Gemini 4 Hasn't Landed — But Alphabet Just Became the Only Unblocked Frontier Model Left

The July delay narrative has quietly been reclassified. Alphabet originally told investors at Google I/O in May 2026 that Gemini 3.5 Pro would ship in June. A Jul 16 Bloomberg scoop found the model 'months behind schedule' — failing on internal coding benchmarks — and Alphabet shares fell 4.4% that day, erasing roughly $200 billion in market capitalization. Two weeks later, on Jul 21, Alphabet disclosed that the 'most ambitious pre-training run' for the next model had begun. That next model is now called Gemini 4, not Gemini 3.5 Pro — a version skip that suggests the post-training results forced an architectural reset rather than a simple continuation.

On Sep 23, new DeepMind leader Koray Kavukcuoglu told The Information's AI Agenda Live Summit that Gemini 4 is in 'refinement' and that the company would release an 'early post-training output' 'as soon as possible.' He did not share a specific date. 9to5Google confirmed the same line the next day. The 'as soon as possible' framing — combined with the lack of a date — is the second under-appreciated shift: it signals Alphabet is choosing iteration speed over calendar precision, the opposite of the Microsoft-style 'ship-by-quarter' cadence.

Cloud is the only line item carrying the thesis. Google Cloud's 82% YoY growth to $24.8B in Q2 materially beat the 64% consensus and validated the TPU-plus-Broadcom strategy. But the multiple compression risk is real: with Alphabet now trading at 76.8x trailing free cash flow — up materially from a year ago — every quarter of negative FCF argues for a richer enterprise revenue story, not a thinner one. Microsoft's FY2027 capex guidance of $255–260B (vs. ~$190B in FY26) shows the same pattern across the Mag-7: the spend is no longer the differentiator. What gets built with it is.

The real alpha in the Gemini 4 story is not benchmark scores — it is access. On Sep 1, 2026, OpenAI disclosed that its upcoming Astra model is the first system to cross the 'Critical' cybersecurity capability threshold; advanced cyber capabilities are now restricted to its 'Daybreak' trusted-access program, with the U.S. government as an early participant. In the same window, Anthropic's Mythos 5.1 has been limited to vetted institutions due to cyber and bioscience capabilities — there is no general release. The Trump administration has begun dictating release access to frontier AI, per a Jul 17 CNBC report covering Anthropic and OpenAI. The market is shifting from 'who has the best model' to 'who is allowed to sell you one.'

This is the under-appreciated re-rating catalyst. If Alphabet ships an ungated frontier-tier model into a market where the two strongest competitors are access-restricted, the addressable buyer base shifts: enterprises that need AI sovereignty — banks, healthcare, defense primes, EU regulated industries — have fewer viable suppliers. The Q2 Cloud beat (+82% YoY) is consistent with Alphabet already capturing this share ahead of Gemini 4's release; the Cloud acceleration did not need Gemini 4 to start.

The Nasdaq-100 quarterly rebalance on Sep 21, 2026 has already taken effect — but the annual reconstitution, expected in October, is the date investors should anchor on. With GOOGL absorbing a 4.4% drawdown on the Jul 16 delay news and a further 6% post-Q2 drop on Jul 23 over capex concerns, Alphabet's NDX weight has been compressed relative to peers during a period when the top-7 concentration is at multi-year highs. Apple, Microsoft, NVIDIA, Alphabet, Amazon, Meta, and Tesla collectively still command a disproportionate share of NDX weight — and any weight redistribution at reconstitution mechanically drives QQQ / QQQM flows.

Alphabet's $200B does not stay at Alphabet. It flows through Broadcom (custom TPU silicon, long-term deal through 2031), NVIDIA (GPUs for training clusters and inference at scale), and TSMC (advanced packaging and leading-edge nodes for both). The most consequential downstream signal of2026 is the Apr 6 expansion of the Anthropic / Google / Broadcom compute deal — a commitment to multiple gigawatts of next-generation TPU capacity coming online starting in 2027, on top of the existing 1GW already expected in 2026.

Short-term (days to quarters): Three catalysts cluster in the October–November window. First, the Nasdaq-100 annual reconstitution locks in NDX weights that ETF flows will chase through year-end. Second, Alphabet's Q3 earnings (expected late October) will reveal whether Cloud growth stays at >70% YoY and whether capex guidance moves into the $220–240B range for2027. Third, the actual Gemini 4 release date — likely late October or early November, per the Sep 23 'as soon as possible' framing — will resolve the gating question: does the model land below the threshold that triggered Astra/Mythos restrictions, or above it?

Long-term (1–3 years): The structural question is whether Alphabet's TPU business — now externally monetized via Anthropic (3.5GW), Meta, and Apple — becomes a standalone profit pool worth a re-rating. If Broadcom's external-TPU franchise scales to $30–50B by 2028, Alphabet's capex becomes an investment in a moated silicon platform, not a NVIDIA hedge. The risk: a Microsoft-led effort to consolidate NVIDIA-only supply chains at the enterprise layer, and a Trump-era export control that fragments the international TPU market. Both are visible tail risks in 2027.