from one of many world’s largest community operators.
For a seed-stage firm that’s already solved the constructing drawback and now wants distribution, that’s tangible worth and complementary to what monetary VC companies are offering. In order that aggressive strain has really sharpened our worth proposition. It forces us to convey extra than simply capital to the desk.
Traditionally, company companions wish to see enterprise readiness, safety compliance and scalability — issues a seed startup hardly ever has. If a seed startup has a totally functioning product however continues to be a two-person workforce, can an enterprise like AT&T really run a pilot with them, or does the company integration timeline turn out to be a bottleneck?
Taneja: It begins with strategic rationale. That has all the time been the entry level for us at AT&T Ventures, and that hasn’t modified. If that’s in place, then it doesn’t all the time require full enterprise readiness to begin a pilot. It may be a structured trial or a extremely focused engagement, relying on the corporate’s stage.
We have now various ongoing proof of ideas with portfolio corporations throughout areas akin to AI-RAN, related infrastructure and pc imaginative and prescient.
The secret’s readability upfront — readability on what the target of the engagement is and the way we measure success. As soon as that’s clear, even early-stage corporations could be built-in right into a studying or testing surroundings with out pointless delay. The aim is to make the AT&T relationship really feel like an accelerant to additional adoption.
If seed is the brand new Collection A by way of product maturity, are you seeing Collection A pricing bleed into the seed spherical? How are you disciplined about valuations when the product appears like a Collection A, however the firm infrastructure continues to be very early?
Taneja: Seed pricing certainly appears completely different than perhaps 4 or 5 years in the past. We’re routinely seeing seed offers priced within the low- to mid-single-digit-million vary at about $20 million to $25 million post-money. That is just about the place Collection A offers had been a couple of years in the past. But it surely’s not essentially unjustified — the make-up and traction of seed-stage corporations are a lot additional alongside than predecessor vintages as we’ve mentioned.
We keep disciplined by being specific about what we’re really underwriting. We’re not simply underwriting the monetary return on this spherical — we’re underwriting the strategic worth of the connection over a five- to 10-year horizon.
Does this firm make AT&T’s community extra clever? Does it open up a brand new buyer phase? Does it validate a thesis we’re constructing round? Are there industrial alternatives past our preliminary thesis? Whenever you body it that manner, it provides us an extended horizon to work with and gives a number of levers to tug.
And truthfully, that’s the place our engineering and product groups play a key position. They assist us decipher whether or not the product that appears like a Collection A is definitely constructed like one, or whether or not it’s an ideal demo sitting on a basis that hasn’t been stress-tested. That technical learn bolsters our conviction when making investments.
A practical AI app on the seed stage nonetheless requires huge infrastructure. Whenever you consider these early-stage corporations, how a lot does their underlying structure and the way they deal with knowledge processing or edge computing issue into your choice?
Taneja: Structure is a key a part of our diligence course of. The way in which we give it some thought actually is determined by the last word use case. Is it for inside use — i.e., a software that AT&T shall be working with in our environments — or is it one thing we’d be distributing or incorporating into some type of product providing?
If the previous, all elements of the structure shall be reviewed, and that is most definitely to happen all through trials and proof of ideas as we develop a technical understanding of the appliance or product. If it’s the latter, then we’re seemingly most serious about understanding how this product structure scales over time and what it means from a price, latency and infrastructure perspective. We like to see corporations embracing edge-related applied sciences, however that doesn’t preclude us from engaged on functions that use conventional knowledge processing strategies.
You’ve spoken earlier than about your curiosity in “bodily AI” and robotics (like Apptronik). The software program lifecycle is definitely compressed by generative AI, however {hardware} and bodily deployment take time. Does this “seed is the brand new Collection A” pattern apply to pure-play software program strictly, or are you seeing AI speed up bodily tech and IoT on the early stage too?
Taneja: Bodily AI is a sector we’ve been fairly a bit, significantly as a result of inference and decisioning in autonomous programs, robotics and related units create a really completely different kind of demand profile on networks.
The software program layer is clearly accelerating — issues like notion, management programs and decisioning are transferring sooner due to AI (the rounds present it!). That may in the end assist pave the best way for the adoption of bodily AI. Nevertheless, the bodily deployment cycle nonetheless takes time, so that you don’t see fairly the identical stage of time compression there.
What’s fascinating for us at AT&T is the intersection — how intelligence is transferring nearer to the sting and the way that adjustments the best way networks have to be architected to deal with these workloads.
Associated Crunchbase queries:
Associated studying:
Illustration: Dom Guzman
Keep updated with latest funding rounds, acquisitions, and extra with the
Crunchbase Day by day.
