Alphabet just became a semiconductor vendor—but its own CEO admitted customers are second priority. Here’s why that should worry investors, excite competitors, and change how we think about AI infrastructure.
Table of Contents
The Headline Everyone Got Wrong
“Google begins selling TPU chips.” Sounds like a straightforward business story, right? Alphabet enters the AI chip market, challenges Nvidia, diversifies revenue. Another tech giant becoming more like its peers.
But read between the lines of Sundar Pichai’s own words, and a very different picture emerges—one that has nothing to do with competing with Nvidia and everything to do with a high-stakes gamble that could reshape the entire AI industry.
Here’s what Pichai actually said when asked how Google allocates its limited TPU supply:
“Our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development.”
Let that sink in. Customers come after AGI.
Not alongside. After.
The Semiconductor Vendor That Doesn’t Want to Sell Chips
Let’s be clear about what just happened. Alphabet:
- Started recognizing hardware revenue for the first time
- Added TPU sales to its $514 billion backlog
- Raised capex guidance to $195–205 billion for 2026
- Saw its free cash flow turn negative (-$5.9 billion) for the first time ever
And yet, the CEO is openly saying that if there’s a conflict between selling a chip to a customer and using it internally for AGI research, the customer loses.
This isn’t a typical semiconductor vendor playbook. Nvidia sells every GPU it can make. AMD and Intel fight for market share. Even Amazon, with its Trainium chips, positions them as a cloud service first and foremost.
Google is doing something unprecedented: selling chips it doesn’t actually want to sell, to customers it considers secondary, because it needs the revenue to fund a bet that may never pay off.
That’s not a hardware strategy. That’s a funding strategy disguised as a product launch.
The CoWoS Bottleneck: Why Google Had No Choice
To understand why this matters, you need to understand the supply chain reality.
Google’s TPUs are fabricated by TSMC, the same company that makes Nvidia’s GPUs. Both rely on CoWoS advanced packaging—a manufacturing process so constrained that Google had to slash its 2026 production target from 4 million to 3 million units.
Here’s the brutal math:
Table
| Resource | Demand | Supply | Gap |
|---|---|---|---|
| TSMC CoWoS capacity | ~4M Google + Millions of Nvidia + Others | Limited | Severe shortage |
| Google’s internal needs | AGI research + Search + YouTube + Cloud | 3M units | Not enough |
| External customer demand | Enterprise AI, startups, governments | Growing rapidly | Unmet |
Google is supply-constrained at the exact moment when AI infrastructure demand is exploding. It can’t make enough chips for itself, let alone customers.
So why sell any at all?
Because negative free cash flow of $5.9 billion is unsustainable, even for Alphabet. The TPU sales aren’t about market expansion—they’re about bridging the gap between AGI ambition and financial reality.
CFO Anat Ashkenazi admitted as much: Google will use “third-party capacity in Q3 as a bridging strategy while we build out more internal capacity.”
“Bridging strategy” is corporate speak for “we’re buying time.”
The AGI Bet: What Happens If Google Is Wrong?
Here’s where this gets genuinely risky.
Google is prioritizing AGI development over everything else. But here’s the uncomfortable truth: nobody knows if AGI is achievable, when it might arrive, or what it will cost.
Consider the scenarios:
Scenario A: AGI arrives in 2–3 years
- Google’s bet pays off massively
- It owns the most advanced AI systems
- Competitors who sold chips instead of building AGI are left behind
- TPU sales become irrelevant because Google keeps everything internal
Scenario B: AGI remains elusive for 5–10+ years
- Google burns through hundreds of billions in capex
- Negative free cash flow becomes a structural problem
- Customers who were deprioritized move to Nvidia, AMD, or custom silicon
- Cloud market share erodes just as AI infrastructure spending peaks
Scenario C: AGI is achieved by someone else first
- OpenAI, Anthropic, or a Chinese lab cracks it first
- Google’s chip allocation strategy looks like a catastrophic miscalculation
- The $514 billion backlog becomes a liability if customers defect
Pichai is essentially saying: “We’re going all-in on a technology that may not exist yet, and if you’re a customer, you get what’s left over.”
That’s not confidence. That’s desperation dressed as vision.
What This Means for the AI Chip Market
Google’s entry as a TPU vendor isn’t the Nvidia killer some headlines suggest. In fact, it might be the best thing that ever happened to Nvidia.
Here’s why:
1. Google just validated the “AI chip shortage” narrative If even Google—with its own fabs partnerships and vertical integration—can’t meet demand, what chance do smaller players have? This reinforces Nvidia’s pricing power and market position.
2. Google’s “customers second” message is a gift to competitors Every enterprise evaluating AI infrastructure just heard the CEO of Google say they’re not the top priority. That pushes buyers toward Nvidia (reliable supply), AWS (Tranium/Inferentia as a service), or Microsoft (Azure with OpenAI partnership).
3. The TPU is a product without a clear market fit Nvidia’s GPUs are general-purpose AI accelerators. Google’s TPUs are optimized specifically for Google’s own AI workloads. Selling them to external customers means either:
- Customizing for diverse workloads (expensive, complex)
- Limiting sales to customers with Google-like needs (small market)
- Accepting that TPUs underperform vs. GPUs on non-Google models (bad optics)
None of these are winning strategies.
4. Negative free cash flow is a flashing red light Alphabet has never had negative free cash flow before. In Q2 2026, it hit -$5.9 billion. For context, that’s worse than Amazon’s infamous “profitless growth” years. The market tolerated Amazon because it was gaining market share. Google is losing money while telling customers they’re not the priority.
The Real Story: Google Is Building a Moat, Not a Business
Strip away the hardware revenue headlines, and here’s what’s actually happening:
Google isn’t trying to become the next Nvidia. It’s trying to ensure that if AGI is real, Google builds it first. The TPU sales are a side effect of needing cash to fund that mission.
This is a capital allocation story, not a product launch story.
Consider the numbers:
- $195–205 billion in 2026 capex (up from $180–190 billion last quarter)
- $514 billion backlog (up $50 billion in one quarter)
- 82% YoY growth in Cloud revenue ($24.8 billion)
- Negative $5.9 billion free cash flow
Google is spending like a startup chasing product-market fit, not like a mature tech giant optimizing for margins. The difference? Startups have nothing to lose. Google has a $2 trillion market cap and a reputation as the most disciplined big tech company.
That discipline is gone. In its place is a moonshot mentality applied to the core business.
Three Takeaways for Different Audiences
If you’re an investor: This is a binary bet. Either Google achieves AGI first and justifies the spending, or it doesn’t and faces a painful reckoning. The TPU sales are a distraction from the real question: Is Alphabet’s AGI timeline realistic?
If you’re an enterprise buyer: Google just told you that you’re not their priority. That doesn’t mean TPUs are bad—it means you should diversify. Don’t build your AI infrastructure strategy around a vendor that admits it’s holding back its best supply for internal use.
If you’re in the AI industry: Google’s move validates the “infrastructure bottleneck” thesis. The real constraint isn’t models or data—it’s compute capacity. Whoever solves the chip supply problem (TSMC expanding CoWoS, Intel catching up, new packaging tech) will capture enormous value. The model race is secondary to the silicon race.
The Bottom Line
Google selling TPUs isn’t a hardware strategy. It’s a financial bridge to fund an AGI bet that may or may not pay off. Pichai’s candor about prioritizing internal research over customers is refreshing—but it’s also a warning sign.
The semiconductor industry has seen this movie before. Companies that prioritize internal dreams over customer needs usually end up as cautionary tales, not market leaders.
Google has earned the benefit of the doubt through decades of successful moonshots. But negative free cash flow, supply constraints, and a “customers second” philosophy are a dangerous combination—especially when the prize (AGI) might be a mirage.
The TPU sales aren’t the story. The desperation behind them is.
What’s your take? Is Google making a brilliant long-term bet, or burning cash on an AGI fantasy? Let me know in the comments.
Recommended for you: