Alphabet’s priorities snapped into focus on its latest earnings call. CEO Sundar Pichai didn’t mince words. The company places artificial general intelligence research first when parceling out its custom silicon. Everything else follows.
That revelation, delivered July 23 during the second-quarter results briefing, underscores a calculated gamble. Google isn’t just building bigger models. It’s hoarding the specialized hardware needed to chase human-level machine intelligence while its cloud business explodes. The approach highlights tensions rippling across the industry. Supply shortages bite hard. Competitors scramble. And the stakes keep climbing.
“In terms of allocating our TPUs … our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development,” Pichai said, according to The Register. “That is the foundation for everything we do.”
The statement landed as Alphabet posted robust numbers. Cloud revenue jumped 82 percent year over year to $24.75 billion. Operating profit there soared 214 percent to $8.8 billion. A backlog of $514 billion in deferred customer commitments signals sustained hunger for AI infrastructure. Search and YouTube ads grew too. Yet free cash flow turned negative for the first time since 2004. Capital spending forecasts rose again.
CFO Anat Ashkenazi pointed to constraints. The company now expects to spend $195 billion to $205 billion this year. That’s up from an earlier projection of $180 billion to $190 billion. “The supply-constrained environment” limits what Google can acquire, she noted. So the firm turns to third-party capacity as a temporary fix. It buys time while internal production scales.
Pichai elaborated on the balancing act. Analysts pressed him on how the company splits resources among research, search, YouTube, and cloud customers. “On allocation, I think the baseline with which we start is what it takes to continue AGI development at the frontier,” he replied. Core products come next. Then cloud workloads, especially those powering Vertex AI, Gemini models for enterprises, analytics, and security tools.
But. The message was unmistakable. AGI comes first. That focus traces back years. Google developed tensor processing units specifically for machine learning workloads. These chips deliver strong performance per watt compared with graphics processors from Nvidia. They power much of the company’s training and inference internally. And now, select customers can buy access too.
The decision to sell some TPUs earlier this year sparked the questions that drew out Pichai’s comments. Goldman Sachs’ Eric Sheridan and Bernstein’s Mark Shmulik both probed the trade-offs. Google’s answer? Protect the moonshot. Serve paying cloud users where possible. Accept short-term pain for long-term positioning.
This isn’t happening in isolation. The entire sector wrestles with similar limits. Reports from earlier this year describe persistent shortages of high-end AI accelerators. Packaging capacity at TSMC remains booked solid. High-bandwidth memory supplies from SK Hynix and others lag demand. Spheron Network’s analysis from April flagged constraints stretching into late 2026 and beyond. Forward orders from hyperscalers have locked up much of Nvidia’s output.
Even Google has felt the pinch from the other side. In March it reportedly told Meta it could not supply all the Gemini inference capacity the social media giant sought. That disclosure, covered by the Financial Times via Yahoo Finance, disrupted some of Meta’s internal projects. It forced prioritization. The episode reveals how tight things have grown. Demand outruns even the largest players’ ability to provision.
Nvidia’s Jensen Huang has acknowledged the strain while projecting growth. “We’ve secured supply for very robust growth of all of those systems,” he said in June, per Reuters. “We have supply for very, very robust growth, but we’re still supply constrained.” His comments came as the company unveiled new chips aimed at PCs and data centers alike.
Google’s own DeepMind unit has published on the path ahead. A June report titled “From AGI to ASI” explores what might follow human-level systems. It discusses scaling, paradigm changes, recursive self-improvement, and multi-agent collectives. The paper frames superintelligence as a plausible extension once AGI arrives. Such thinking animates the resource allocation Pichai described.
Demis Hassabis, DeepMind’s CEO, offered a timeline in a Stanford talk. “Maybe 2030, plus or minus a year, which is astounding to think, really,” he said in June, as reported by Business Insider. “I think that will be such an enormous transformative technology; it’s gonna effectively be a new human era.” His remarks echo the urgency felt inside Google.
The company has also worked on frameworks to measure progress. A March post on its blog detailed a cognitive taxonomy for evaluating AI capabilities. The goal? Ground AGI benchmarks in scientific principles drawn from cognitive science. That effort, while academic on the surface, supports the serious engineering bets behind the scenes.
Yet questions linger about execution. Negative free cash flow surprised some investors. Shares slipped in after-hours trading. The capital intensity of chasing frontier models collides with near-term financial optics. Google Cloud’s growth helps. So does AI-enhanced search, which Pichai said lifts query volumes and cuts per-response costs through optimizations.
Still, the hoarding strategy carries risks. Cloud customers expect reliable access. Large deals sometimes require Google to absorb high short-term costs for third-party capacity. Pichai framed those choices as positive over multi-year horizons. “The incremental opportunities they are bringing to us, while a short-term cost over a few months may be very high, in the lifetime of the deal, as we bring more capacity on, is highly ROI positive,” he explained.
Industry observers note the broader pattern. Hyperscalers stockpile chips, reserve future production, and build massive data centers. Power availability, land, and cooling present additional bottlenecks. A Bloomberg discussion on AI buildout constraints from January highlighted land, power, and supply chain as triple threats. Those pressures show no quick relief.
Google’s approach stands out because its hardware is proprietary. TPUs give it an edge in efficiency for certain workloads. By reserving the majority for internal AGI work, the company signals confidence that breakthroughs there will pay off across products. Search. Ads. Cloud offerings. Even hardware sales down the road.
Competitors pursue parallel paths. OpenAI, Anthropic, Meta, and Microsoft pour resources into their own models and infrastructure. All face the same chip scarcity. Some turn to custom silicon. Others double down on Nvidia GPUs. The arms race intensifies.
And the timeline compresses. What once seemed distant now appears measurable in years. Pichai’s comments make clear Google won’t cede the lead. It will ration its most valuable compute resource to protect that position. The rest of the business must adapt around it.
That calculation defines this moment in AI development. Massive spending. Scarce components. Corporate strategy built on the bet that intelligence at scale unlocks outsized returns. Whether the bet pays off remains unseen. But the resource flows tell their own story. Google is all in.
WebProNews is an iEntry Publication