Why Google, Why Now
Reflections from Google Cloud Next on why TBS is reactivating its Google partnership for the agentic AI era.
By Ivan Zhitiyanov, CEO of Telelink Business Services (TBS) — a European IT services firm headquartered in Sofia, Bulgaria, operating across seven countries.
I will start with a confession: I am not a fan of large industry events. Google Cloud Next is one of the biggest of them — tens of thousands of attendees, hundreds of sessions, the full Las Vegas treatment — which is exactly the kind of event I usually avoid. It took real motivation to fly 24 hours and sit through a 10-hour time difference for four nights. I did it anyway, and I want to be honest about why.
Why I went to Google Cloud Next
Partly because the distance — physical and temporal — is useful, a 10-hour time difference from Sofia means that for five straight days, my full attention is on one thing. In this case, Google.
I had never been to a Google event. I wanted to see the client base with my own eyes, feel the partnership mass, meet the leadership, and form a first-hand impression of where Google is heading as a company.
And I went with a thesis. I believed, going in, that Google was building a very strong AI portfolio. I believed they are becoming more and more a B2B company. I believed, as the challenger in the hyperscaler space, they would have a more open view of partners than the larger hyperscalers do. We work with the other hyperscalers every day, and I will be direct: most of them do not care about partners the way they claim to. I hoped Google was different. It seems I was right.
My job at the event was to learn. In the sessions, I am a student — I hear a term, a concept, a framework, and during the breaks, the nights, and the early mornings (wide awake at 4:30 a.m. thanks to the jet lag), I read, I research, I verify. Honestly, I learned more from my own reading than from the sessions themselves. But the sessions set the questions, and the reading filled in the answers. Both were necessary.
Cloud and on-premises in Europe — where we actually are
Before I say anything about Google, I want to correct a common misconception about how on-premises and cloud are really playing out in our markets.
Eastern Europe has always preferred on-premises infrastructure. That is not changing, and I do not expect it to. What is more interesting — and what we are increasingly seeing and hearing in our Western European conversations — is that on-prem and hybrid are no longer the unfashionable answer there either. Clients in Western Europe are taking a harder look at how much of their estate genuinely belongs in the public cloud, and which parts make more sense closer to home. Anyone selling a “pure cloud future” story right now is not paying attention.
But here is the honest truth for us, and for the AI Consulting Practice we are building: building infrastructure for AI is expensive — for now. The costs are coming down, and I expect the trend to continue, but we are still very far from affordable when it comes to standing up a proof-of-concept. And hardware is only part of the problem. If you want to run an AI stack on-premises in an open-source environment, you also need to solve governance, scalability, and optimization — which is not a weekend project. It is weeks of work, real money, and serious engineering before you even have a working foundation.
Which means moving from concept to a real, usable AI environment on-premises is not easy. It takes time, capital, and risk tolerance that most clients — reasonably — are not willing to commit before they know the use case works.
That is why hyperscalers matter for us right now.
There is also the blunt commercial reality. Memory is expensive, GPU capacity is scarce, and nobody negotiates better prices with the silicon vendors than the hyperscalers themselves. Google negotiates better than TBS can. Let us not pretend otherwise.
So we need hyperscalers — Google in particular, but also Microsoft — to enable AI projects at the proof stage.
Why we start AI projects in the cloud
Our approach to AI consulting is straightforward and worth describing in plain terms.
We help clients prove value before they commit to infrastructure. That means starting with a small workload in the cloud — fast to deploy, cheap to validate, easy to stop if it does not work. The hyperscaler is doing exactly what it is good at: absorbing the upfront cost, the procurement complexity, and the risk that the use case does not pan out.
Once the value is proven, the client has options.
For some clients, growing the workload inside the cloud will continue to make sense — economically, operationally, and from a risk perspective. We are perfectly happy for them to stay there.
For other clients, the math will tip toward on-premises or hybrid as the workload matures and the ROI becomes clear. Particularly in our region, where sovereignty, control, and TCO discipline matter, on-prem can become the better answer once the use case is proven and the volume is real. The honest version of this is that some workloads will need refactoring on the way back, especially if they were architected casually inside the cloud — but with Google Distributed Cloud, that movement can be made considerably cleaner.
So the role of the hyperscaler, for us, is to open the door — and then to stay or not, depending on what each client actually needs. That is why it matters that the hyperscaler we choose supports this movement cleanly: cloud for the proof, a real sovereignty story for those who need it, and a credible bridge between the two.
The lock-in risk is real, and I take it seriously. But avoiding lock-in while still capturing the speed and economics of the hyperscaler is exactly what a system integrator is for. I believe we can strike the right balance between sovereignty and functionality. That is our job.
What I was looking for on Google
I went to Las Vegas with a list of criteria. A hyperscaler we commit to for our AI practice needs to bring:
- A genuine agentic AI platform — not demoware. Works with Gemini, but also with frontier and open-source models, because we do not lock clients into a single model ecosystem.
- Security embedded end-to-end, not bolted on.
- A real pay-as-you-grow consumption model — and I mean a real one. Not just pay for the VM while it runs, but pay for the agent only when the agent actually works. That is a meaningfully different economic model, and it strengthens the proof-stage story significantly. Google’s serverless compute (Cloud Run) and their Agent Engine, both part of the Gemini Enterprise platform, are built around this principle.
- IaaS and PaaS depth underneath the agents, so when a use case needs custom infrastructure, we can build around the agentic layer instead of being constrained by it.
- Sovereign options across the full ladder. I was not skeptical about sovereignty going in — I already knew what Google had been building. The week reinforced my existing belief: a workload can start in Google Cloud, be proven there, and then grow into confidential computing or fully on-premises through Google Distributed Cloud when the scale, regulation, or economics demand it. That is the movement our clients actually want.
- Openness — multi-cloud, multi-model, multi-tool. Because our clients already run M365, already have Azure investments, and already use specialized tools across dozens of categories. Our job is to integrate those investments, not replace them.
Seven criteria. Google meets all of them, and — more importantly — they meet them as a single coherent architecture rather than as seven separate products.
Technical observations from Google Cloud Next
A few specific things from the week are worth calling out, because they will matter to anyone in our industry thinking seriously about what is possible now.
The Gemini Enterprise agent platform
This is the core of Google’s B2B AI story, and it is a genuinely complete platform rather than a collection of features. Security, identity, governance, and visibility are built in, and the visibility works for both the engineers building the agents and the users interacting with them. Add A2A (their open inter-agent protocol) and ADK (a multi-language, multi-model SDK) and you have real connective tissue for enterprise AI regardless of which cloud, model, or framework an organization chooses. For a system integrator, this is the first platform I have seen that feels ready to be deployed into a serious enterprise client without us having to write the missing governance layers ourselves.
Looker — the surprise of the week
I went in with modest expectations for Looker. I came out thinking it is one of the more interesting pieces of the entire portfolio.
Two things changed my view. First, the semantic abstraction layer — a curated, governed representation of business data where definitions like “revenue,” “customer,” or “churn” are standardized once and reused everywhere, instead of being redefined inconsistently in every dashboard and query. Second, the ability to reason across up to five domains (datasets) in a single data agent, which means a conversational AI can answer real business questions that cross multiple areas — finance, sales, operations — instead of being trapped inside a single dataset.
The combination is powerful. In my opinion, this is the first version of conversational analytics that actually works for non-data-experts, because the agent reasons over curated business definitions rather than raw SQL. Self-service analytics has been promised for two decades and has consistently disappointed. This might be the version that does not.
Looker is not cheap, and in clients with strong incumbent BI tooling, the adoption argument is not automatic. But as a semantic foundation for AI agents, rather than as a traditional BI tool, it opens conversations that were not previously possible.
The security portfolio
Google has assembled a genuinely impressive security portfolio. It now sits as one of the most complete code-to-cloud security stories in the market, combining Mandiant’s threat intelligence (already arguably the industry benchmark), Google Security Operations for SIEM and SOAR on a single data foundation, Security Command Center for GCP-native posture, the recent Wiz acquisition for cloud security posture management across multiple clouds, and the new agentic security agents announced at the event for continuous threat hunting, triage, and environment auditing. That last piece — agentic security — is the clearest signal that Google is building its security layer for the AI era rather than retrofitting legacy tooling onto it.
For us, the most immediate interest is in AI security, and specifically in the security primitives that come built into the agent platform. When we deliver a Gemini Enterprise solution, the security layer is part of the package rather than an afterthought, and that matters. As our cloud business grows over time, the broader security stack will become more relevant — but for our current AI consulting work, it is the agent platform’s native security that we lean on first.
Google Distributed Cloud
The edge and on-premises story is finally real. Google Distributed Cloud is a working product — not a concept — and it is deployable at a footprint that is sized for realistic clients in our region, not only for the largest Western European institutions. Combined with Customer-Managed Encryption Keys, External Key Management, and Confidential Computing, Google now offers a complete sovereignty ladder where each tier corresponds to a different regulatory requirement. That means we can have a consultative sovereignty conversation with clients rather than a binary one.
Google’s structural position in the AI market
One observation I want to share separately, because it underlies everything else.
In my view, four ingredients matter most for a serious AI partner today: a frontier foundation model, a global hyperscaler, your own AI chip, and a meaningful application footprint to deploy AI into real workflows. Underneath all of it, you also need the capital to keep investing in all four at the same time.
Google has all four. They own Gemini. They own Google Cloud. They own the TPU. They own Workspace, which is increasingly an AI delivery surface in its own right. And they have the balance sheet to keep building. On top of that, they are a major investor in Anthropic, which gives them a strategic relationship with one of the most important AI labs in the world.
Among Google’s direct competitors, only Amazon has even two of the four ingredients — hyperscaler and chip. No one else has the full combination.
The chip piece matters more than most people realise. Owning the silicon means Google can produce tokens at a lower marginal cost than any of their competitors. That cost advantage flows directly into the pay-as-you-grow economics that make hyperscalers attractive at the proof stage in the first place. It is not a small thing. It is structural.
When we choose a partner for the next decade of AI consulting, this is the kind of advantage that matters. We are not betting on a single product or a single quarter. We are betting on the company in this market with the most complete vertical integration — model, silicon, cloud, applications, and capital all under one roof. That kind of integration tends to compound over time.
Reactivating the TBS–Google partnership
TBS has been a Google partner for some time, but the relationship has not been a priority on either side. I want to be honest about that.
Four years ago, TBS was a different company — primarily an infrastructure integrator, not yet serious about an AI practice. Google Cloud was a different company too — strong on IaaS and Workspace, but not yet the comprehensive enterprise platform it has become. Both of us have changed. The fit that was not there before is there now.
The regional team in East Europe is genuinely strong — business-oriented, commercially realistic, and in my personal experience, people I have known for years. My belief is that they have the right motivation and the right acumen, and if we show up as an active, committed partner, they will support us as appropriate. That is how partnerships should work, and I am optimistic about what we can build together.
And just as importantly: this does not mean we are becoming a Google-only partner. We will continue our partnership with Microsoft where it makes sense for us and for our clients, and we are actively exploring new partnerships in other parts of the ecosystem. Our job is not to pick a single side; it is to bring the right combination to each client. What this week has clarified is that Google now earns a much larger place in that combination than it did before.
What this means
I will keep it simple.
Google has built products we believe in. We have clients with needs those products can meet. Google is partner-oriented in a way that makes the economics work for a system integrator like us. And the structural position they hold in the AI market is, in my view, the strongest of any single vendor.
So we are going to double down on building capabilities, demos, and proof-of-concepts with Google. Our engineers will be learning the platform. Our practice leads will be investing in certifications. Our lab will have Google Distributed Cloud running in it. And our clients will be seeing more of Google in our recommendations and our solutions — because we believe it makes sense.
That is the short version. The long version plays out over the next twelve to eighteen months. More to come, and soon.
🤘
A few honest caveats
I am sure some of these arguments will draw pushback. Two in particular are worth acknowledging openly.
On Microsoft’s place in the four-ingredients picture
Some readers will fairly object that Microsoft has more than I gave them credit for — Azure as a hyperscaler, M365 as a major applications layer, Maia as their own chip, and a deep partnership with OpenAI as their model. By a generous reading, that is also four out of four.
By a stricter reading, two of those four are weaker than they appear. An OpenAI partnership is not the same as owning a frontier model — it is a relationship with a separate company that has its own commercial agenda, and that agenda has been visibly unstable in the recent past. Maia is real silicon, but it is years behind TPU and Trainium at the scale that matters for partner economics.
I land on the stricter reading. But I will acknowledge that the generous reading is defensible. Either way, we continue to partner with Microsoft and take their AI direction seriously. They are not where Google is today, but they are not standing still either.
On NVIDIA and vertical integration
Another reasonable reader will point out that NVIDIA has its own form of vertical integration — silicon, CUDA, DGX Cloud, and growing investments across the model layer. That is also true.
The reason I do not put NVIDIA in the same comparison is that NVIDIA is not a hyperscaler with a customer-facing applications layer for enterprises. They are an extraordinary company at a different point in the AI value chain. The four-ingredients argument I am making is specifically about the hyperscaler-as-AI-partner decision — and within that decision, Google’s combination is what stands out today.
Neither caveat changes the core argument. They just make sure I am making it with my eyes open.



