Key points
- Alphabet (GOOGL) shares rose about 3% Monday after The Information reported a new AI chip project called Frozen v2.
- Frozen v2 would embed Gemini's architecture directly into the chip, targeting 6 to 10 times the efficiency of Google's current TPUs.
- Google is targeting deployment as early as 2028, as an addition to its TPU lineup, not a replacement for it.
- Alphabet reports Q2 2026 earnings Wednesday, July 22, with Google Cloud growth the number that matters most.
Alphabet (GOOGL) shares opened higher Monday, touched $359.67 intraday, and were still up about 3% by midmorning, changing hands near $357. The move followed a report from The Information that Google is developing a new kind of AI chip, one built specifically around its own Gemini models rather than as a general-purpose accelerator. Bloomberg, CNBC and others picked the story up through the morning. Google has not confirmed any of it. The stock moved anyway, and by Wednesday, Alphabet has to back that reaction up with an actual earnings report.
A chip with Gemini built into the hardware
The project is reportedly called Frozen v2. According to The Information, Google engineers are working to embed parts of Gemini's own architecture directly into the chip's silicon, instead of running the model as software on general-purpose AI hardware. The goal is fewer calculation steps and less data movement inside the server for every prompt the model answers.
Google's engineers reportedly project the chip could serve six to ten times more tokens per unit of power than the company's current Tensor Processing Units, the custom AI chips Google already runs at scale across its own products and Google Cloud. That would make Frozen v2 a specialized branch of Google's chip lineup rather than a TPU replacement. Google can still update the model's behavior by shipping new weights, so the hardware would not go stale every time Gemini gets retrained.
The tradeoff is real. Locking a model's architecture into physical silicon ties that hardware to Gemini's current design in a way general-purpose TPUs are not tied. If Google's AI team wants a meaningfully different architecture two years from now, Frozen v2 will not bend the way general-purpose chips can.
Why Google wants this badly enough to try it
The timing points to a specific problem. Google has reportedly been rationing its own compute internally, tight enough that Google Cloud has had to turn away outside business it otherwise would have taken. Sundar Pichai told analysts on Alphabet's Q1 call that the company was compute constrained and would have booked more cloud revenue with more capacity. A chip six to ten times more efficient per token would ease exactly that bottleneck, freeing up power and floor space that would otherwise go toward buying and running more general-purpose hardware.
Google is not the only one making this bet. Anthropic is reportedly in early talks with Samsung to build its own custom AI chip, following a similar move by OpenAI, and Amazon and Meta already run their own silicon in Trainium and MTIA. Renting Nvidia's chips at scale gets expensive fast, and owning hardware tuned to your own models is one of the only ways to bring that cost down.
The market's answer, and Google's own caveat
Google's response to the report didn't say much either way: "not every lab project reaches commercial production." Frozen v2 is reportedly targeted for 2028, two years out, plenty of time for a project like this to get shelved before it ever ships.
That did not stop the stock from moving. A roughly 3% gain on a company Alphabet's size is a large single-day swing for an unconfirmed report on a chip nobody has seen, with no guarantee it ever ships. Investors are betting Google's compute shortage has a real fix coming. The AI trade has been nervous about that exact question for weeks.
Wednesday's earnings are the actual test
Alphabet reports second-quarter results Wednesday, July 22, after the close, with a call scheduled for 4:30pm Eastern. Wall Street is looking for roughly $2.87 in earnings per share and revenue near $116.8 billion. The number that will matter more than either of those is Google Cloud's growth rate. Cloud revenue grew 63% in the first quarter to just over $20 billion, with a backlog that swelled past $460 billion, and at least one analyst has since pushed a growth estimate as high as 70% for this report.
We covered the full earnings calendar for the week, which also includes Tesla (TSLA) and IBM Wednesday, Intel (INTC) and Nokia (NOK) Thursday, and a lineup that touches nearly every corner of the market. But Alphabet's own report carries extra weight this week because of Monday's chip news specifically. If Google Cloud keeps growing near 60% while capital spending guidance holds, that is a company credibly racing to fix its own compute shortage on more than one front at once. If Cloud growth slows or spending guidance turns cautious, Monday's 3% gain on an unconfirmed report two years from shipping will look like exactly what skeptics will say it always was.
What to watch
Alphabet has guided full-year 2026 capital expenditures as high as $190 billion, part of the roughly $725 billion big tech is spending on AI data centers this year. Frozen v2, if it is real and if it ships on time, is a bet that some of that spending curve can eventually bend down per unit of AI output delivered. Nvidia's own next-generation rack was reported delayed to 2028 as well, which makes this the second major "2028" chip story to land in two weeks. Whether either one actually ships on that timeline is a question that will not get answered for years. What gets answered Wednesday is much narrower: whether Alphabet's current business backs up the excitement investors showed on Monday.
Sources
- Bloomberg: Google plans new chip to boost AI efficiency, Information says
- CNBC: Alphabet stock pops on report it's developing a more efficient AI chip
- TechCrunch: Anthropic is discussing a new custom chip with Samsung
- CNBC: Alphabet ups 2026 capex to as much as $190 billion
- Earnings dates, consensus EPS, and revenue estimates via brokerage earnings-calendar data, July 20, 2026
- Stock prices and market cap via live market data, July 20, 2026
Stock prices are as of midmorning trading, July 20, 2026, and may have moved further by market close. This is general market commentary and not investment advice. Always do your own research and consider speaking with a licensed financial professional before making any investment decision.
