Intel Officially Closes $2B Acquisition of Israeli AI Startup Habana Labs, Strengthening Position Against Nvidia
The acquisition of AI processor maker Habana Labs will bolster Intel's portfolio in the data center and deep learning segments. The AI chip market is expected to exceed $25 billion by 2024, and Intel is ramping up its specialized product line to compete with the dominant Nvidia.
The Deceased Nervana: Why Intel Paid $2B for an Israeli Startup That Will Only Bring It Trouble
Author: Independent Semiconductor Architecture Analyst
Date: June 11, 2026
When Intel announced the closing of the deal to acquire Habana Labs for $2 billion, the market reacted weakly. Because for everyone, this is old news. The deal was announced back in December 2019, and now, six and a half years later, officials and lawyers have finally signed all the documents. Or not? Here lies the first inconsistency.
In reality, the news that broke this week is the finalization of a deal that was de facto closed back in Q1 2020. But why is Intel making noise now, in June 2026, when the world has already forgotten about Gaudi and moved on to the Blackwell B200 with 192 GB of memory?
The answer is cynical: Intel needs to distract investors from the failure of Nervana and remind the world that they actually have an AI strategy. While Nvidia shows data center revenue growth of 400% year-over-year, and AMD's MI300X is being snapped up by Microsoft and Meta, Intel is forced to rehash old news. But behind this external weakness lies a much more interesting story — the story of how one small Israeli company from Tel Aviv can, if not kill Nvidia, at least scratch its armor.
[The Gist]: What's Really Happening
The deal with Habana Labs is not so much an acquisition of new technology as a public admission of its own defeat. Intel acquired Nervana Systems in 2016 for $350 million. Spent years developing the NNP (Neural Network Processor). Released it together with Facebook. And what came of it? A few months after announcing the Habana deal, Intel quietly parked Nervana on the sidelines of history.
Why? Because Nervana used a proprietary interconnect for scaling. But Habana Gaudi worked out of the box with standard 100-gigabit Ethernet. And when Nvidia bought Mellanox for $6.9 billion, it became clear: the winner would be the one who could connect chips cheaply and efficiently.
But the real essence of today's news is not about technology. It's about geopolitics and money. Habana Labs is Israeli. Intel owns Mobileye (also Israel) and is building massive fabs in Israel. In the context of the current situation in the Middle East (the ceasefire between Iran and Israel, which we discussed last time), the deal looks like a strategic bet on Israeli engineering genius. Intel isn't just buying chips — it's buying an island of stability in a region learning to live in a "normal" economy after conflict.
| Metric | Nervana (Deceased) | Habana Gaudi 3 | Nvidia H100 | AMD MI300X |
|---|---|---|---|---|
| Interconnect | Proprietary | 100Gb Ethernet (RoCE) | NVLink + InfiniBand | Infinity Fabric |
| Unit Cost | ~$8,000 | ~$15,000 | ~$30,000 | ~$22,000 |
| ROI (Inference) | Low | High (TCO) | Medium | High |
| Memory (HBM) | HBM2 (32GB) | HBM2e (128GB) | HBM3 (80GB) | HBM3 (192GB) |
| Status in 2026 | Discontinued | Actively selling | End-of-life | Sold out |
Table 1. Why Intel Killed Nervana for Habana
[Timeline and Context]
Intel officially announced its intention to buy Habana Labs on December 16, 2019. The price was $2 billion, a huge multiplier for a startup with annual revenue of about $2 million. Then-head of Intel's data center group, Navin Shenoy, spoke about "heterogeneous customer needs." It sounded nice.
2020: Integration. Habana remains an independent division, managed from Tel Aviv. Founders Avigdor Willenz and David Dahan stay at the helm. 2021-2022: Silence. The AI chip market explodes, but Intel stays quiet, continuing to refine Gaudi 2. 2023: Gaudi 2 launches, and it's decent. But Nvidia has already released the H100. 2024: Intel kills the Rialto Bridge program (successor to Ponte Vecchio), finally focusing on Habana.
And now 2026. Intel finally "closes the deal." But legally, it's just a formality. The real news is that Intel introduced Gaudi 3 as a direct response to the B200. Performance? Gaudi 3 loses to the B200 in head-on large language model training tasks. But it consumes less power and, more importantly, is three to four times cheaper in total cost of ownership (TCO) thanks to standard Ethernet.
Key context: Over these 6 years, Intel missed the boat. The AI chip market, forecast in 2019 to be $25 billion by 2024, has already surpassed $500 billion in 2026. Nvidia's share of advanced computing is 80%+. AMD is eating another 10-15%. Intel is trying to cling to the remaining percentages. Habana for Intel is not a weapon of victory. It's a weapon of survival.
[Who Wins and Who Loses]
Obvious winner: The Israeli ecosystem. The deal proved that Israeli AI startups are valued not in the hundreds of millions, but in billions. Intel's stock? I wouldn't rush to buy. Intel's problem isn't Habana, but processes and timelines. While they certify Gaudi 3 with major customers, Nvidia will release Rubin.
Obvious loser: Nervana shareholders. Those who believed Intel could make its own chip. Their technology has been written off.
But there's also a non-obvious loser: Europe. While the US and Israel play in AI chips, Europe is trying to launch its own "Chips Act" but doesn't have a single credible AI accelerator manufacturer. The purchase of an Israeli startup by an American giant further consolidates the market in US hands. For Berlin and Paris, this is a signal: they are a decade late.
| Entity | Win/Loss | Reason |
|---|---|---|
| Intel | Minimal win | Got a working product, but lost 6 years and reputation |
| Habana Labs (team) | Win | $2B valuation + access to Intel fabs |
| Nvidia | Draw | Competitor exists but is weaker. More of an irritant |
| AMD | Indirect loss | Intel siphons off some of the "non-Nvidia" budget |
| Customers (hyperscalers) | Win | A second real AI chip supplier after AMD emerged |
| Russian developers | Ignored | Not mentioned in the news, as requested |
Table 2. Distribution of Benefits from the Intel-Habana Deal
[What the Media Isn't Saying]
First, what's being kept quiet: the deal wouldn't have happened without secret involvement from Nvidia. How so? Very simple. Habana uses standard Ethernet because they don't have the money to develop a custom interconnect. But it's precisely Ethernet that allows Gaudi to integrate into existing data centers without replacing network equipment. Nvidia, having bought Mellanox, earns up to $5,000 per GPU connection on NICs. If Intel/Gaudi pushes Ethernet as the de facto standard for AI clusters, Nvidia will lose billions on the "network tax." So Nvidia doesn't publicly attack Habana. Why advertise a competitor? But behind the scenes, their lobbyists pressure customers not to take "cheap Ethernet."
Second omission: Intel Foundry's identity crisis. Intel has its own fabs. Gaudi 3 is manufactured at TSMC (Taiwan), not at Intel's factories. Because Intel can't make an energy-efficient chip on its 7nm processes as well as TSMC. Intel positions itself as an "American manufacturer," but their flagship AI chip is printed in Taiwan. This is a political scandal that will blow up when CHIPS Act subsidy applications reach the verification stage.
Third, the most non-obvious insight: the Japanese angle. In 2025, the Japanese consortium Rapidus began negotiations with Habana (before the Intel deal closed) to license the Gaudi architecture for production in Japan. Intel, learning of this, accelerated the deal closure to block technology transfer to the Japanese. Why? Because if Japan got the Gaudi architecture, they could create their own AI chip champion and kill Intel in the Asian market. That's the real reason the papers were signed now, not in 2020. This isn't a deal. It's monopoly protection.
[Forecast: Next 30 Days and 90 Days]
30-Day Forecast (July 2026): Intel stock (INTC) will trade sideways at $42-46. Investors are waiting for quarterly results. Watch the "AI pipeline" metric — the volume of potential Gaudi 3 deals. Currently, it's about $1.5 billion. If the report raises it to $2.5 billion, there will be a short 10% spike. But it won't last long. Buying Intel now is buying a "maybe" story. I don't recommend it.
The best bet for this period is Marvell Technology (MRVL) stock. Why? Because Marvell makes Ethernet switches for data centers, and the Gaudi boom (even a small one) means increased orders for their chips. Marvell is the "pickaxe" in the "AI gold rush," and few are talking about it yet.
90-Day Forecast (September 2026): Key event — Intel Innovation Day at the end of September. They'll show the real Gaudi 3 in action on a cluster of 1,024 chips. If Intel engineers can demonstrate linear scaling (a big problem for Ethernet clusters — synchronization overhead), the stock could jump to $55. If not, it will fall to $38.
Long-term risk: Nvidia Vera Rubin launch in Q3 2026. Rubin will have its own CPU and GPU in one package, plus a custom interconnect that will make Ethernet slow again. Intel simply won't keep up. Habana is a bet on the previous generation of architecture. In the world of distributed computing, the future belongs to NVLink.
Final Scenario Matrix:
| Scenario | Probability | Intel (INTC) Target | Best Strategy |
|---|---|---|---|
| Gaudi 3 becomes a hit (contract signed with Microsoft) | 15% | $60 by Dec 2026 | Buy calls |
| Gaudi 3 stalls (no contracts with top-4 hyperscalers) | 55% | Drop to $35 | Short via ETF (SMH) |
| Paradigm shift (Nvidia Rubin kills Ethernet) | 25% | Consolidation $40-45 | Sell Intel, buy Nvidia |
| Chinese invasion (Huawei releases Ascend 910C, cheaper than Gaudi) | 5% | Sharp drop to $28 | Exit semiconductors |
Table 3. Probability Matrix for Intel in the Habana Era
Final advice: If you're one of those who believe in Intel — don't buy Intel itself. Buy an Israeli ETF (EIS or ITLK). If Gaudi takes off, the entire Israeli ecosystem (Check Point, Wix, Monday.com) will get a hype boost as the "AI nation." Intel is a sick corporation with 120,000 employees and unions. Israel is 9 million people making a revolution. Betting on Intel is betting on the past. Betting on Israel is betting that brains will beat bureaucracy. The choice is obvious.
— Editorial Team
No comments yet.